{ "base": { "dir": "/workspace/models/model-2.5-6", "bpw": 2.5287238241261742 }, "alts": [ { "dir": "/workspace/models/model-3.5-6", "bpw": 3.528303125766898 } ], "groups": [ { "idx": 0, "layers": [ "model.language_model.layers.0.self_attn.q_proj", "model.language_model.layers.0.self_attn.k_proj", "model.language_model.layers.0.self_attn.v_proj", "model.language_model.layers.0.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0014086388051509913, "dbits": 18874368 } ] }, { "idx": 1, "layers": [ "model.language_model.layers.0.mlp.experts.0.gate_proj", "model.language_model.layers.0.mlp.experts.1.gate_proj", "model.language_model.layers.0.mlp.experts.2.gate_proj", "model.language_model.layers.0.mlp.experts.3.gate_proj", "model.language_model.layers.0.mlp.experts.4.gate_proj", "model.language_model.layers.0.mlp.experts.5.gate_proj", "model.language_model.layers.0.mlp.experts.6.gate_proj", "model.language_model.layers.0.mlp.experts.7.gate_proj", "model.language_model.layers.0.mlp.experts.8.gate_proj", "model.language_model.layers.0.mlp.experts.9.gate_proj", "model.language_model.layers.0.mlp.experts.10.gate_proj", "model.language_model.layers.0.mlp.experts.11.gate_proj", "model.language_model.layers.0.mlp.experts.12.gate_proj", "model.language_model.layers.0.mlp.experts.13.gate_proj", "model.language_model.layers.0.mlp.experts.14.gate_proj", "model.language_model.layers.0.mlp.experts.15.gate_proj", "model.language_model.layers.0.mlp.experts.16.gate_proj", "model.language_model.layers.0.mlp.experts.17.gate_proj", "model.language_model.layers.0.mlp.experts.18.gate_proj", "model.language_model.layers.0.mlp.experts.19.gate_proj", "model.language_model.layers.0.mlp.experts.20.gate_proj", "model.language_model.layers.0.mlp.experts.21.gate_proj", "model.language_model.layers.0.mlp.experts.22.gate_proj", "model.language_model.layers.0.mlp.experts.23.gate_proj", "model.language_model.layers.0.mlp.experts.24.gate_proj", "model.language_model.layers.0.mlp.experts.25.gate_proj", "model.language_model.layers.0.mlp.experts.26.gate_proj", "model.language_model.layers.0.mlp.experts.27.gate_proj", "model.language_model.layers.0.mlp.experts.28.gate_proj", "model.language_model.layers.0.mlp.experts.29.gate_proj", "model.language_model.layers.0.mlp.experts.30.gate_proj", "model.language_model.layers.0.mlp.experts.31.gate_proj", "model.language_model.layers.0.mlp.experts.32.gate_proj", "model.language_model.layers.0.mlp.experts.33.gate_proj", "model.language_model.layers.0.mlp.experts.34.gate_proj", "model.language_model.layers.0.mlp.experts.35.gate_proj", "model.language_model.layers.0.mlp.experts.36.gate_proj", "model.language_model.layers.0.mlp.experts.37.gate_proj", "model.language_model.layers.0.mlp.experts.38.gate_proj", "model.language_model.layers.0.mlp.experts.39.gate_proj", "model.language_model.layers.0.mlp.experts.40.gate_proj", "model.language_model.layers.0.mlp.experts.41.gate_proj", "model.language_model.layers.0.mlp.experts.42.gate_proj", "model.language_model.layers.0.mlp.experts.43.gate_proj", "model.language_model.layers.0.mlp.experts.44.gate_proj", "model.language_model.layers.0.mlp.experts.45.gate_proj", "model.language_model.layers.0.mlp.experts.46.gate_proj", "model.language_model.layers.0.mlp.experts.47.gate_proj", "model.language_model.layers.0.mlp.experts.48.gate_proj", "model.language_model.layers.0.mlp.experts.49.gate_proj", "model.language_model.layers.0.mlp.experts.50.gate_proj", "model.language_model.layers.0.mlp.experts.51.gate_proj", "model.language_model.layers.0.mlp.experts.52.gate_proj", "model.language_model.layers.0.mlp.experts.53.gate_proj", "model.language_model.layers.0.mlp.experts.54.gate_proj", "model.language_model.layers.0.mlp.experts.55.gate_proj", "model.language_model.layers.0.mlp.experts.56.gate_proj", "model.language_model.layers.0.mlp.experts.57.gate_proj", "model.language_model.layers.0.mlp.experts.58.gate_proj", "model.language_model.layers.0.mlp.experts.59.gate_proj", "model.language_model.layers.0.mlp.experts.60.gate_proj", "model.language_model.layers.0.mlp.experts.61.gate_proj", "model.language_model.layers.0.mlp.experts.62.gate_proj", "model.language_model.layers.0.mlp.experts.63.gate_proj", "model.language_model.layers.0.mlp.experts.64.gate_proj", "model.language_model.layers.0.mlp.experts.65.gate_proj", "model.language_model.layers.0.mlp.experts.66.gate_proj", "model.language_model.layers.0.mlp.experts.67.gate_proj", "model.language_model.layers.0.mlp.experts.68.gate_proj", "model.language_model.layers.0.mlp.experts.69.gate_proj", "model.language_model.layers.0.mlp.experts.70.gate_proj", "model.language_model.layers.0.mlp.experts.71.gate_proj", "model.language_model.layers.0.mlp.experts.72.gate_proj", "model.language_model.layers.0.mlp.experts.73.gate_proj", "model.language_model.layers.0.mlp.experts.74.gate_proj", "model.language_model.layers.0.mlp.experts.75.gate_proj", "model.language_model.layers.0.mlp.experts.76.gate_proj", "model.language_model.layers.0.mlp.experts.77.gate_proj", "model.language_model.layers.0.mlp.experts.78.gate_proj", "model.language_model.layers.0.mlp.experts.79.gate_proj", "model.language_model.layers.0.mlp.experts.80.gate_proj", "model.language_model.layers.0.mlp.experts.81.gate_proj", "model.language_model.layers.0.mlp.experts.82.gate_proj", "model.language_model.layers.0.mlp.experts.83.gate_proj", "model.language_model.layers.0.mlp.experts.84.gate_proj", "model.language_model.layers.0.mlp.experts.85.gate_proj", "model.language_model.layers.0.mlp.experts.86.gate_proj", "model.language_model.layers.0.mlp.experts.87.gate_proj", "model.language_model.layers.0.mlp.experts.88.gate_proj", "model.language_model.layers.0.mlp.experts.89.gate_proj", "model.language_model.layers.0.mlp.experts.90.gate_proj", "model.language_model.layers.0.mlp.experts.91.gate_proj", "model.language_model.layers.0.mlp.experts.92.gate_proj", "model.language_model.layers.0.mlp.experts.93.gate_proj", "model.language_model.layers.0.mlp.experts.94.gate_proj", "model.language_model.layers.0.mlp.experts.95.gate_proj", "model.language_model.layers.0.mlp.experts.96.gate_proj", "model.language_model.layers.0.mlp.experts.97.gate_proj", "model.language_model.layers.0.mlp.experts.98.gate_proj", "model.language_model.layers.0.mlp.experts.99.gate_proj", "model.language_model.layers.0.mlp.experts.100.gate_proj", "model.language_model.layers.0.mlp.experts.101.gate_proj", "model.language_model.layers.0.mlp.experts.102.gate_proj", "model.language_model.layers.0.mlp.experts.103.gate_proj", "model.language_model.layers.0.mlp.experts.104.gate_proj", "model.language_model.layers.0.mlp.experts.105.gate_proj", "model.language_model.layers.0.mlp.experts.106.gate_proj", "model.language_model.layers.0.mlp.experts.107.gate_proj", "model.language_model.layers.0.mlp.experts.108.gate_proj", "model.language_model.layers.0.mlp.experts.109.gate_proj", "model.language_model.layers.0.mlp.experts.110.gate_proj", "model.language_model.layers.0.mlp.experts.111.gate_proj", "model.language_model.layers.0.mlp.experts.112.gate_proj", "model.language_model.layers.0.mlp.experts.113.gate_proj", "model.language_model.layers.0.mlp.experts.114.gate_proj", "model.language_model.layers.0.mlp.experts.115.gate_proj", "model.language_model.layers.0.mlp.experts.116.gate_proj", "model.language_model.layers.0.mlp.experts.117.gate_proj", "model.language_model.layers.0.mlp.experts.118.gate_proj", "model.language_model.layers.0.mlp.experts.119.gate_proj", "model.language_model.layers.0.mlp.experts.120.gate_proj", "model.language_model.layers.0.mlp.experts.121.gate_proj", "model.language_model.layers.0.mlp.experts.122.gate_proj", "model.language_model.layers.0.mlp.experts.123.gate_proj", "model.language_model.layers.0.mlp.experts.124.gate_proj", "model.language_model.layers.0.mlp.experts.125.gate_proj", "model.language_model.layers.0.mlp.experts.126.gate_proj", "model.language_model.layers.0.mlp.experts.127.gate_proj", "model.language_model.layers.0.mlp.experts.0.up_proj", "model.language_model.layers.0.mlp.experts.1.up_proj", "model.language_model.layers.0.mlp.experts.2.up_proj", "model.language_model.layers.0.mlp.experts.3.up_proj", "model.language_model.layers.0.mlp.experts.4.up_proj", "model.language_model.layers.0.mlp.experts.5.up_proj", "model.language_model.layers.0.mlp.experts.6.up_proj", "model.language_model.layers.0.mlp.experts.7.up_proj", "model.language_model.layers.0.mlp.experts.8.up_proj", "model.language_model.layers.0.mlp.experts.9.up_proj", "model.language_model.layers.0.mlp.experts.10.up_proj", "model.language_model.layers.0.mlp.experts.11.up_proj", "model.language_model.layers.0.mlp.experts.12.up_proj", "model.language_model.layers.0.mlp.experts.13.up_proj", "model.language_model.layers.0.mlp.experts.14.up_proj", "model.language_model.layers.0.mlp.experts.15.up_proj", "model.language_model.layers.0.mlp.experts.16.up_proj", "model.language_model.layers.0.mlp.experts.17.up_proj", "model.language_model.layers.0.mlp.experts.18.up_proj", "model.language_model.layers.0.mlp.experts.19.up_proj", "model.language_model.layers.0.mlp.experts.20.up_proj", "model.language_model.layers.0.mlp.experts.21.up_proj", "model.language_model.layers.0.mlp.experts.22.up_proj", "model.language_model.layers.0.mlp.experts.23.up_proj", "model.language_model.layers.0.mlp.experts.24.up_proj", "model.language_model.layers.0.mlp.experts.25.up_proj", "model.language_model.layers.0.mlp.experts.26.up_proj", "model.language_model.layers.0.mlp.experts.27.up_proj", "model.language_model.layers.0.mlp.experts.28.up_proj", "model.language_model.layers.0.mlp.experts.29.up_proj", "model.language_model.layers.0.mlp.experts.30.up_proj", "model.language_model.layers.0.mlp.experts.31.up_proj", "model.language_model.layers.0.mlp.experts.32.up_proj", "model.language_model.layers.0.mlp.experts.33.up_proj", "model.language_model.layers.0.mlp.experts.34.up_proj", "model.language_model.layers.0.mlp.experts.35.up_proj", "model.language_model.layers.0.mlp.experts.36.up_proj", "model.language_model.layers.0.mlp.experts.37.up_proj", "model.language_model.layers.0.mlp.experts.38.up_proj", "model.language_model.layers.0.mlp.experts.39.up_proj", "model.language_model.layers.0.mlp.experts.40.up_proj", "model.language_model.layers.0.mlp.experts.41.up_proj", "model.language_model.layers.0.mlp.experts.42.up_proj", "model.language_model.layers.0.mlp.experts.43.up_proj", "model.language_model.layers.0.mlp.experts.44.up_proj", "model.language_model.layers.0.mlp.experts.45.up_proj", "model.language_model.layers.0.mlp.experts.46.up_proj", "model.language_model.layers.0.mlp.experts.47.up_proj", "model.language_model.layers.0.mlp.experts.48.up_proj", "model.language_model.layers.0.mlp.experts.49.up_proj", "model.language_model.layers.0.mlp.experts.50.up_proj", "model.language_model.layers.0.mlp.experts.51.up_proj", "model.language_model.layers.0.mlp.experts.52.up_proj", "model.language_model.layers.0.mlp.experts.53.up_proj", "model.language_model.layers.0.mlp.experts.54.up_proj", "model.language_model.layers.0.mlp.experts.55.up_proj", "model.language_model.layers.0.mlp.experts.56.up_proj", "model.language_model.layers.0.mlp.experts.57.up_proj", "model.language_model.layers.0.mlp.experts.58.up_proj", "model.language_model.layers.0.mlp.experts.59.up_proj", "model.language_model.layers.0.mlp.experts.60.up_proj", "model.language_model.layers.0.mlp.experts.61.up_proj", "model.language_model.layers.0.mlp.experts.62.up_proj", "model.language_model.layers.0.mlp.experts.63.up_proj", "model.language_model.layers.0.mlp.experts.64.up_proj", "model.language_model.layers.0.mlp.experts.65.up_proj", "model.language_model.layers.0.mlp.experts.66.up_proj", "model.language_model.layers.0.mlp.experts.67.up_proj", "model.language_model.layers.0.mlp.experts.68.up_proj", "model.language_model.layers.0.mlp.experts.69.up_proj", "model.language_model.layers.0.mlp.experts.70.up_proj", "model.language_model.layers.0.mlp.experts.71.up_proj", "model.language_model.layers.0.mlp.experts.72.up_proj", "model.language_model.layers.0.mlp.experts.73.up_proj", "model.language_model.layers.0.mlp.experts.74.up_proj", "model.language_model.layers.0.mlp.experts.75.up_proj", "model.language_model.layers.0.mlp.experts.76.up_proj", "model.language_model.layers.0.mlp.experts.77.up_proj", "model.language_model.layers.0.mlp.experts.78.up_proj", "model.language_model.layers.0.mlp.experts.79.up_proj", "model.language_model.layers.0.mlp.experts.80.up_proj", "model.language_model.layers.0.mlp.experts.81.up_proj", "model.language_model.layers.0.mlp.experts.82.up_proj", "model.language_model.layers.0.mlp.experts.83.up_proj", "model.language_model.layers.0.mlp.experts.84.up_proj", "model.language_model.layers.0.mlp.experts.85.up_proj", "model.language_model.layers.0.mlp.experts.86.up_proj", "model.language_model.layers.0.mlp.experts.87.up_proj", "model.language_model.layers.0.mlp.experts.88.up_proj", "model.language_model.layers.0.mlp.experts.89.up_proj", "model.language_model.layers.0.mlp.experts.90.up_proj", "model.language_model.layers.0.mlp.experts.91.up_proj", "model.language_model.layers.0.mlp.experts.92.up_proj", "model.language_model.layers.0.mlp.experts.93.up_proj", "model.language_model.layers.0.mlp.experts.94.up_proj", "model.language_model.layers.0.mlp.experts.95.up_proj", "model.language_model.layers.0.mlp.experts.96.up_proj", "model.language_model.layers.0.mlp.experts.97.up_proj", "model.language_model.layers.0.mlp.experts.98.up_proj", "model.language_model.layers.0.mlp.experts.99.up_proj", "model.language_model.layers.0.mlp.experts.100.up_proj", "model.language_model.layers.0.mlp.experts.101.up_proj", "model.language_model.layers.0.mlp.experts.102.up_proj", "model.language_model.layers.0.mlp.experts.103.up_proj", "model.language_model.layers.0.mlp.experts.104.up_proj", "model.language_model.layers.0.mlp.experts.105.up_proj", "model.language_model.layers.0.mlp.experts.106.up_proj", "model.language_model.layers.0.mlp.experts.107.up_proj", "model.language_model.layers.0.mlp.experts.108.up_proj", "model.language_model.layers.0.mlp.experts.109.up_proj", "model.language_model.layers.0.mlp.experts.110.up_proj", "model.language_model.layers.0.mlp.experts.111.up_proj", "model.language_model.layers.0.mlp.experts.112.up_proj", "model.language_model.layers.0.mlp.experts.113.up_proj", "model.language_model.layers.0.mlp.experts.114.up_proj", "model.language_model.layers.0.mlp.experts.115.up_proj", "model.language_model.layers.0.mlp.experts.116.up_proj", "model.language_model.layers.0.mlp.experts.117.up_proj", "model.language_model.layers.0.mlp.experts.118.up_proj", "model.language_model.layers.0.mlp.experts.119.up_proj", "model.language_model.layers.0.mlp.experts.120.up_proj", "model.language_model.layers.0.mlp.experts.121.up_proj", "model.language_model.layers.0.mlp.experts.122.up_proj", "model.language_model.layers.0.mlp.experts.123.up_proj", "model.language_model.layers.0.mlp.experts.124.up_proj", "model.language_model.layers.0.mlp.experts.125.up_proj", "model.language_model.layers.0.mlp.experts.126.up_proj", "model.language_model.layers.0.mlp.experts.127.up_proj", "model.language_model.layers.0.mlp.experts.0.down_proj", "model.language_model.layers.0.mlp.experts.1.down_proj", "model.language_model.layers.0.mlp.experts.2.down_proj", "model.language_model.layers.0.mlp.experts.3.down_proj", "model.language_model.layers.0.mlp.experts.4.down_proj", "model.language_model.layers.0.mlp.experts.5.down_proj", "model.language_model.layers.0.mlp.experts.6.down_proj", "model.language_model.layers.0.mlp.experts.7.down_proj", "model.language_model.layers.0.mlp.experts.8.down_proj", "model.language_model.layers.0.mlp.experts.9.down_proj", "model.language_model.layers.0.mlp.experts.10.down_proj", "model.language_model.layers.0.mlp.experts.11.down_proj", "model.language_model.layers.0.mlp.experts.12.down_proj", "model.language_model.layers.0.mlp.experts.13.down_proj", "model.language_model.layers.0.mlp.experts.14.down_proj", "model.language_model.layers.0.mlp.experts.15.down_proj", "model.language_model.layers.0.mlp.experts.16.down_proj", "model.language_model.layers.0.mlp.experts.17.down_proj", "model.language_model.layers.0.mlp.experts.18.down_proj", "model.language_model.layers.0.mlp.experts.19.down_proj", "model.language_model.layers.0.mlp.experts.20.down_proj", "model.language_model.layers.0.mlp.experts.21.down_proj", "model.language_model.layers.0.mlp.experts.22.down_proj", "model.language_model.layers.0.mlp.experts.23.down_proj", "model.language_model.layers.0.mlp.experts.24.down_proj", "model.language_model.layers.0.mlp.experts.25.down_proj", "model.language_model.layers.0.mlp.experts.26.down_proj", "model.language_model.layers.0.mlp.experts.27.down_proj", "model.language_model.layers.0.mlp.experts.28.down_proj", "model.language_model.layers.0.mlp.experts.29.down_proj", "model.language_model.layers.0.mlp.experts.30.down_proj", "model.language_model.layers.0.mlp.experts.31.down_proj", "model.language_model.layers.0.mlp.experts.32.down_proj", "model.language_model.layers.0.mlp.experts.33.down_proj", "model.language_model.layers.0.mlp.experts.34.down_proj", "model.language_model.layers.0.mlp.experts.35.down_proj", "model.language_model.layers.0.mlp.experts.36.down_proj", "model.language_model.layers.0.mlp.experts.37.down_proj", "model.language_model.layers.0.mlp.experts.38.down_proj", "model.language_model.layers.0.mlp.experts.39.down_proj", "model.language_model.layers.0.mlp.experts.40.down_proj", "model.language_model.layers.0.mlp.experts.41.down_proj", "model.language_model.layers.0.mlp.experts.42.down_proj", "model.language_model.layers.0.mlp.experts.43.down_proj", "model.language_model.layers.0.mlp.experts.44.down_proj", "model.language_model.layers.0.mlp.experts.45.down_proj", "model.language_model.layers.0.mlp.experts.46.down_proj", "model.language_model.layers.0.mlp.experts.47.down_proj", "model.language_model.layers.0.mlp.experts.48.down_proj", "model.language_model.layers.0.mlp.experts.49.down_proj", "model.language_model.layers.0.mlp.experts.50.down_proj", "model.language_model.layers.0.mlp.experts.51.down_proj", "model.language_model.layers.0.mlp.experts.52.down_proj", "model.language_model.layers.0.mlp.experts.53.down_proj", "model.language_model.layers.0.mlp.experts.54.down_proj", "model.language_model.layers.0.mlp.experts.55.down_proj", "model.language_model.layers.0.mlp.experts.56.down_proj", "model.language_model.layers.0.mlp.experts.57.down_proj", "model.language_model.layers.0.mlp.experts.58.down_proj", "model.language_model.layers.0.mlp.experts.59.down_proj", "model.language_model.layers.0.mlp.experts.60.down_proj", "model.language_model.layers.0.mlp.experts.61.down_proj", "model.language_model.layers.0.mlp.experts.62.down_proj", "model.language_model.layers.0.mlp.experts.63.down_proj", "model.language_model.layers.0.mlp.experts.64.down_proj", "model.language_model.layers.0.mlp.experts.65.down_proj", "model.language_model.layers.0.mlp.experts.66.down_proj", "model.language_model.layers.0.mlp.experts.67.down_proj", "model.language_model.layers.0.mlp.experts.68.down_proj", "model.language_model.layers.0.mlp.experts.69.down_proj", "model.language_model.layers.0.mlp.experts.70.down_proj", "model.language_model.layers.0.mlp.experts.71.down_proj", "model.language_model.layers.0.mlp.experts.72.down_proj", "model.language_model.layers.0.mlp.experts.73.down_proj", "model.language_model.layers.0.mlp.experts.74.down_proj", "model.language_model.layers.0.mlp.experts.75.down_proj", "model.language_model.layers.0.mlp.experts.76.down_proj", "model.language_model.layers.0.mlp.experts.77.down_proj", "model.language_model.layers.0.mlp.experts.78.down_proj", "model.language_model.layers.0.mlp.experts.79.down_proj", "model.language_model.layers.0.mlp.experts.80.down_proj", "model.language_model.layers.0.mlp.experts.81.down_proj", "model.language_model.layers.0.mlp.experts.82.down_proj", "model.language_model.layers.0.mlp.experts.83.down_proj", "model.language_model.layers.0.mlp.experts.84.down_proj", "model.language_model.layers.0.mlp.experts.85.down_proj", "model.language_model.layers.0.mlp.experts.86.down_proj", "model.language_model.layers.0.mlp.experts.87.down_proj", "model.language_model.layers.0.mlp.experts.88.down_proj", "model.language_model.layers.0.mlp.experts.89.down_proj", "model.language_model.layers.0.mlp.experts.90.down_proj", "model.language_model.layers.0.mlp.experts.91.down_proj", "model.language_model.layers.0.mlp.experts.92.down_proj", "model.language_model.layers.0.mlp.experts.93.down_proj", "model.language_model.layers.0.mlp.experts.94.down_proj", "model.language_model.layers.0.mlp.experts.95.down_proj", "model.language_model.layers.0.mlp.experts.96.down_proj", "model.language_model.layers.0.mlp.experts.97.down_proj", "model.language_model.layers.0.mlp.experts.98.down_proj", "model.language_model.layers.0.mlp.experts.99.down_proj", "model.language_model.layers.0.mlp.experts.100.down_proj", "model.language_model.layers.0.mlp.experts.101.down_proj", "model.language_model.layers.0.mlp.experts.102.down_proj", "model.language_model.layers.0.mlp.experts.103.down_proj", "model.language_model.layers.0.mlp.experts.104.down_proj", "model.language_model.layers.0.mlp.experts.105.down_proj", "model.language_model.layers.0.mlp.experts.106.down_proj", "model.language_model.layers.0.mlp.experts.107.down_proj", "model.language_model.layers.0.mlp.experts.108.down_proj", "model.language_model.layers.0.mlp.experts.109.down_proj", "model.language_model.layers.0.mlp.experts.110.down_proj", "model.language_model.layers.0.mlp.experts.111.down_proj", "model.language_model.layers.0.mlp.experts.112.down_proj", "model.language_model.layers.0.mlp.experts.113.down_proj", "model.language_model.layers.0.mlp.experts.114.down_proj", "model.language_model.layers.0.mlp.experts.115.down_proj", "model.language_model.layers.0.mlp.experts.116.down_proj", "model.language_model.layers.0.mlp.experts.117.down_proj", "model.language_model.layers.0.mlp.experts.118.down_proj", "model.language_model.layers.0.mlp.experts.119.down_proj", "model.language_model.layers.0.mlp.experts.120.down_proj", "model.language_model.layers.0.mlp.experts.121.down_proj", "model.language_model.layers.0.mlp.experts.122.down_proj", "model.language_model.layers.0.mlp.experts.123.down_proj", "model.language_model.layers.0.mlp.experts.124.down_proj", "model.language_model.layers.0.mlp.experts.125.down_proj", "model.language_model.layers.0.mlp.experts.126.down_proj", "model.language_model.layers.0.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.009487887099385264, "dbits": 603979776 } ] }, { "idx": 2, "layers": [ "model.language_model.layers.1.self_attn.q_proj", "model.language_model.layers.1.self_attn.k_proj", "model.language_model.layers.1.self_attn.v_proj", "model.language_model.layers.1.self_attn.o_proj" ], "candidates": [ { "dkld": 0.0004498978145420496, "dbits": 18874368 } ] }, { "idx": 3, "layers": [ "model.language_model.layers.1.mlp.experts.0.gate_proj", "model.language_model.layers.1.mlp.experts.1.gate_proj", "model.language_model.layers.1.mlp.experts.2.gate_proj", "model.language_model.layers.1.mlp.experts.3.gate_proj", "model.language_model.layers.1.mlp.experts.4.gate_proj", "model.language_model.layers.1.mlp.experts.5.gate_proj", "model.language_model.layers.1.mlp.experts.6.gate_proj", "model.language_model.layers.1.mlp.experts.7.gate_proj", "model.language_model.layers.1.mlp.experts.8.gate_proj", "model.language_model.layers.1.mlp.experts.9.gate_proj", "model.language_model.layers.1.mlp.experts.10.gate_proj", "model.language_model.layers.1.mlp.experts.11.gate_proj", "model.language_model.layers.1.mlp.experts.12.gate_proj", "model.language_model.layers.1.mlp.experts.13.gate_proj", "model.language_model.layers.1.mlp.experts.14.gate_proj", "model.language_model.layers.1.mlp.experts.15.gate_proj", "model.language_model.layers.1.mlp.experts.16.gate_proj", "model.language_model.layers.1.mlp.experts.17.gate_proj", "model.language_model.layers.1.mlp.experts.18.gate_proj", "model.language_model.layers.1.mlp.experts.19.gate_proj", "model.language_model.layers.1.mlp.experts.20.gate_proj", "model.language_model.layers.1.mlp.experts.21.gate_proj", "model.language_model.layers.1.mlp.experts.22.gate_proj", "model.language_model.layers.1.mlp.experts.23.gate_proj", "model.language_model.layers.1.mlp.experts.24.gate_proj", "model.language_model.layers.1.mlp.experts.25.gate_proj", "model.language_model.layers.1.mlp.experts.26.gate_proj", "model.language_model.layers.1.mlp.experts.27.gate_proj", "model.language_model.layers.1.mlp.experts.28.gate_proj", "model.language_model.layers.1.mlp.experts.29.gate_proj", "model.language_model.layers.1.mlp.experts.30.gate_proj", "model.language_model.layers.1.mlp.experts.31.gate_proj", "model.language_model.layers.1.mlp.experts.32.gate_proj", "model.language_model.layers.1.mlp.experts.33.gate_proj", "model.language_model.layers.1.mlp.experts.34.gate_proj", "model.language_model.layers.1.mlp.experts.35.gate_proj", "model.language_model.layers.1.mlp.experts.36.gate_proj", "model.language_model.layers.1.mlp.experts.37.gate_proj", "model.language_model.layers.1.mlp.experts.38.gate_proj", "model.language_model.layers.1.mlp.experts.39.gate_proj", "model.language_model.layers.1.mlp.experts.40.gate_proj", "model.language_model.layers.1.mlp.experts.41.gate_proj", "model.language_model.layers.1.mlp.experts.42.gate_proj", "model.language_model.layers.1.mlp.experts.43.gate_proj", "model.language_model.layers.1.mlp.experts.44.gate_proj", "model.language_model.layers.1.mlp.experts.45.gate_proj", "model.language_model.layers.1.mlp.experts.46.gate_proj", "model.language_model.layers.1.mlp.experts.47.gate_proj", "model.language_model.layers.1.mlp.experts.48.gate_proj", "model.language_model.layers.1.mlp.experts.49.gate_proj", "model.language_model.layers.1.mlp.experts.50.gate_proj", "model.language_model.layers.1.mlp.experts.51.gate_proj", "model.language_model.layers.1.mlp.experts.52.gate_proj", "model.language_model.layers.1.mlp.experts.53.gate_proj", "model.language_model.layers.1.mlp.experts.54.gate_proj", "model.language_model.layers.1.mlp.experts.55.gate_proj", "model.language_model.layers.1.mlp.experts.56.gate_proj", "model.language_model.layers.1.mlp.experts.57.gate_proj", "model.language_model.layers.1.mlp.experts.58.gate_proj", "model.language_model.layers.1.mlp.experts.59.gate_proj", "model.language_model.layers.1.mlp.experts.60.gate_proj", "model.language_model.layers.1.mlp.experts.61.gate_proj", "model.language_model.layers.1.mlp.experts.62.gate_proj", "model.language_model.layers.1.mlp.experts.63.gate_proj", "model.language_model.layers.1.mlp.experts.64.gate_proj", "model.language_model.layers.1.mlp.experts.65.gate_proj", "model.language_model.layers.1.mlp.experts.66.gate_proj", "model.language_model.layers.1.mlp.experts.67.gate_proj", "model.language_model.layers.1.mlp.experts.68.gate_proj", "model.language_model.layers.1.mlp.experts.69.gate_proj", "model.language_model.layers.1.mlp.experts.70.gate_proj", "model.language_model.layers.1.mlp.experts.71.gate_proj", "model.language_model.layers.1.mlp.experts.72.gate_proj", "model.language_model.layers.1.mlp.experts.73.gate_proj", "model.language_model.layers.1.mlp.experts.74.gate_proj", "model.language_model.layers.1.mlp.experts.75.gate_proj", "model.language_model.layers.1.mlp.experts.76.gate_proj", "model.language_model.layers.1.mlp.experts.77.gate_proj", "model.language_model.layers.1.mlp.experts.78.gate_proj", "model.language_model.layers.1.mlp.experts.79.gate_proj", "model.language_model.layers.1.mlp.experts.80.gate_proj", "model.language_model.layers.1.mlp.experts.81.gate_proj", "model.language_model.layers.1.mlp.experts.82.gate_proj", "model.language_model.layers.1.mlp.experts.83.gate_proj", "model.language_model.layers.1.mlp.experts.84.gate_proj", "model.language_model.layers.1.mlp.experts.85.gate_proj", "model.language_model.layers.1.mlp.experts.86.gate_proj", "model.language_model.layers.1.mlp.experts.87.gate_proj", "model.language_model.layers.1.mlp.experts.88.gate_proj", "model.language_model.layers.1.mlp.experts.89.gate_proj", "model.language_model.layers.1.mlp.experts.90.gate_proj", "model.language_model.layers.1.mlp.experts.91.gate_proj", "model.language_model.layers.1.mlp.experts.92.gate_proj", "model.language_model.layers.1.mlp.experts.93.gate_proj", "model.language_model.layers.1.mlp.experts.94.gate_proj", "model.language_model.layers.1.mlp.experts.95.gate_proj", "model.language_model.layers.1.mlp.experts.96.gate_proj", "model.language_model.layers.1.mlp.experts.97.gate_proj", "model.language_model.layers.1.mlp.experts.98.gate_proj", "model.language_model.layers.1.mlp.experts.99.gate_proj", "model.language_model.layers.1.mlp.experts.100.gate_proj", "model.language_model.layers.1.mlp.experts.101.gate_proj", "model.language_model.layers.1.mlp.experts.102.gate_proj", "model.language_model.layers.1.mlp.experts.103.gate_proj", "model.language_model.layers.1.mlp.experts.104.gate_proj", "model.language_model.layers.1.mlp.experts.105.gate_proj", "model.language_model.layers.1.mlp.experts.106.gate_proj", "model.language_model.layers.1.mlp.experts.107.gate_proj", "model.language_model.layers.1.mlp.experts.108.gate_proj", "model.language_model.layers.1.mlp.experts.109.gate_proj", "model.language_model.layers.1.mlp.experts.110.gate_proj", "model.language_model.layers.1.mlp.experts.111.gate_proj", "model.language_model.layers.1.mlp.experts.112.gate_proj", "model.language_model.layers.1.mlp.experts.113.gate_proj", "model.language_model.layers.1.mlp.experts.114.gate_proj", "model.language_model.layers.1.mlp.experts.115.gate_proj", "model.language_model.layers.1.mlp.experts.116.gate_proj", "model.language_model.layers.1.mlp.experts.117.gate_proj", "model.language_model.layers.1.mlp.experts.118.gate_proj", "model.language_model.layers.1.mlp.experts.119.gate_proj", "model.language_model.layers.1.mlp.experts.120.gate_proj", "model.language_model.layers.1.mlp.experts.121.gate_proj", "model.language_model.layers.1.mlp.experts.122.gate_proj", "model.language_model.layers.1.mlp.experts.123.gate_proj", "model.language_model.layers.1.mlp.experts.124.gate_proj", "model.language_model.layers.1.mlp.experts.125.gate_proj", "model.language_model.layers.1.mlp.experts.126.gate_proj", "model.language_model.layers.1.mlp.experts.127.gate_proj", "model.language_model.layers.1.mlp.experts.0.up_proj", "model.language_model.layers.1.mlp.experts.1.up_proj", "model.language_model.layers.1.mlp.experts.2.up_proj", "model.language_model.layers.1.mlp.experts.3.up_proj", "model.language_model.layers.1.mlp.experts.4.up_proj", "model.language_model.layers.1.mlp.experts.5.up_proj", "model.language_model.layers.1.mlp.experts.6.up_proj", "model.language_model.layers.1.mlp.experts.7.up_proj", "model.language_model.layers.1.mlp.experts.8.up_proj", "model.language_model.layers.1.mlp.experts.9.up_proj", "model.language_model.layers.1.mlp.experts.10.up_proj", "model.language_model.layers.1.mlp.experts.11.up_proj", "model.language_model.layers.1.mlp.experts.12.up_proj", "model.language_model.layers.1.mlp.experts.13.up_proj", "model.language_model.layers.1.mlp.experts.14.up_proj", "model.language_model.layers.1.mlp.experts.15.up_proj", "model.language_model.layers.1.mlp.experts.16.up_proj", "model.language_model.layers.1.mlp.experts.17.up_proj", "model.language_model.layers.1.mlp.experts.18.up_proj", "model.language_model.layers.1.mlp.experts.19.up_proj", "model.language_model.layers.1.mlp.experts.20.up_proj", "model.language_model.layers.1.mlp.experts.21.up_proj", "model.language_model.layers.1.mlp.experts.22.up_proj", "model.language_model.layers.1.mlp.experts.23.up_proj", "model.language_model.layers.1.mlp.experts.24.up_proj", "model.language_model.layers.1.mlp.experts.25.up_proj", "model.language_model.layers.1.mlp.experts.26.up_proj", "model.language_model.layers.1.mlp.experts.27.up_proj", "model.language_model.layers.1.mlp.experts.28.up_proj", "model.language_model.layers.1.mlp.experts.29.up_proj", "model.language_model.layers.1.mlp.experts.30.up_proj", "model.language_model.layers.1.mlp.experts.31.up_proj", "model.language_model.layers.1.mlp.experts.32.up_proj", "model.language_model.layers.1.mlp.experts.33.up_proj", "model.language_model.layers.1.mlp.experts.34.up_proj", "model.language_model.layers.1.mlp.experts.35.up_proj", "model.language_model.layers.1.mlp.experts.36.up_proj", "model.language_model.layers.1.mlp.experts.37.up_proj", "model.language_model.layers.1.mlp.experts.38.up_proj", "model.language_model.layers.1.mlp.experts.39.up_proj", "model.language_model.layers.1.mlp.experts.40.up_proj", "model.language_model.layers.1.mlp.experts.41.up_proj", "model.language_model.layers.1.mlp.experts.42.up_proj", "model.language_model.layers.1.mlp.experts.43.up_proj", "model.language_model.layers.1.mlp.experts.44.up_proj", "model.language_model.layers.1.mlp.experts.45.up_proj", "model.language_model.layers.1.mlp.experts.46.up_proj", "model.language_model.layers.1.mlp.experts.47.up_proj", "model.language_model.layers.1.mlp.experts.48.up_proj", "model.language_model.layers.1.mlp.experts.49.up_proj", "model.language_model.layers.1.mlp.experts.50.up_proj", "model.language_model.layers.1.mlp.experts.51.up_proj", "model.language_model.layers.1.mlp.experts.52.up_proj", "model.language_model.layers.1.mlp.experts.53.up_proj", "model.language_model.layers.1.mlp.experts.54.up_proj", "model.language_model.layers.1.mlp.experts.55.up_proj", "model.language_model.layers.1.mlp.experts.56.up_proj", "model.language_model.layers.1.mlp.experts.57.up_proj", "model.language_model.layers.1.mlp.experts.58.up_proj", "model.language_model.layers.1.mlp.experts.59.up_proj", "model.language_model.layers.1.mlp.experts.60.up_proj", "model.language_model.layers.1.mlp.experts.61.up_proj", "model.language_model.layers.1.mlp.experts.62.up_proj", "model.language_model.layers.1.mlp.experts.63.up_proj", "model.language_model.layers.1.mlp.experts.64.up_proj", "model.language_model.layers.1.mlp.experts.65.up_proj", "model.language_model.layers.1.mlp.experts.66.up_proj", "model.language_model.layers.1.mlp.experts.67.up_proj", "model.language_model.layers.1.mlp.experts.68.up_proj", "model.language_model.layers.1.mlp.experts.69.up_proj", "model.language_model.layers.1.mlp.experts.70.up_proj", "model.language_model.layers.1.mlp.experts.71.up_proj", "model.language_model.layers.1.mlp.experts.72.up_proj", "model.language_model.layers.1.mlp.experts.73.up_proj", "model.language_model.layers.1.mlp.experts.74.up_proj", "model.language_model.layers.1.mlp.experts.75.up_proj", "model.language_model.layers.1.mlp.experts.76.up_proj", "model.language_model.layers.1.mlp.experts.77.up_proj", "model.language_model.layers.1.mlp.experts.78.up_proj", "model.language_model.layers.1.mlp.experts.79.up_proj", "model.language_model.layers.1.mlp.experts.80.up_proj", "model.language_model.layers.1.mlp.experts.81.up_proj", "model.language_model.layers.1.mlp.experts.82.up_proj", "model.language_model.layers.1.mlp.experts.83.up_proj", "model.language_model.layers.1.mlp.experts.84.up_proj", "model.language_model.layers.1.mlp.experts.85.up_proj", "model.language_model.layers.1.mlp.experts.86.up_proj", "model.language_model.layers.1.mlp.experts.87.up_proj", "model.language_model.layers.1.mlp.experts.88.up_proj", "model.language_model.layers.1.mlp.experts.89.up_proj", "model.language_model.layers.1.mlp.experts.90.up_proj", "model.language_model.layers.1.mlp.experts.91.up_proj", "model.language_model.layers.1.mlp.experts.92.up_proj", "model.language_model.layers.1.mlp.experts.93.up_proj", "model.language_model.layers.1.mlp.experts.94.up_proj", "model.language_model.layers.1.mlp.experts.95.up_proj", "model.language_model.layers.1.mlp.experts.96.up_proj", "model.language_model.layers.1.mlp.experts.97.up_proj", "model.language_model.layers.1.mlp.experts.98.up_proj", "model.language_model.layers.1.mlp.experts.99.up_proj", "model.language_model.layers.1.mlp.experts.100.up_proj", "model.language_model.layers.1.mlp.experts.101.up_proj", "model.language_model.layers.1.mlp.experts.102.up_proj", "model.language_model.layers.1.mlp.experts.103.up_proj", "model.language_model.layers.1.mlp.experts.104.up_proj", "model.language_model.layers.1.mlp.experts.105.up_proj", "model.language_model.layers.1.mlp.experts.106.up_proj", "model.language_model.layers.1.mlp.experts.107.up_proj", "model.language_model.layers.1.mlp.experts.108.up_proj", "model.language_model.layers.1.mlp.experts.109.up_proj", "model.language_model.layers.1.mlp.experts.110.up_proj", "model.language_model.layers.1.mlp.experts.111.up_proj", "model.language_model.layers.1.mlp.experts.112.up_proj", "model.language_model.layers.1.mlp.experts.113.up_proj", "model.language_model.layers.1.mlp.experts.114.up_proj", "model.language_model.layers.1.mlp.experts.115.up_proj", "model.language_model.layers.1.mlp.experts.116.up_proj", "model.language_model.layers.1.mlp.experts.117.up_proj", "model.language_model.layers.1.mlp.experts.118.up_proj", "model.language_model.layers.1.mlp.experts.119.up_proj", "model.language_model.layers.1.mlp.experts.120.up_proj", "model.language_model.layers.1.mlp.experts.121.up_proj", "model.language_model.layers.1.mlp.experts.122.up_proj", "model.language_model.layers.1.mlp.experts.123.up_proj", "model.language_model.layers.1.mlp.experts.124.up_proj", "model.language_model.layers.1.mlp.experts.125.up_proj", "model.language_model.layers.1.mlp.experts.126.up_proj", "model.language_model.layers.1.mlp.experts.127.up_proj", "model.language_model.layers.1.mlp.experts.0.down_proj", "model.language_model.layers.1.mlp.experts.1.down_proj", "model.language_model.layers.1.mlp.experts.2.down_proj", "model.language_model.layers.1.mlp.experts.3.down_proj", "model.language_model.layers.1.mlp.experts.4.down_proj", "model.language_model.layers.1.mlp.experts.5.down_proj", "model.language_model.layers.1.mlp.experts.6.down_proj", "model.language_model.layers.1.mlp.experts.7.down_proj", "model.language_model.layers.1.mlp.experts.8.down_proj", "model.language_model.layers.1.mlp.experts.9.down_proj", "model.language_model.layers.1.mlp.experts.10.down_proj", "model.language_model.layers.1.mlp.experts.11.down_proj", "model.language_model.layers.1.mlp.experts.12.down_proj", "model.language_model.layers.1.mlp.experts.13.down_proj", "model.language_model.layers.1.mlp.experts.14.down_proj", "model.language_model.layers.1.mlp.experts.15.down_proj", "model.language_model.layers.1.mlp.experts.16.down_proj", "model.language_model.layers.1.mlp.experts.17.down_proj", "model.language_model.layers.1.mlp.experts.18.down_proj", "model.language_model.layers.1.mlp.experts.19.down_proj", "model.language_model.layers.1.mlp.experts.20.down_proj", "model.language_model.layers.1.mlp.experts.21.down_proj", "model.language_model.layers.1.mlp.experts.22.down_proj", "model.language_model.layers.1.mlp.experts.23.down_proj", "model.language_model.layers.1.mlp.experts.24.down_proj", "model.language_model.layers.1.mlp.experts.25.down_proj", "model.language_model.layers.1.mlp.experts.26.down_proj", "model.language_model.layers.1.mlp.experts.27.down_proj", "model.language_model.layers.1.mlp.experts.28.down_proj", "model.language_model.layers.1.mlp.experts.29.down_proj", "model.language_model.layers.1.mlp.experts.30.down_proj", "model.language_model.layers.1.mlp.experts.31.down_proj", "model.language_model.layers.1.mlp.experts.32.down_proj", "model.language_model.layers.1.mlp.experts.33.down_proj", "model.language_model.layers.1.mlp.experts.34.down_proj", "model.language_model.layers.1.mlp.experts.35.down_proj", "model.language_model.layers.1.mlp.experts.36.down_proj", "model.language_model.layers.1.mlp.experts.37.down_proj", "model.language_model.layers.1.mlp.experts.38.down_proj", "model.language_model.layers.1.mlp.experts.39.down_proj", "model.language_model.layers.1.mlp.experts.40.down_proj", "model.language_model.layers.1.mlp.experts.41.down_proj", "model.language_model.layers.1.mlp.experts.42.down_proj", "model.language_model.layers.1.mlp.experts.43.down_proj", "model.language_model.layers.1.mlp.experts.44.down_proj", "model.language_model.layers.1.mlp.experts.45.down_proj", "model.language_model.layers.1.mlp.experts.46.down_proj", "model.language_model.layers.1.mlp.experts.47.down_proj", "model.language_model.layers.1.mlp.experts.48.down_proj", "model.language_model.layers.1.mlp.experts.49.down_proj", "model.language_model.layers.1.mlp.experts.50.down_proj", "model.language_model.layers.1.mlp.experts.51.down_proj", "model.language_model.layers.1.mlp.experts.52.down_proj", "model.language_model.layers.1.mlp.experts.53.down_proj", "model.language_model.layers.1.mlp.experts.54.down_proj", "model.language_model.layers.1.mlp.experts.55.down_proj", "model.language_model.layers.1.mlp.experts.56.down_proj", "model.language_model.layers.1.mlp.experts.57.down_proj", "model.language_model.layers.1.mlp.experts.58.down_proj", "model.language_model.layers.1.mlp.experts.59.down_proj", "model.language_model.layers.1.mlp.experts.60.down_proj", "model.language_model.layers.1.mlp.experts.61.down_proj", "model.language_model.layers.1.mlp.experts.62.down_proj", "model.language_model.layers.1.mlp.experts.63.down_proj", "model.language_model.layers.1.mlp.experts.64.down_proj", "model.language_model.layers.1.mlp.experts.65.down_proj", "model.language_model.layers.1.mlp.experts.66.down_proj", "model.language_model.layers.1.mlp.experts.67.down_proj", "model.language_model.layers.1.mlp.experts.68.down_proj", "model.language_model.layers.1.mlp.experts.69.down_proj", "model.language_model.layers.1.mlp.experts.70.down_proj", "model.language_model.layers.1.mlp.experts.71.down_proj", "model.language_model.layers.1.mlp.experts.72.down_proj", "model.language_model.layers.1.mlp.experts.73.down_proj", "model.language_model.layers.1.mlp.experts.74.down_proj", "model.language_model.layers.1.mlp.experts.75.down_proj", "model.language_model.layers.1.mlp.experts.76.down_proj", "model.language_model.layers.1.mlp.experts.77.down_proj", "model.language_model.layers.1.mlp.experts.78.down_proj", "model.language_model.layers.1.mlp.experts.79.down_proj", "model.language_model.layers.1.mlp.experts.80.down_proj", "model.language_model.layers.1.mlp.experts.81.down_proj", "model.language_model.layers.1.mlp.experts.82.down_proj", "model.language_model.layers.1.mlp.experts.83.down_proj", "model.language_model.layers.1.mlp.experts.84.down_proj", "model.language_model.layers.1.mlp.experts.85.down_proj", "model.language_model.layers.1.mlp.experts.86.down_proj", "model.language_model.layers.1.mlp.experts.87.down_proj", "model.language_model.layers.1.mlp.experts.88.down_proj", "model.language_model.layers.1.mlp.experts.89.down_proj", "model.language_model.layers.1.mlp.experts.90.down_proj", "model.language_model.layers.1.mlp.experts.91.down_proj", "model.language_model.layers.1.mlp.experts.92.down_proj", "model.language_model.layers.1.mlp.experts.93.down_proj", "model.language_model.layers.1.mlp.experts.94.down_proj", "model.language_model.layers.1.mlp.experts.95.down_proj", "model.language_model.layers.1.mlp.experts.96.down_proj", "model.language_model.layers.1.mlp.experts.97.down_proj", "model.language_model.layers.1.mlp.experts.98.down_proj", "model.language_model.layers.1.mlp.experts.99.down_proj", "model.language_model.layers.1.mlp.experts.100.down_proj", "model.language_model.layers.1.mlp.experts.101.down_proj", "model.language_model.layers.1.mlp.experts.102.down_proj", "model.language_model.layers.1.mlp.experts.103.down_proj", "model.language_model.layers.1.mlp.experts.104.down_proj", "model.language_model.layers.1.mlp.experts.105.down_proj", "model.language_model.layers.1.mlp.experts.106.down_proj", "model.language_model.layers.1.mlp.experts.107.down_proj", "model.language_model.layers.1.mlp.experts.108.down_proj", "model.language_model.layers.1.mlp.experts.109.down_proj", "model.language_model.layers.1.mlp.experts.110.down_proj", "model.language_model.layers.1.mlp.experts.111.down_proj", "model.language_model.layers.1.mlp.experts.112.down_proj", "model.language_model.layers.1.mlp.experts.113.down_proj", "model.language_model.layers.1.mlp.experts.114.down_proj", "model.language_model.layers.1.mlp.experts.115.down_proj", "model.language_model.layers.1.mlp.experts.116.down_proj", "model.language_model.layers.1.mlp.experts.117.down_proj", "model.language_model.layers.1.mlp.experts.118.down_proj", "model.language_model.layers.1.mlp.experts.119.down_proj", "model.language_model.layers.1.mlp.experts.120.down_proj", "model.language_model.layers.1.mlp.experts.121.down_proj", "model.language_model.layers.1.mlp.experts.122.down_proj", "model.language_model.layers.1.mlp.experts.123.down_proj", "model.language_model.layers.1.mlp.experts.124.down_proj", "model.language_model.layers.1.mlp.experts.125.down_proj", "model.language_model.layers.1.mlp.experts.126.down_proj", "model.language_model.layers.1.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.007288339640945191, "dbits": 603979776 } ] }, { "idx": 4, "layers": [ "model.language_model.layers.2.self_attn.q_proj", "model.language_model.layers.2.self_attn.k_proj", "model.language_model.layers.2.self_attn.v_proj", "model.language_model.layers.2.self_attn.o_proj" ], "candidates": [ { "dkld": 0.0009018868207931463, "dbits": 18874368 } ] }, { "idx": 5, "layers": [ "model.language_model.layers.2.mlp.experts.0.gate_proj", "model.language_model.layers.2.mlp.experts.1.gate_proj", "model.language_model.layers.2.mlp.experts.2.gate_proj", "model.language_model.layers.2.mlp.experts.3.gate_proj", "model.language_model.layers.2.mlp.experts.4.gate_proj", "model.language_model.layers.2.mlp.experts.5.gate_proj", "model.language_model.layers.2.mlp.experts.6.gate_proj", "model.language_model.layers.2.mlp.experts.7.gate_proj", "model.language_model.layers.2.mlp.experts.8.gate_proj", "model.language_model.layers.2.mlp.experts.9.gate_proj", "model.language_model.layers.2.mlp.experts.10.gate_proj", "model.language_model.layers.2.mlp.experts.11.gate_proj", "model.language_model.layers.2.mlp.experts.12.gate_proj", "model.language_model.layers.2.mlp.experts.13.gate_proj", "model.language_model.layers.2.mlp.experts.14.gate_proj", "model.language_model.layers.2.mlp.experts.15.gate_proj", "model.language_model.layers.2.mlp.experts.16.gate_proj", "model.language_model.layers.2.mlp.experts.17.gate_proj", "model.language_model.layers.2.mlp.experts.18.gate_proj", "model.language_model.layers.2.mlp.experts.19.gate_proj", "model.language_model.layers.2.mlp.experts.20.gate_proj", "model.language_model.layers.2.mlp.experts.21.gate_proj", "model.language_model.layers.2.mlp.experts.22.gate_proj", "model.language_model.layers.2.mlp.experts.23.gate_proj", "model.language_model.layers.2.mlp.experts.24.gate_proj", "model.language_model.layers.2.mlp.experts.25.gate_proj", "model.language_model.layers.2.mlp.experts.26.gate_proj", "model.language_model.layers.2.mlp.experts.27.gate_proj", "model.language_model.layers.2.mlp.experts.28.gate_proj", "model.language_model.layers.2.mlp.experts.29.gate_proj", "model.language_model.layers.2.mlp.experts.30.gate_proj", "model.language_model.layers.2.mlp.experts.31.gate_proj", "model.language_model.layers.2.mlp.experts.32.gate_proj", "model.language_model.layers.2.mlp.experts.33.gate_proj", "model.language_model.layers.2.mlp.experts.34.gate_proj", "model.language_model.layers.2.mlp.experts.35.gate_proj", "model.language_model.layers.2.mlp.experts.36.gate_proj", "model.language_model.layers.2.mlp.experts.37.gate_proj", "model.language_model.layers.2.mlp.experts.38.gate_proj", "model.language_model.layers.2.mlp.experts.39.gate_proj", "model.language_model.layers.2.mlp.experts.40.gate_proj", "model.language_model.layers.2.mlp.experts.41.gate_proj", "model.language_model.layers.2.mlp.experts.42.gate_proj", "model.language_model.layers.2.mlp.experts.43.gate_proj", "model.language_model.layers.2.mlp.experts.44.gate_proj", "model.language_model.layers.2.mlp.experts.45.gate_proj", "model.language_model.layers.2.mlp.experts.46.gate_proj", "model.language_model.layers.2.mlp.experts.47.gate_proj", "model.language_model.layers.2.mlp.experts.48.gate_proj", "model.language_model.layers.2.mlp.experts.49.gate_proj", "model.language_model.layers.2.mlp.experts.50.gate_proj", "model.language_model.layers.2.mlp.experts.51.gate_proj", "model.language_model.layers.2.mlp.experts.52.gate_proj", "model.language_model.layers.2.mlp.experts.53.gate_proj", "model.language_model.layers.2.mlp.experts.54.gate_proj", "model.language_model.layers.2.mlp.experts.55.gate_proj", "model.language_model.layers.2.mlp.experts.56.gate_proj", "model.language_model.layers.2.mlp.experts.57.gate_proj", "model.language_model.layers.2.mlp.experts.58.gate_proj", "model.language_model.layers.2.mlp.experts.59.gate_proj", "model.language_model.layers.2.mlp.experts.60.gate_proj", "model.language_model.layers.2.mlp.experts.61.gate_proj", "model.language_model.layers.2.mlp.experts.62.gate_proj", "model.language_model.layers.2.mlp.experts.63.gate_proj", "model.language_model.layers.2.mlp.experts.64.gate_proj", "model.language_model.layers.2.mlp.experts.65.gate_proj", "model.language_model.layers.2.mlp.experts.66.gate_proj", "model.language_model.layers.2.mlp.experts.67.gate_proj", "model.language_model.layers.2.mlp.experts.68.gate_proj", "model.language_model.layers.2.mlp.experts.69.gate_proj", "model.language_model.layers.2.mlp.experts.70.gate_proj", "model.language_model.layers.2.mlp.experts.71.gate_proj", "model.language_model.layers.2.mlp.experts.72.gate_proj", "model.language_model.layers.2.mlp.experts.73.gate_proj", "model.language_model.layers.2.mlp.experts.74.gate_proj", "model.language_model.layers.2.mlp.experts.75.gate_proj", "model.language_model.layers.2.mlp.experts.76.gate_proj", "model.language_model.layers.2.mlp.experts.77.gate_proj", "model.language_model.layers.2.mlp.experts.78.gate_proj", "model.language_model.layers.2.mlp.experts.79.gate_proj", "model.language_model.layers.2.mlp.experts.80.gate_proj", "model.language_model.layers.2.mlp.experts.81.gate_proj", "model.language_model.layers.2.mlp.experts.82.gate_proj", "model.language_model.layers.2.mlp.experts.83.gate_proj", "model.language_model.layers.2.mlp.experts.84.gate_proj", "model.language_model.layers.2.mlp.experts.85.gate_proj", "model.language_model.layers.2.mlp.experts.86.gate_proj", "model.language_model.layers.2.mlp.experts.87.gate_proj", "model.language_model.layers.2.mlp.experts.88.gate_proj", "model.language_model.layers.2.mlp.experts.89.gate_proj", "model.language_model.layers.2.mlp.experts.90.gate_proj", "model.language_model.layers.2.mlp.experts.91.gate_proj", "model.language_model.layers.2.mlp.experts.92.gate_proj", "model.language_model.layers.2.mlp.experts.93.gate_proj", "model.language_model.layers.2.mlp.experts.94.gate_proj", "model.language_model.layers.2.mlp.experts.95.gate_proj", "model.language_model.layers.2.mlp.experts.96.gate_proj", "model.language_model.layers.2.mlp.experts.97.gate_proj", "model.language_model.layers.2.mlp.experts.98.gate_proj", "model.language_model.layers.2.mlp.experts.99.gate_proj", "model.language_model.layers.2.mlp.experts.100.gate_proj", "model.language_model.layers.2.mlp.experts.101.gate_proj", "model.language_model.layers.2.mlp.experts.102.gate_proj", "model.language_model.layers.2.mlp.experts.103.gate_proj", "model.language_model.layers.2.mlp.experts.104.gate_proj", "model.language_model.layers.2.mlp.experts.105.gate_proj", "model.language_model.layers.2.mlp.experts.106.gate_proj", "model.language_model.layers.2.mlp.experts.107.gate_proj", "model.language_model.layers.2.mlp.experts.108.gate_proj", "model.language_model.layers.2.mlp.experts.109.gate_proj", "model.language_model.layers.2.mlp.experts.110.gate_proj", "model.language_model.layers.2.mlp.experts.111.gate_proj", "model.language_model.layers.2.mlp.experts.112.gate_proj", "model.language_model.layers.2.mlp.experts.113.gate_proj", "model.language_model.layers.2.mlp.experts.114.gate_proj", "model.language_model.layers.2.mlp.experts.115.gate_proj", "model.language_model.layers.2.mlp.experts.116.gate_proj", "model.language_model.layers.2.mlp.experts.117.gate_proj", "model.language_model.layers.2.mlp.experts.118.gate_proj", "model.language_model.layers.2.mlp.experts.119.gate_proj", "model.language_model.layers.2.mlp.experts.120.gate_proj", "model.language_model.layers.2.mlp.experts.121.gate_proj", "model.language_model.layers.2.mlp.experts.122.gate_proj", "model.language_model.layers.2.mlp.experts.123.gate_proj", "model.language_model.layers.2.mlp.experts.124.gate_proj", "model.language_model.layers.2.mlp.experts.125.gate_proj", "model.language_model.layers.2.mlp.experts.126.gate_proj", "model.language_model.layers.2.mlp.experts.127.gate_proj", "model.language_model.layers.2.mlp.experts.0.up_proj", "model.language_model.layers.2.mlp.experts.1.up_proj", "model.language_model.layers.2.mlp.experts.2.up_proj", "model.language_model.layers.2.mlp.experts.3.up_proj", "model.language_model.layers.2.mlp.experts.4.up_proj", "model.language_model.layers.2.mlp.experts.5.up_proj", "model.language_model.layers.2.mlp.experts.6.up_proj", "model.language_model.layers.2.mlp.experts.7.up_proj", "model.language_model.layers.2.mlp.experts.8.up_proj", "model.language_model.layers.2.mlp.experts.9.up_proj", "model.language_model.layers.2.mlp.experts.10.up_proj", "model.language_model.layers.2.mlp.experts.11.up_proj", "model.language_model.layers.2.mlp.experts.12.up_proj", "model.language_model.layers.2.mlp.experts.13.up_proj", "model.language_model.layers.2.mlp.experts.14.up_proj", "model.language_model.layers.2.mlp.experts.15.up_proj", "model.language_model.layers.2.mlp.experts.16.up_proj", "model.language_model.layers.2.mlp.experts.17.up_proj", "model.language_model.layers.2.mlp.experts.18.up_proj", "model.language_model.layers.2.mlp.experts.19.up_proj", "model.language_model.layers.2.mlp.experts.20.up_proj", "model.language_model.layers.2.mlp.experts.21.up_proj", "model.language_model.layers.2.mlp.experts.22.up_proj", "model.language_model.layers.2.mlp.experts.23.up_proj", "model.language_model.layers.2.mlp.experts.24.up_proj", "model.language_model.layers.2.mlp.experts.25.up_proj", "model.language_model.layers.2.mlp.experts.26.up_proj", "model.language_model.layers.2.mlp.experts.27.up_proj", "model.language_model.layers.2.mlp.experts.28.up_proj", "model.language_model.layers.2.mlp.experts.29.up_proj", "model.language_model.layers.2.mlp.experts.30.up_proj", "model.language_model.layers.2.mlp.experts.31.up_proj", "model.language_model.layers.2.mlp.experts.32.up_proj", "model.language_model.layers.2.mlp.experts.33.up_proj", "model.language_model.layers.2.mlp.experts.34.up_proj", "model.language_model.layers.2.mlp.experts.35.up_proj", "model.language_model.layers.2.mlp.experts.36.up_proj", "model.language_model.layers.2.mlp.experts.37.up_proj", "model.language_model.layers.2.mlp.experts.38.up_proj", "model.language_model.layers.2.mlp.experts.39.up_proj", "model.language_model.layers.2.mlp.experts.40.up_proj", "model.language_model.layers.2.mlp.experts.41.up_proj", "model.language_model.layers.2.mlp.experts.42.up_proj", "model.language_model.layers.2.mlp.experts.43.up_proj", "model.language_model.layers.2.mlp.experts.44.up_proj", "model.language_model.layers.2.mlp.experts.45.up_proj", "model.language_model.layers.2.mlp.experts.46.up_proj", "model.language_model.layers.2.mlp.experts.47.up_proj", "model.language_model.layers.2.mlp.experts.48.up_proj", "model.language_model.layers.2.mlp.experts.49.up_proj", "model.language_model.layers.2.mlp.experts.50.up_proj", "model.language_model.layers.2.mlp.experts.51.up_proj", "model.language_model.layers.2.mlp.experts.52.up_proj", "model.language_model.layers.2.mlp.experts.53.up_proj", "model.language_model.layers.2.mlp.experts.54.up_proj", "model.language_model.layers.2.mlp.experts.55.up_proj", "model.language_model.layers.2.mlp.experts.56.up_proj", "model.language_model.layers.2.mlp.experts.57.up_proj", "model.language_model.layers.2.mlp.experts.58.up_proj", "model.language_model.layers.2.mlp.experts.59.up_proj", "model.language_model.layers.2.mlp.experts.60.up_proj", "model.language_model.layers.2.mlp.experts.61.up_proj", "model.language_model.layers.2.mlp.experts.62.up_proj", "model.language_model.layers.2.mlp.experts.63.up_proj", "model.language_model.layers.2.mlp.experts.64.up_proj", "model.language_model.layers.2.mlp.experts.65.up_proj", "model.language_model.layers.2.mlp.experts.66.up_proj", "model.language_model.layers.2.mlp.experts.67.up_proj", "model.language_model.layers.2.mlp.experts.68.up_proj", "model.language_model.layers.2.mlp.experts.69.up_proj", "model.language_model.layers.2.mlp.experts.70.up_proj", "model.language_model.layers.2.mlp.experts.71.up_proj", "model.language_model.layers.2.mlp.experts.72.up_proj", "model.language_model.layers.2.mlp.experts.73.up_proj", "model.language_model.layers.2.mlp.experts.74.up_proj", "model.language_model.layers.2.mlp.experts.75.up_proj", "model.language_model.layers.2.mlp.experts.76.up_proj", "model.language_model.layers.2.mlp.experts.77.up_proj", "model.language_model.layers.2.mlp.experts.78.up_proj", "model.language_model.layers.2.mlp.experts.79.up_proj", "model.language_model.layers.2.mlp.experts.80.up_proj", "model.language_model.layers.2.mlp.experts.81.up_proj", "model.language_model.layers.2.mlp.experts.82.up_proj", "model.language_model.layers.2.mlp.experts.83.up_proj", "model.language_model.layers.2.mlp.experts.84.up_proj", "model.language_model.layers.2.mlp.experts.85.up_proj", "model.language_model.layers.2.mlp.experts.86.up_proj", "model.language_model.layers.2.mlp.experts.87.up_proj", "model.language_model.layers.2.mlp.experts.88.up_proj", "model.language_model.layers.2.mlp.experts.89.up_proj", "model.language_model.layers.2.mlp.experts.90.up_proj", "model.language_model.layers.2.mlp.experts.91.up_proj", "model.language_model.layers.2.mlp.experts.92.up_proj", "model.language_model.layers.2.mlp.experts.93.up_proj", "model.language_model.layers.2.mlp.experts.94.up_proj", "model.language_model.layers.2.mlp.experts.95.up_proj", "model.language_model.layers.2.mlp.experts.96.up_proj", "model.language_model.layers.2.mlp.experts.97.up_proj", "model.language_model.layers.2.mlp.experts.98.up_proj", "model.language_model.layers.2.mlp.experts.99.up_proj", "model.language_model.layers.2.mlp.experts.100.up_proj", "model.language_model.layers.2.mlp.experts.101.up_proj", "model.language_model.layers.2.mlp.experts.102.up_proj", "model.language_model.layers.2.mlp.experts.103.up_proj", "model.language_model.layers.2.mlp.experts.104.up_proj", "model.language_model.layers.2.mlp.experts.105.up_proj", "model.language_model.layers.2.mlp.experts.106.up_proj", "model.language_model.layers.2.mlp.experts.107.up_proj", "model.language_model.layers.2.mlp.experts.108.up_proj", "model.language_model.layers.2.mlp.experts.109.up_proj", "model.language_model.layers.2.mlp.experts.110.up_proj", "model.language_model.layers.2.mlp.experts.111.up_proj", "model.language_model.layers.2.mlp.experts.112.up_proj", "model.language_model.layers.2.mlp.experts.113.up_proj", "model.language_model.layers.2.mlp.experts.114.up_proj", "model.language_model.layers.2.mlp.experts.115.up_proj", "model.language_model.layers.2.mlp.experts.116.up_proj", "model.language_model.layers.2.mlp.experts.117.up_proj", "model.language_model.layers.2.mlp.experts.118.up_proj", "model.language_model.layers.2.mlp.experts.119.up_proj", "model.language_model.layers.2.mlp.experts.120.up_proj", "model.language_model.layers.2.mlp.experts.121.up_proj", "model.language_model.layers.2.mlp.experts.122.up_proj", "model.language_model.layers.2.mlp.experts.123.up_proj", "model.language_model.layers.2.mlp.experts.124.up_proj", "model.language_model.layers.2.mlp.experts.125.up_proj", "model.language_model.layers.2.mlp.experts.126.up_proj", "model.language_model.layers.2.mlp.experts.127.up_proj", "model.language_model.layers.2.mlp.experts.0.down_proj", "model.language_model.layers.2.mlp.experts.1.down_proj", "model.language_model.layers.2.mlp.experts.2.down_proj", "model.language_model.layers.2.mlp.experts.3.down_proj", "model.language_model.layers.2.mlp.experts.4.down_proj", "model.language_model.layers.2.mlp.experts.5.down_proj", "model.language_model.layers.2.mlp.experts.6.down_proj", "model.language_model.layers.2.mlp.experts.7.down_proj", "model.language_model.layers.2.mlp.experts.8.down_proj", "model.language_model.layers.2.mlp.experts.9.down_proj", "model.language_model.layers.2.mlp.experts.10.down_proj", "model.language_model.layers.2.mlp.experts.11.down_proj", "model.language_model.layers.2.mlp.experts.12.down_proj", "model.language_model.layers.2.mlp.experts.13.down_proj", "model.language_model.layers.2.mlp.experts.14.down_proj", "model.language_model.layers.2.mlp.experts.15.down_proj", "model.language_model.layers.2.mlp.experts.16.down_proj", "model.language_model.layers.2.mlp.experts.17.down_proj", "model.language_model.layers.2.mlp.experts.18.down_proj", "model.language_model.layers.2.mlp.experts.19.down_proj", "model.language_model.layers.2.mlp.experts.20.down_proj", "model.language_model.layers.2.mlp.experts.21.down_proj", "model.language_model.layers.2.mlp.experts.22.down_proj", "model.language_model.layers.2.mlp.experts.23.down_proj", "model.language_model.layers.2.mlp.experts.24.down_proj", "model.language_model.layers.2.mlp.experts.25.down_proj", "model.language_model.layers.2.mlp.experts.26.down_proj", "model.language_model.layers.2.mlp.experts.27.down_proj", "model.language_model.layers.2.mlp.experts.28.down_proj", "model.language_model.layers.2.mlp.experts.29.down_proj", "model.language_model.layers.2.mlp.experts.30.down_proj", "model.language_model.layers.2.mlp.experts.31.down_proj", "model.language_model.layers.2.mlp.experts.32.down_proj", "model.language_model.layers.2.mlp.experts.33.down_proj", "model.language_model.layers.2.mlp.experts.34.down_proj", "model.language_model.layers.2.mlp.experts.35.down_proj", "model.language_model.layers.2.mlp.experts.36.down_proj", "model.language_model.layers.2.mlp.experts.37.down_proj", "model.language_model.layers.2.mlp.experts.38.down_proj", "model.language_model.layers.2.mlp.experts.39.down_proj", "model.language_model.layers.2.mlp.experts.40.down_proj", "model.language_model.layers.2.mlp.experts.41.down_proj", "model.language_model.layers.2.mlp.experts.42.down_proj", "model.language_model.layers.2.mlp.experts.43.down_proj", "model.language_model.layers.2.mlp.experts.44.down_proj", "model.language_model.layers.2.mlp.experts.45.down_proj", "model.language_model.layers.2.mlp.experts.46.down_proj", "model.language_model.layers.2.mlp.experts.47.down_proj", "model.language_model.layers.2.mlp.experts.48.down_proj", "model.language_model.layers.2.mlp.experts.49.down_proj", "model.language_model.layers.2.mlp.experts.50.down_proj", "model.language_model.layers.2.mlp.experts.51.down_proj", "model.language_model.layers.2.mlp.experts.52.down_proj", "model.language_model.layers.2.mlp.experts.53.down_proj", "model.language_model.layers.2.mlp.experts.54.down_proj", "model.language_model.layers.2.mlp.experts.55.down_proj", "model.language_model.layers.2.mlp.experts.56.down_proj", "model.language_model.layers.2.mlp.experts.57.down_proj", "model.language_model.layers.2.mlp.experts.58.down_proj", "model.language_model.layers.2.mlp.experts.59.down_proj", "model.language_model.layers.2.mlp.experts.60.down_proj", "model.language_model.layers.2.mlp.experts.61.down_proj", "model.language_model.layers.2.mlp.experts.62.down_proj", "model.language_model.layers.2.mlp.experts.63.down_proj", "model.language_model.layers.2.mlp.experts.64.down_proj", "model.language_model.layers.2.mlp.experts.65.down_proj", "model.language_model.layers.2.mlp.experts.66.down_proj", "model.language_model.layers.2.mlp.experts.67.down_proj", "model.language_model.layers.2.mlp.experts.68.down_proj", "model.language_model.layers.2.mlp.experts.69.down_proj", "model.language_model.layers.2.mlp.experts.70.down_proj", "model.language_model.layers.2.mlp.experts.71.down_proj", "model.language_model.layers.2.mlp.experts.72.down_proj", "model.language_model.layers.2.mlp.experts.73.down_proj", "model.language_model.layers.2.mlp.experts.74.down_proj", "model.language_model.layers.2.mlp.experts.75.down_proj", "model.language_model.layers.2.mlp.experts.76.down_proj", "model.language_model.layers.2.mlp.experts.77.down_proj", "model.language_model.layers.2.mlp.experts.78.down_proj", "model.language_model.layers.2.mlp.experts.79.down_proj", "model.language_model.layers.2.mlp.experts.80.down_proj", "model.language_model.layers.2.mlp.experts.81.down_proj", "model.language_model.layers.2.mlp.experts.82.down_proj", "model.language_model.layers.2.mlp.experts.83.down_proj", "model.language_model.layers.2.mlp.experts.84.down_proj", "model.language_model.layers.2.mlp.experts.85.down_proj", "model.language_model.layers.2.mlp.experts.86.down_proj", "model.language_model.layers.2.mlp.experts.87.down_proj", "model.language_model.layers.2.mlp.experts.88.down_proj", "model.language_model.layers.2.mlp.experts.89.down_proj", "model.language_model.layers.2.mlp.experts.90.down_proj", "model.language_model.layers.2.mlp.experts.91.down_proj", "model.language_model.layers.2.mlp.experts.92.down_proj", "model.language_model.layers.2.mlp.experts.93.down_proj", "model.language_model.layers.2.mlp.experts.94.down_proj", "model.language_model.layers.2.mlp.experts.95.down_proj", "model.language_model.layers.2.mlp.experts.96.down_proj", "model.language_model.layers.2.mlp.experts.97.down_proj", "model.language_model.layers.2.mlp.experts.98.down_proj", "model.language_model.layers.2.mlp.experts.99.down_proj", "model.language_model.layers.2.mlp.experts.100.down_proj", "model.language_model.layers.2.mlp.experts.101.down_proj", "model.language_model.layers.2.mlp.experts.102.down_proj", "model.language_model.layers.2.mlp.experts.103.down_proj", "model.language_model.layers.2.mlp.experts.104.down_proj", "model.language_model.layers.2.mlp.experts.105.down_proj", "model.language_model.layers.2.mlp.experts.106.down_proj", "model.language_model.layers.2.mlp.experts.107.down_proj", "model.language_model.layers.2.mlp.experts.108.down_proj", "model.language_model.layers.2.mlp.experts.109.down_proj", "model.language_model.layers.2.mlp.experts.110.down_proj", "model.language_model.layers.2.mlp.experts.111.down_proj", "model.language_model.layers.2.mlp.experts.112.down_proj", "model.language_model.layers.2.mlp.experts.113.down_proj", "model.language_model.layers.2.mlp.experts.114.down_proj", "model.language_model.layers.2.mlp.experts.115.down_proj", "model.language_model.layers.2.mlp.experts.116.down_proj", "model.language_model.layers.2.mlp.experts.117.down_proj", "model.language_model.layers.2.mlp.experts.118.down_proj", "model.language_model.layers.2.mlp.experts.119.down_proj", "model.language_model.layers.2.mlp.experts.120.down_proj", "model.language_model.layers.2.mlp.experts.121.down_proj", "model.language_model.layers.2.mlp.experts.122.down_proj", "model.language_model.layers.2.mlp.experts.123.down_proj", "model.language_model.layers.2.mlp.experts.124.down_proj", "model.language_model.layers.2.mlp.experts.125.down_proj", "model.language_model.layers.2.mlp.experts.126.down_proj", "model.language_model.layers.2.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.006383895780891186, "dbits": 603979776 } ] }, { "idx": 6, "layers": [ "model.language_model.layers.3.self_attn.q_proj", "model.language_model.layers.3.self_attn.k_proj", "model.language_model.layers.3.self_attn.v_proj", "model.language_model.layers.3.self_attn.o_proj" ], "candidates": [ { "dkld": -0.002999580930918455, "dbits": 18874368 } ] }, { "idx": 7, "layers": [ "model.language_model.layers.3.mlp.experts.0.gate_proj", "model.language_model.layers.3.mlp.experts.1.gate_proj", "model.language_model.layers.3.mlp.experts.2.gate_proj", "model.language_model.layers.3.mlp.experts.3.gate_proj", "model.language_model.layers.3.mlp.experts.4.gate_proj", "model.language_model.layers.3.mlp.experts.5.gate_proj", "model.language_model.layers.3.mlp.experts.6.gate_proj", "model.language_model.layers.3.mlp.experts.7.gate_proj", "model.language_model.layers.3.mlp.experts.8.gate_proj", "model.language_model.layers.3.mlp.experts.9.gate_proj", "model.language_model.layers.3.mlp.experts.10.gate_proj", "model.language_model.layers.3.mlp.experts.11.gate_proj", "model.language_model.layers.3.mlp.experts.12.gate_proj", "model.language_model.layers.3.mlp.experts.13.gate_proj", "model.language_model.layers.3.mlp.experts.14.gate_proj", "model.language_model.layers.3.mlp.experts.15.gate_proj", "model.language_model.layers.3.mlp.experts.16.gate_proj", "model.language_model.layers.3.mlp.experts.17.gate_proj", "model.language_model.layers.3.mlp.experts.18.gate_proj", "model.language_model.layers.3.mlp.experts.19.gate_proj", "model.language_model.layers.3.mlp.experts.20.gate_proj", "model.language_model.layers.3.mlp.experts.21.gate_proj", "model.language_model.layers.3.mlp.experts.22.gate_proj", "model.language_model.layers.3.mlp.experts.23.gate_proj", "model.language_model.layers.3.mlp.experts.24.gate_proj", "model.language_model.layers.3.mlp.experts.25.gate_proj", "model.language_model.layers.3.mlp.experts.26.gate_proj", "model.language_model.layers.3.mlp.experts.27.gate_proj", "model.language_model.layers.3.mlp.experts.28.gate_proj", "model.language_model.layers.3.mlp.experts.29.gate_proj", "model.language_model.layers.3.mlp.experts.30.gate_proj", "model.language_model.layers.3.mlp.experts.31.gate_proj", "model.language_model.layers.3.mlp.experts.32.gate_proj", "model.language_model.layers.3.mlp.experts.33.gate_proj", "model.language_model.layers.3.mlp.experts.34.gate_proj", "model.language_model.layers.3.mlp.experts.35.gate_proj", "model.language_model.layers.3.mlp.experts.36.gate_proj", "model.language_model.layers.3.mlp.experts.37.gate_proj", "model.language_model.layers.3.mlp.experts.38.gate_proj", "model.language_model.layers.3.mlp.experts.39.gate_proj", "model.language_model.layers.3.mlp.experts.40.gate_proj", "model.language_model.layers.3.mlp.experts.41.gate_proj", "model.language_model.layers.3.mlp.experts.42.gate_proj", "model.language_model.layers.3.mlp.experts.43.gate_proj", "model.language_model.layers.3.mlp.experts.44.gate_proj", "model.language_model.layers.3.mlp.experts.45.gate_proj", "model.language_model.layers.3.mlp.experts.46.gate_proj", "model.language_model.layers.3.mlp.experts.47.gate_proj", "model.language_model.layers.3.mlp.experts.48.gate_proj", "model.language_model.layers.3.mlp.experts.49.gate_proj", "model.language_model.layers.3.mlp.experts.50.gate_proj", "model.language_model.layers.3.mlp.experts.51.gate_proj", "model.language_model.layers.3.mlp.experts.52.gate_proj", "model.language_model.layers.3.mlp.experts.53.gate_proj", "model.language_model.layers.3.mlp.experts.54.gate_proj", "model.language_model.layers.3.mlp.experts.55.gate_proj", "model.language_model.layers.3.mlp.experts.56.gate_proj", "model.language_model.layers.3.mlp.experts.57.gate_proj", "model.language_model.layers.3.mlp.experts.58.gate_proj", "model.language_model.layers.3.mlp.experts.59.gate_proj", "model.language_model.layers.3.mlp.experts.60.gate_proj", "model.language_model.layers.3.mlp.experts.61.gate_proj", "model.language_model.layers.3.mlp.experts.62.gate_proj", "model.language_model.layers.3.mlp.experts.63.gate_proj", "model.language_model.layers.3.mlp.experts.64.gate_proj", "model.language_model.layers.3.mlp.experts.65.gate_proj", "model.language_model.layers.3.mlp.experts.66.gate_proj", "model.language_model.layers.3.mlp.experts.67.gate_proj", "model.language_model.layers.3.mlp.experts.68.gate_proj", "model.language_model.layers.3.mlp.experts.69.gate_proj", "model.language_model.layers.3.mlp.experts.70.gate_proj", "model.language_model.layers.3.mlp.experts.71.gate_proj", "model.language_model.layers.3.mlp.experts.72.gate_proj", "model.language_model.layers.3.mlp.experts.73.gate_proj", "model.language_model.layers.3.mlp.experts.74.gate_proj", "model.language_model.layers.3.mlp.experts.75.gate_proj", "model.language_model.layers.3.mlp.experts.76.gate_proj", "model.language_model.layers.3.mlp.experts.77.gate_proj", "model.language_model.layers.3.mlp.experts.78.gate_proj", "model.language_model.layers.3.mlp.experts.79.gate_proj", "model.language_model.layers.3.mlp.experts.80.gate_proj", "model.language_model.layers.3.mlp.experts.81.gate_proj", "model.language_model.layers.3.mlp.experts.82.gate_proj", "model.language_model.layers.3.mlp.experts.83.gate_proj", "model.language_model.layers.3.mlp.experts.84.gate_proj", "model.language_model.layers.3.mlp.experts.85.gate_proj", "model.language_model.layers.3.mlp.experts.86.gate_proj", "model.language_model.layers.3.mlp.experts.87.gate_proj", "model.language_model.layers.3.mlp.experts.88.gate_proj", "model.language_model.layers.3.mlp.experts.89.gate_proj", "model.language_model.layers.3.mlp.experts.90.gate_proj", "model.language_model.layers.3.mlp.experts.91.gate_proj", "model.language_model.layers.3.mlp.experts.92.gate_proj", "model.language_model.layers.3.mlp.experts.93.gate_proj", "model.language_model.layers.3.mlp.experts.94.gate_proj", "model.language_model.layers.3.mlp.experts.95.gate_proj", "model.language_model.layers.3.mlp.experts.96.gate_proj", "model.language_model.layers.3.mlp.experts.97.gate_proj", "model.language_model.layers.3.mlp.experts.98.gate_proj", "model.language_model.layers.3.mlp.experts.99.gate_proj", "model.language_model.layers.3.mlp.experts.100.gate_proj", "model.language_model.layers.3.mlp.experts.101.gate_proj", "model.language_model.layers.3.mlp.experts.102.gate_proj", "model.language_model.layers.3.mlp.experts.103.gate_proj", "model.language_model.layers.3.mlp.experts.104.gate_proj", "model.language_model.layers.3.mlp.experts.105.gate_proj", "model.language_model.layers.3.mlp.experts.106.gate_proj", "model.language_model.layers.3.mlp.experts.107.gate_proj", "model.language_model.layers.3.mlp.experts.108.gate_proj", "model.language_model.layers.3.mlp.experts.109.gate_proj", "model.language_model.layers.3.mlp.experts.110.gate_proj", "model.language_model.layers.3.mlp.experts.111.gate_proj", "model.language_model.layers.3.mlp.experts.112.gate_proj", "model.language_model.layers.3.mlp.experts.113.gate_proj", "model.language_model.layers.3.mlp.experts.114.gate_proj", "model.language_model.layers.3.mlp.experts.115.gate_proj", "model.language_model.layers.3.mlp.experts.116.gate_proj", "model.language_model.layers.3.mlp.experts.117.gate_proj", "model.language_model.layers.3.mlp.experts.118.gate_proj", "model.language_model.layers.3.mlp.experts.119.gate_proj", "model.language_model.layers.3.mlp.experts.120.gate_proj", "model.language_model.layers.3.mlp.experts.121.gate_proj", "model.language_model.layers.3.mlp.experts.122.gate_proj", "model.language_model.layers.3.mlp.experts.123.gate_proj", "model.language_model.layers.3.mlp.experts.124.gate_proj", "model.language_model.layers.3.mlp.experts.125.gate_proj", "model.language_model.layers.3.mlp.experts.126.gate_proj", "model.language_model.layers.3.mlp.experts.127.gate_proj", "model.language_model.layers.3.mlp.experts.0.up_proj", "model.language_model.layers.3.mlp.experts.1.up_proj", "model.language_model.layers.3.mlp.experts.2.up_proj", "model.language_model.layers.3.mlp.experts.3.up_proj", "model.language_model.layers.3.mlp.experts.4.up_proj", "model.language_model.layers.3.mlp.experts.5.up_proj", "model.language_model.layers.3.mlp.experts.6.up_proj", "model.language_model.layers.3.mlp.experts.7.up_proj", "model.language_model.layers.3.mlp.experts.8.up_proj", "model.language_model.layers.3.mlp.experts.9.up_proj", "model.language_model.layers.3.mlp.experts.10.up_proj", "model.language_model.layers.3.mlp.experts.11.up_proj", "model.language_model.layers.3.mlp.experts.12.up_proj", "model.language_model.layers.3.mlp.experts.13.up_proj", "model.language_model.layers.3.mlp.experts.14.up_proj", "model.language_model.layers.3.mlp.experts.15.up_proj", "model.language_model.layers.3.mlp.experts.16.up_proj", "model.language_model.layers.3.mlp.experts.17.up_proj", "model.language_model.layers.3.mlp.experts.18.up_proj", "model.language_model.layers.3.mlp.experts.19.up_proj", "model.language_model.layers.3.mlp.experts.20.up_proj", "model.language_model.layers.3.mlp.experts.21.up_proj", "model.language_model.layers.3.mlp.experts.22.up_proj", "model.language_model.layers.3.mlp.experts.23.up_proj", "model.language_model.layers.3.mlp.experts.24.up_proj", "model.language_model.layers.3.mlp.experts.25.up_proj", "model.language_model.layers.3.mlp.experts.26.up_proj", "model.language_model.layers.3.mlp.experts.27.up_proj", "model.language_model.layers.3.mlp.experts.28.up_proj", "model.language_model.layers.3.mlp.experts.29.up_proj", "model.language_model.layers.3.mlp.experts.30.up_proj", "model.language_model.layers.3.mlp.experts.31.up_proj", "model.language_model.layers.3.mlp.experts.32.up_proj", "model.language_model.layers.3.mlp.experts.33.up_proj", "model.language_model.layers.3.mlp.experts.34.up_proj", "model.language_model.layers.3.mlp.experts.35.up_proj", "model.language_model.layers.3.mlp.experts.36.up_proj", "model.language_model.layers.3.mlp.experts.37.up_proj", "model.language_model.layers.3.mlp.experts.38.up_proj", "model.language_model.layers.3.mlp.experts.39.up_proj", "model.language_model.layers.3.mlp.experts.40.up_proj", "model.language_model.layers.3.mlp.experts.41.up_proj", "model.language_model.layers.3.mlp.experts.42.up_proj", "model.language_model.layers.3.mlp.experts.43.up_proj", "model.language_model.layers.3.mlp.experts.44.up_proj", "model.language_model.layers.3.mlp.experts.45.up_proj", "model.language_model.layers.3.mlp.experts.46.up_proj", "model.language_model.layers.3.mlp.experts.47.up_proj", "model.language_model.layers.3.mlp.experts.48.up_proj", "model.language_model.layers.3.mlp.experts.49.up_proj", "model.language_model.layers.3.mlp.experts.50.up_proj", "model.language_model.layers.3.mlp.experts.51.up_proj", "model.language_model.layers.3.mlp.experts.52.up_proj", "model.language_model.layers.3.mlp.experts.53.up_proj", "model.language_model.layers.3.mlp.experts.54.up_proj", "model.language_model.layers.3.mlp.experts.55.up_proj", "model.language_model.layers.3.mlp.experts.56.up_proj", "model.language_model.layers.3.mlp.experts.57.up_proj", "model.language_model.layers.3.mlp.experts.58.up_proj", "model.language_model.layers.3.mlp.experts.59.up_proj", "model.language_model.layers.3.mlp.experts.60.up_proj", "model.language_model.layers.3.mlp.experts.61.up_proj", "model.language_model.layers.3.mlp.experts.62.up_proj", "model.language_model.layers.3.mlp.experts.63.up_proj", "model.language_model.layers.3.mlp.experts.64.up_proj", "model.language_model.layers.3.mlp.experts.65.up_proj", "model.language_model.layers.3.mlp.experts.66.up_proj", "model.language_model.layers.3.mlp.experts.67.up_proj", "model.language_model.layers.3.mlp.experts.68.up_proj", "model.language_model.layers.3.mlp.experts.69.up_proj", "model.language_model.layers.3.mlp.experts.70.up_proj", "model.language_model.layers.3.mlp.experts.71.up_proj", "model.language_model.layers.3.mlp.experts.72.up_proj", "model.language_model.layers.3.mlp.experts.73.up_proj", "model.language_model.layers.3.mlp.experts.74.up_proj", "model.language_model.layers.3.mlp.experts.75.up_proj", "model.language_model.layers.3.mlp.experts.76.up_proj", "model.language_model.layers.3.mlp.experts.77.up_proj", "model.language_model.layers.3.mlp.experts.78.up_proj", "model.language_model.layers.3.mlp.experts.79.up_proj", "model.language_model.layers.3.mlp.experts.80.up_proj", "model.language_model.layers.3.mlp.experts.81.up_proj", "model.language_model.layers.3.mlp.experts.82.up_proj", "model.language_model.layers.3.mlp.experts.83.up_proj", "model.language_model.layers.3.mlp.experts.84.up_proj", "model.language_model.layers.3.mlp.experts.85.up_proj", "model.language_model.layers.3.mlp.experts.86.up_proj", "model.language_model.layers.3.mlp.experts.87.up_proj", "model.language_model.layers.3.mlp.experts.88.up_proj", "model.language_model.layers.3.mlp.experts.89.up_proj", "model.language_model.layers.3.mlp.experts.90.up_proj", "model.language_model.layers.3.mlp.experts.91.up_proj", "model.language_model.layers.3.mlp.experts.92.up_proj", "model.language_model.layers.3.mlp.experts.93.up_proj", "model.language_model.layers.3.mlp.experts.94.up_proj", "model.language_model.layers.3.mlp.experts.95.up_proj", "model.language_model.layers.3.mlp.experts.96.up_proj", "model.language_model.layers.3.mlp.experts.97.up_proj", "model.language_model.layers.3.mlp.experts.98.up_proj", "model.language_model.layers.3.mlp.experts.99.up_proj", "model.language_model.layers.3.mlp.experts.100.up_proj", "model.language_model.layers.3.mlp.experts.101.up_proj", "model.language_model.layers.3.mlp.experts.102.up_proj", "model.language_model.layers.3.mlp.experts.103.up_proj", "model.language_model.layers.3.mlp.experts.104.up_proj", "model.language_model.layers.3.mlp.experts.105.up_proj", "model.language_model.layers.3.mlp.experts.106.up_proj", "model.language_model.layers.3.mlp.experts.107.up_proj", "model.language_model.layers.3.mlp.experts.108.up_proj", "model.language_model.layers.3.mlp.experts.109.up_proj", "model.language_model.layers.3.mlp.experts.110.up_proj", "model.language_model.layers.3.mlp.experts.111.up_proj", "model.language_model.layers.3.mlp.experts.112.up_proj", "model.language_model.layers.3.mlp.experts.113.up_proj", "model.language_model.layers.3.mlp.experts.114.up_proj", "model.language_model.layers.3.mlp.experts.115.up_proj", "model.language_model.layers.3.mlp.experts.116.up_proj", "model.language_model.layers.3.mlp.experts.117.up_proj", "model.language_model.layers.3.mlp.experts.118.up_proj", "model.language_model.layers.3.mlp.experts.119.up_proj", "model.language_model.layers.3.mlp.experts.120.up_proj", "model.language_model.layers.3.mlp.experts.121.up_proj", "model.language_model.layers.3.mlp.experts.122.up_proj", "model.language_model.layers.3.mlp.experts.123.up_proj", "model.language_model.layers.3.mlp.experts.124.up_proj", "model.language_model.layers.3.mlp.experts.125.up_proj", "model.language_model.layers.3.mlp.experts.126.up_proj", "model.language_model.layers.3.mlp.experts.127.up_proj", "model.language_model.layers.3.mlp.experts.0.down_proj", "model.language_model.layers.3.mlp.experts.1.down_proj", "model.language_model.layers.3.mlp.experts.2.down_proj", "model.language_model.layers.3.mlp.experts.3.down_proj", "model.language_model.layers.3.mlp.experts.4.down_proj", "model.language_model.layers.3.mlp.experts.5.down_proj", "model.language_model.layers.3.mlp.experts.6.down_proj", "model.language_model.layers.3.mlp.experts.7.down_proj", "model.language_model.layers.3.mlp.experts.8.down_proj", "model.language_model.layers.3.mlp.experts.9.down_proj", "model.language_model.layers.3.mlp.experts.10.down_proj", "model.language_model.layers.3.mlp.experts.11.down_proj", "model.language_model.layers.3.mlp.experts.12.down_proj", "model.language_model.layers.3.mlp.experts.13.down_proj", "model.language_model.layers.3.mlp.experts.14.down_proj", "model.language_model.layers.3.mlp.experts.15.down_proj", "model.language_model.layers.3.mlp.experts.16.down_proj", "model.language_model.layers.3.mlp.experts.17.down_proj", "model.language_model.layers.3.mlp.experts.18.down_proj", "model.language_model.layers.3.mlp.experts.19.down_proj", "model.language_model.layers.3.mlp.experts.20.down_proj", "model.language_model.layers.3.mlp.experts.21.down_proj", "model.language_model.layers.3.mlp.experts.22.down_proj", "model.language_model.layers.3.mlp.experts.23.down_proj", "model.language_model.layers.3.mlp.experts.24.down_proj", "model.language_model.layers.3.mlp.experts.25.down_proj", "model.language_model.layers.3.mlp.experts.26.down_proj", "model.language_model.layers.3.mlp.experts.27.down_proj", "model.language_model.layers.3.mlp.experts.28.down_proj", "model.language_model.layers.3.mlp.experts.29.down_proj", "model.language_model.layers.3.mlp.experts.30.down_proj", "model.language_model.layers.3.mlp.experts.31.down_proj", "model.language_model.layers.3.mlp.experts.32.down_proj", "model.language_model.layers.3.mlp.experts.33.down_proj", "model.language_model.layers.3.mlp.experts.34.down_proj", "model.language_model.layers.3.mlp.experts.35.down_proj", "model.language_model.layers.3.mlp.experts.36.down_proj", "model.language_model.layers.3.mlp.experts.37.down_proj", "model.language_model.layers.3.mlp.experts.38.down_proj", "model.language_model.layers.3.mlp.experts.39.down_proj", "model.language_model.layers.3.mlp.experts.40.down_proj", "model.language_model.layers.3.mlp.experts.41.down_proj", "model.language_model.layers.3.mlp.experts.42.down_proj", "model.language_model.layers.3.mlp.experts.43.down_proj", "model.language_model.layers.3.mlp.experts.44.down_proj", "model.language_model.layers.3.mlp.experts.45.down_proj", "model.language_model.layers.3.mlp.experts.46.down_proj", "model.language_model.layers.3.mlp.experts.47.down_proj", "model.language_model.layers.3.mlp.experts.48.down_proj", "model.language_model.layers.3.mlp.experts.49.down_proj", "model.language_model.layers.3.mlp.experts.50.down_proj", "model.language_model.layers.3.mlp.experts.51.down_proj", "model.language_model.layers.3.mlp.experts.52.down_proj", "model.language_model.layers.3.mlp.experts.53.down_proj", "model.language_model.layers.3.mlp.experts.54.down_proj", "model.language_model.layers.3.mlp.experts.55.down_proj", "model.language_model.layers.3.mlp.experts.56.down_proj", "model.language_model.layers.3.mlp.experts.57.down_proj", "model.language_model.layers.3.mlp.experts.58.down_proj", "model.language_model.layers.3.mlp.experts.59.down_proj", "model.language_model.layers.3.mlp.experts.60.down_proj", "model.language_model.layers.3.mlp.experts.61.down_proj", "model.language_model.layers.3.mlp.experts.62.down_proj", "model.language_model.layers.3.mlp.experts.63.down_proj", "model.language_model.layers.3.mlp.experts.64.down_proj", "model.language_model.layers.3.mlp.experts.65.down_proj", "model.language_model.layers.3.mlp.experts.66.down_proj", "model.language_model.layers.3.mlp.experts.67.down_proj", "model.language_model.layers.3.mlp.experts.68.down_proj", "model.language_model.layers.3.mlp.experts.69.down_proj", "model.language_model.layers.3.mlp.experts.70.down_proj", "model.language_model.layers.3.mlp.experts.71.down_proj", "model.language_model.layers.3.mlp.experts.72.down_proj", "model.language_model.layers.3.mlp.experts.73.down_proj", "model.language_model.layers.3.mlp.experts.74.down_proj", "model.language_model.layers.3.mlp.experts.75.down_proj", "model.language_model.layers.3.mlp.experts.76.down_proj", "model.language_model.layers.3.mlp.experts.77.down_proj", "model.language_model.layers.3.mlp.experts.78.down_proj", "model.language_model.layers.3.mlp.experts.79.down_proj", "model.language_model.layers.3.mlp.experts.80.down_proj", "model.language_model.layers.3.mlp.experts.81.down_proj", "model.language_model.layers.3.mlp.experts.82.down_proj", "model.language_model.layers.3.mlp.experts.83.down_proj", "model.language_model.layers.3.mlp.experts.84.down_proj", "model.language_model.layers.3.mlp.experts.85.down_proj", "model.language_model.layers.3.mlp.experts.86.down_proj", "model.language_model.layers.3.mlp.experts.87.down_proj", "model.language_model.layers.3.mlp.experts.88.down_proj", "model.language_model.layers.3.mlp.experts.89.down_proj", "model.language_model.layers.3.mlp.experts.90.down_proj", "model.language_model.layers.3.mlp.experts.91.down_proj", "model.language_model.layers.3.mlp.experts.92.down_proj", "model.language_model.layers.3.mlp.experts.93.down_proj", "model.language_model.layers.3.mlp.experts.94.down_proj", "model.language_model.layers.3.mlp.experts.95.down_proj", "model.language_model.layers.3.mlp.experts.96.down_proj", "model.language_model.layers.3.mlp.experts.97.down_proj", "model.language_model.layers.3.mlp.experts.98.down_proj", "model.language_model.layers.3.mlp.experts.99.down_proj", "model.language_model.layers.3.mlp.experts.100.down_proj", "model.language_model.layers.3.mlp.experts.101.down_proj", "model.language_model.layers.3.mlp.experts.102.down_proj", "model.language_model.layers.3.mlp.experts.103.down_proj", "model.language_model.layers.3.mlp.experts.104.down_proj", "model.language_model.layers.3.mlp.experts.105.down_proj", "model.language_model.layers.3.mlp.experts.106.down_proj", "model.language_model.layers.3.mlp.experts.107.down_proj", "model.language_model.layers.3.mlp.experts.108.down_proj", "model.language_model.layers.3.mlp.experts.109.down_proj", "model.language_model.layers.3.mlp.experts.110.down_proj", "model.language_model.layers.3.mlp.experts.111.down_proj", "model.language_model.layers.3.mlp.experts.112.down_proj", "model.language_model.layers.3.mlp.experts.113.down_proj", "model.language_model.layers.3.mlp.experts.114.down_proj", "model.language_model.layers.3.mlp.experts.115.down_proj", "model.language_model.layers.3.mlp.experts.116.down_proj", "model.language_model.layers.3.mlp.experts.117.down_proj", "model.language_model.layers.3.mlp.experts.118.down_proj", "model.language_model.layers.3.mlp.experts.119.down_proj", "model.language_model.layers.3.mlp.experts.120.down_proj", "model.language_model.layers.3.mlp.experts.121.down_proj", "model.language_model.layers.3.mlp.experts.122.down_proj", "model.language_model.layers.3.mlp.experts.123.down_proj", "model.language_model.layers.3.mlp.experts.124.down_proj", "model.language_model.layers.3.mlp.experts.125.down_proj", "model.language_model.layers.3.mlp.experts.126.down_proj", "model.language_model.layers.3.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.005279565975069997, "dbits": 603979776 } ] }, { "idx": 8, "layers": [ "model.language_model.layers.4.self_attn.q_proj", "model.language_model.layers.4.self_attn.k_proj", "model.language_model.layers.4.self_attn.v_proj", "model.language_model.layers.4.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0009297920390963499, "dbits": 18874368 } ] }, { "idx": 9, "layers": [ "model.language_model.layers.4.mlp.experts.0.gate_proj", "model.language_model.layers.4.mlp.experts.1.gate_proj", "model.language_model.layers.4.mlp.experts.2.gate_proj", "model.language_model.layers.4.mlp.experts.3.gate_proj", "model.language_model.layers.4.mlp.experts.4.gate_proj", "model.language_model.layers.4.mlp.experts.5.gate_proj", "model.language_model.layers.4.mlp.experts.6.gate_proj", "model.language_model.layers.4.mlp.experts.7.gate_proj", "model.language_model.layers.4.mlp.experts.8.gate_proj", "model.language_model.layers.4.mlp.experts.9.gate_proj", "model.language_model.layers.4.mlp.experts.10.gate_proj", "model.language_model.layers.4.mlp.experts.11.gate_proj", "model.language_model.layers.4.mlp.experts.12.gate_proj", "model.language_model.layers.4.mlp.experts.13.gate_proj", "model.language_model.layers.4.mlp.experts.14.gate_proj", "model.language_model.layers.4.mlp.experts.15.gate_proj", "model.language_model.layers.4.mlp.experts.16.gate_proj", "model.language_model.layers.4.mlp.experts.17.gate_proj", "model.language_model.layers.4.mlp.experts.18.gate_proj", "model.language_model.layers.4.mlp.experts.19.gate_proj", "model.language_model.layers.4.mlp.experts.20.gate_proj", "model.language_model.layers.4.mlp.experts.21.gate_proj", "model.language_model.layers.4.mlp.experts.22.gate_proj", "model.language_model.layers.4.mlp.experts.23.gate_proj", "model.language_model.layers.4.mlp.experts.24.gate_proj", "model.language_model.layers.4.mlp.experts.25.gate_proj", "model.language_model.layers.4.mlp.experts.26.gate_proj", "model.language_model.layers.4.mlp.experts.27.gate_proj", "model.language_model.layers.4.mlp.experts.28.gate_proj", "model.language_model.layers.4.mlp.experts.29.gate_proj", "model.language_model.layers.4.mlp.experts.30.gate_proj", "model.language_model.layers.4.mlp.experts.31.gate_proj", "model.language_model.layers.4.mlp.experts.32.gate_proj", "model.language_model.layers.4.mlp.experts.33.gate_proj", "model.language_model.layers.4.mlp.experts.34.gate_proj", "model.language_model.layers.4.mlp.experts.35.gate_proj", "model.language_model.layers.4.mlp.experts.36.gate_proj", "model.language_model.layers.4.mlp.experts.37.gate_proj", "model.language_model.layers.4.mlp.experts.38.gate_proj", "model.language_model.layers.4.mlp.experts.39.gate_proj", "model.language_model.layers.4.mlp.experts.40.gate_proj", "model.language_model.layers.4.mlp.experts.41.gate_proj", "model.language_model.layers.4.mlp.experts.42.gate_proj", "model.language_model.layers.4.mlp.experts.43.gate_proj", "model.language_model.layers.4.mlp.experts.44.gate_proj", "model.language_model.layers.4.mlp.experts.45.gate_proj", "model.language_model.layers.4.mlp.experts.46.gate_proj", "model.language_model.layers.4.mlp.experts.47.gate_proj", "model.language_model.layers.4.mlp.experts.48.gate_proj", "model.language_model.layers.4.mlp.experts.49.gate_proj", "model.language_model.layers.4.mlp.experts.50.gate_proj", "model.language_model.layers.4.mlp.experts.51.gate_proj", "model.language_model.layers.4.mlp.experts.52.gate_proj", "model.language_model.layers.4.mlp.experts.53.gate_proj", "model.language_model.layers.4.mlp.experts.54.gate_proj", "model.language_model.layers.4.mlp.experts.55.gate_proj", "model.language_model.layers.4.mlp.experts.56.gate_proj", "model.language_model.layers.4.mlp.experts.57.gate_proj", "model.language_model.layers.4.mlp.experts.58.gate_proj", "model.language_model.layers.4.mlp.experts.59.gate_proj", "model.language_model.layers.4.mlp.experts.60.gate_proj", "model.language_model.layers.4.mlp.experts.61.gate_proj", "model.language_model.layers.4.mlp.experts.62.gate_proj", "model.language_model.layers.4.mlp.experts.63.gate_proj", "model.language_model.layers.4.mlp.experts.64.gate_proj", "model.language_model.layers.4.mlp.experts.65.gate_proj", "model.language_model.layers.4.mlp.experts.66.gate_proj", "model.language_model.layers.4.mlp.experts.67.gate_proj", "model.language_model.layers.4.mlp.experts.68.gate_proj", "model.language_model.layers.4.mlp.experts.69.gate_proj", "model.language_model.layers.4.mlp.experts.70.gate_proj", "model.language_model.layers.4.mlp.experts.71.gate_proj", "model.language_model.layers.4.mlp.experts.72.gate_proj", "model.language_model.layers.4.mlp.experts.73.gate_proj", "model.language_model.layers.4.mlp.experts.74.gate_proj", "model.language_model.layers.4.mlp.experts.75.gate_proj", "model.language_model.layers.4.mlp.experts.76.gate_proj", "model.language_model.layers.4.mlp.experts.77.gate_proj", "model.language_model.layers.4.mlp.experts.78.gate_proj", "model.language_model.layers.4.mlp.experts.79.gate_proj", "model.language_model.layers.4.mlp.experts.80.gate_proj", "model.language_model.layers.4.mlp.experts.81.gate_proj", "model.language_model.layers.4.mlp.experts.82.gate_proj", "model.language_model.layers.4.mlp.experts.83.gate_proj", "model.language_model.layers.4.mlp.experts.84.gate_proj", "model.language_model.layers.4.mlp.experts.85.gate_proj", "model.language_model.layers.4.mlp.experts.86.gate_proj", "model.language_model.layers.4.mlp.experts.87.gate_proj", "model.language_model.layers.4.mlp.experts.88.gate_proj", "model.language_model.layers.4.mlp.experts.89.gate_proj", "model.language_model.layers.4.mlp.experts.90.gate_proj", "model.language_model.layers.4.mlp.experts.91.gate_proj", "model.language_model.layers.4.mlp.experts.92.gate_proj", "model.language_model.layers.4.mlp.experts.93.gate_proj", "model.language_model.layers.4.mlp.experts.94.gate_proj", "model.language_model.layers.4.mlp.experts.95.gate_proj", "model.language_model.layers.4.mlp.experts.96.gate_proj", "model.language_model.layers.4.mlp.experts.97.gate_proj", "model.language_model.layers.4.mlp.experts.98.gate_proj", "model.language_model.layers.4.mlp.experts.99.gate_proj", "model.language_model.layers.4.mlp.experts.100.gate_proj", "model.language_model.layers.4.mlp.experts.101.gate_proj", "model.language_model.layers.4.mlp.experts.102.gate_proj", "model.language_model.layers.4.mlp.experts.103.gate_proj", "model.language_model.layers.4.mlp.experts.104.gate_proj", "model.language_model.layers.4.mlp.experts.105.gate_proj", "model.language_model.layers.4.mlp.experts.106.gate_proj", "model.language_model.layers.4.mlp.experts.107.gate_proj", "model.language_model.layers.4.mlp.experts.108.gate_proj", "model.language_model.layers.4.mlp.experts.109.gate_proj", "model.language_model.layers.4.mlp.experts.110.gate_proj", "model.language_model.layers.4.mlp.experts.111.gate_proj", "model.language_model.layers.4.mlp.experts.112.gate_proj", "model.language_model.layers.4.mlp.experts.113.gate_proj", "model.language_model.layers.4.mlp.experts.114.gate_proj", "model.language_model.layers.4.mlp.experts.115.gate_proj", "model.language_model.layers.4.mlp.experts.116.gate_proj", "model.language_model.layers.4.mlp.experts.117.gate_proj", "model.language_model.layers.4.mlp.experts.118.gate_proj", "model.language_model.layers.4.mlp.experts.119.gate_proj", "model.language_model.layers.4.mlp.experts.120.gate_proj", "model.language_model.layers.4.mlp.experts.121.gate_proj", "model.language_model.layers.4.mlp.experts.122.gate_proj", "model.language_model.layers.4.mlp.experts.123.gate_proj", "model.language_model.layers.4.mlp.experts.124.gate_proj", "model.language_model.layers.4.mlp.experts.125.gate_proj", "model.language_model.layers.4.mlp.experts.126.gate_proj", "model.language_model.layers.4.mlp.experts.127.gate_proj", "model.language_model.layers.4.mlp.experts.0.up_proj", "model.language_model.layers.4.mlp.experts.1.up_proj", "model.language_model.layers.4.mlp.experts.2.up_proj", "model.language_model.layers.4.mlp.experts.3.up_proj", "model.language_model.layers.4.mlp.experts.4.up_proj", "model.language_model.layers.4.mlp.experts.5.up_proj", "model.language_model.layers.4.mlp.experts.6.up_proj", "model.language_model.layers.4.mlp.experts.7.up_proj", "model.language_model.layers.4.mlp.experts.8.up_proj", "model.language_model.layers.4.mlp.experts.9.up_proj", "model.language_model.layers.4.mlp.experts.10.up_proj", "model.language_model.layers.4.mlp.experts.11.up_proj", "model.language_model.layers.4.mlp.experts.12.up_proj", "model.language_model.layers.4.mlp.experts.13.up_proj", "model.language_model.layers.4.mlp.experts.14.up_proj", "model.language_model.layers.4.mlp.experts.15.up_proj", "model.language_model.layers.4.mlp.experts.16.up_proj", "model.language_model.layers.4.mlp.experts.17.up_proj", "model.language_model.layers.4.mlp.experts.18.up_proj", "model.language_model.layers.4.mlp.experts.19.up_proj", "model.language_model.layers.4.mlp.experts.20.up_proj", "model.language_model.layers.4.mlp.experts.21.up_proj", "model.language_model.layers.4.mlp.experts.22.up_proj", "model.language_model.layers.4.mlp.experts.23.up_proj", "model.language_model.layers.4.mlp.experts.24.up_proj", "model.language_model.layers.4.mlp.experts.25.up_proj", "model.language_model.layers.4.mlp.experts.26.up_proj", "model.language_model.layers.4.mlp.experts.27.up_proj", "model.language_model.layers.4.mlp.experts.28.up_proj", "model.language_model.layers.4.mlp.experts.29.up_proj", "model.language_model.layers.4.mlp.experts.30.up_proj", "model.language_model.layers.4.mlp.experts.31.up_proj", "model.language_model.layers.4.mlp.experts.32.up_proj", "model.language_model.layers.4.mlp.experts.33.up_proj", "model.language_model.layers.4.mlp.experts.34.up_proj", "model.language_model.layers.4.mlp.experts.35.up_proj", "model.language_model.layers.4.mlp.experts.36.up_proj", "model.language_model.layers.4.mlp.experts.37.up_proj", "model.language_model.layers.4.mlp.experts.38.up_proj", "model.language_model.layers.4.mlp.experts.39.up_proj", "model.language_model.layers.4.mlp.experts.40.up_proj", "model.language_model.layers.4.mlp.experts.41.up_proj", "model.language_model.layers.4.mlp.experts.42.up_proj", "model.language_model.layers.4.mlp.experts.43.up_proj", "model.language_model.layers.4.mlp.experts.44.up_proj", "model.language_model.layers.4.mlp.experts.45.up_proj", "model.language_model.layers.4.mlp.experts.46.up_proj", "model.language_model.layers.4.mlp.experts.47.up_proj", "model.language_model.layers.4.mlp.experts.48.up_proj", "model.language_model.layers.4.mlp.experts.49.up_proj", "model.language_model.layers.4.mlp.experts.50.up_proj", "model.language_model.layers.4.mlp.experts.51.up_proj", "model.language_model.layers.4.mlp.experts.52.up_proj", "model.language_model.layers.4.mlp.experts.53.up_proj", "model.language_model.layers.4.mlp.experts.54.up_proj", "model.language_model.layers.4.mlp.experts.55.up_proj", "model.language_model.layers.4.mlp.experts.56.up_proj", "model.language_model.layers.4.mlp.experts.57.up_proj", "model.language_model.layers.4.mlp.experts.58.up_proj", "model.language_model.layers.4.mlp.experts.59.up_proj", "model.language_model.layers.4.mlp.experts.60.up_proj", "model.language_model.layers.4.mlp.experts.61.up_proj", "model.language_model.layers.4.mlp.experts.62.up_proj", "model.language_model.layers.4.mlp.experts.63.up_proj", "model.language_model.layers.4.mlp.experts.64.up_proj", "model.language_model.layers.4.mlp.experts.65.up_proj", "model.language_model.layers.4.mlp.experts.66.up_proj", "model.language_model.layers.4.mlp.experts.67.up_proj", "model.language_model.layers.4.mlp.experts.68.up_proj", "model.language_model.layers.4.mlp.experts.69.up_proj", "model.language_model.layers.4.mlp.experts.70.up_proj", "model.language_model.layers.4.mlp.experts.71.up_proj", "model.language_model.layers.4.mlp.experts.72.up_proj", "model.language_model.layers.4.mlp.experts.73.up_proj", "model.language_model.layers.4.mlp.experts.74.up_proj", "model.language_model.layers.4.mlp.experts.75.up_proj", "model.language_model.layers.4.mlp.experts.76.up_proj", "model.language_model.layers.4.mlp.experts.77.up_proj", "model.language_model.layers.4.mlp.experts.78.up_proj", "model.language_model.layers.4.mlp.experts.79.up_proj", "model.language_model.layers.4.mlp.experts.80.up_proj", "model.language_model.layers.4.mlp.experts.81.up_proj", "model.language_model.layers.4.mlp.experts.82.up_proj", "model.language_model.layers.4.mlp.experts.83.up_proj", "model.language_model.layers.4.mlp.experts.84.up_proj", "model.language_model.layers.4.mlp.experts.85.up_proj", "model.language_model.layers.4.mlp.experts.86.up_proj", "model.language_model.layers.4.mlp.experts.87.up_proj", "model.language_model.layers.4.mlp.experts.88.up_proj", "model.language_model.layers.4.mlp.experts.89.up_proj", "model.language_model.layers.4.mlp.experts.90.up_proj", "model.language_model.layers.4.mlp.experts.91.up_proj", "model.language_model.layers.4.mlp.experts.92.up_proj", "model.language_model.layers.4.mlp.experts.93.up_proj", "model.language_model.layers.4.mlp.experts.94.up_proj", "model.language_model.layers.4.mlp.experts.95.up_proj", "model.language_model.layers.4.mlp.experts.96.up_proj", "model.language_model.layers.4.mlp.experts.97.up_proj", "model.language_model.layers.4.mlp.experts.98.up_proj", "model.language_model.layers.4.mlp.experts.99.up_proj", "model.language_model.layers.4.mlp.experts.100.up_proj", "model.language_model.layers.4.mlp.experts.101.up_proj", "model.language_model.layers.4.mlp.experts.102.up_proj", "model.language_model.layers.4.mlp.experts.103.up_proj", "model.language_model.layers.4.mlp.experts.104.up_proj", "model.language_model.layers.4.mlp.experts.105.up_proj", "model.language_model.layers.4.mlp.experts.106.up_proj", "model.language_model.layers.4.mlp.experts.107.up_proj", "model.language_model.layers.4.mlp.experts.108.up_proj", "model.language_model.layers.4.mlp.experts.109.up_proj", "model.language_model.layers.4.mlp.experts.110.up_proj", "model.language_model.layers.4.mlp.experts.111.up_proj", "model.language_model.layers.4.mlp.experts.112.up_proj", "model.language_model.layers.4.mlp.experts.113.up_proj", "model.language_model.layers.4.mlp.experts.114.up_proj", "model.language_model.layers.4.mlp.experts.115.up_proj", "model.language_model.layers.4.mlp.experts.116.up_proj", "model.language_model.layers.4.mlp.experts.117.up_proj", "model.language_model.layers.4.mlp.experts.118.up_proj", "model.language_model.layers.4.mlp.experts.119.up_proj", "model.language_model.layers.4.mlp.experts.120.up_proj", "model.language_model.layers.4.mlp.experts.121.up_proj", "model.language_model.layers.4.mlp.experts.122.up_proj", "model.language_model.layers.4.mlp.experts.123.up_proj", "model.language_model.layers.4.mlp.experts.124.up_proj", "model.language_model.layers.4.mlp.experts.125.up_proj", "model.language_model.layers.4.mlp.experts.126.up_proj", "model.language_model.layers.4.mlp.experts.127.up_proj", "model.language_model.layers.4.mlp.experts.0.down_proj", "model.language_model.layers.4.mlp.experts.1.down_proj", "model.language_model.layers.4.mlp.experts.2.down_proj", "model.language_model.layers.4.mlp.experts.3.down_proj", "model.language_model.layers.4.mlp.experts.4.down_proj", "model.language_model.layers.4.mlp.experts.5.down_proj", "model.language_model.layers.4.mlp.experts.6.down_proj", "model.language_model.layers.4.mlp.experts.7.down_proj", "model.language_model.layers.4.mlp.experts.8.down_proj", "model.language_model.layers.4.mlp.experts.9.down_proj", "model.language_model.layers.4.mlp.experts.10.down_proj", "model.language_model.layers.4.mlp.experts.11.down_proj", "model.language_model.layers.4.mlp.experts.12.down_proj", "model.language_model.layers.4.mlp.experts.13.down_proj", "model.language_model.layers.4.mlp.experts.14.down_proj", "model.language_model.layers.4.mlp.experts.15.down_proj", "model.language_model.layers.4.mlp.experts.16.down_proj", "model.language_model.layers.4.mlp.experts.17.down_proj", "model.language_model.layers.4.mlp.experts.18.down_proj", "model.language_model.layers.4.mlp.experts.19.down_proj", "model.language_model.layers.4.mlp.experts.20.down_proj", "model.language_model.layers.4.mlp.experts.21.down_proj", "model.language_model.layers.4.mlp.experts.22.down_proj", "model.language_model.layers.4.mlp.experts.23.down_proj", "model.language_model.layers.4.mlp.experts.24.down_proj", "model.language_model.layers.4.mlp.experts.25.down_proj", "model.language_model.layers.4.mlp.experts.26.down_proj", "model.language_model.layers.4.mlp.experts.27.down_proj", "model.language_model.layers.4.mlp.experts.28.down_proj", "model.language_model.layers.4.mlp.experts.29.down_proj", "model.language_model.layers.4.mlp.experts.30.down_proj", "model.language_model.layers.4.mlp.experts.31.down_proj", "model.language_model.layers.4.mlp.experts.32.down_proj", "model.language_model.layers.4.mlp.experts.33.down_proj", "model.language_model.layers.4.mlp.experts.34.down_proj", "model.language_model.layers.4.mlp.experts.35.down_proj", "model.language_model.layers.4.mlp.experts.36.down_proj", "model.language_model.layers.4.mlp.experts.37.down_proj", "model.language_model.layers.4.mlp.experts.38.down_proj", "model.language_model.layers.4.mlp.experts.39.down_proj", "model.language_model.layers.4.mlp.experts.40.down_proj", "model.language_model.layers.4.mlp.experts.41.down_proj", "model.language_model.layers.4.mlp.experts.42.down_proj", "model.language_model.layers.4.mlp.experts.43.down_proj", "model.language_model.layers.4.mlp.experts.44.down_proj", "model.language_model.layers.4.mlp.experts.45.down_proj", "model.language_model.layers.4.mlp.experts.46.down_proj", "model.language_model.layers.4.mlp.experts.47.down_proj", "model.language_model.layers.4.mlp.experts.48.down_proj", "model.language_model.layers.4.mlp.experts.49.down_proj", "model.language_model.layers.4.mlp.experts.50.down_proj", "model.language_model.layers.4.mlp.experts.51.down_proj", "model.language_model.layers.4.mlp.experts.52.down_proj", "model.language_model.layers.4.mlp.experts.53.down_proj", "model.language_model.layers.4.mlp.experts.54.down_proj", "model.language_model.layers.4.mlp.experts.55.down_proj", "model.language_model.layers.4.mlp.experts.56.down_proj", "model.language_model.layers.4.mlp.experts.57.down_proj", "model.language_model.layers.4.mlp.experts.58.down_proj", "model.language_model.layers.4.mlp.experts.59.down_proj", "model.language_model.layers.4.mlp.experts.60.down_proj", "model.language_model.layers.4.mlp.experts.61.down_proj", "model.language_model.layers.4.mlp.experts.62.down_proj", "model.language_model.layers.4.mlp.experts.63.down_proj", "model.language_model.layers.4.mlp.experts.64.down_proj", "model.language_model.layers.4.mlp.experts.65.down_proj", "model.language_model.layers.4.mlp.experts.66.down_proj", "model.language_model.layers.4.mlp.experts.67.down_proj", "model.language_model.layers.4.mlp.experts.68.down_proj", "model.language_model.layers.4.mlp.experts.69.down_proj", "model.language_model.layers.4.mlp.experts.70.down_proj", "model.language_model.layers.4.mlp.experts.71.down_proj", "model.language_model.layers.4.mlp.experts.72.down_proj", "model.language_model.layers.4.mlp.experts.73.down_proj", "model.language_model.layers.4.mlp.experts.74.down_proj", "model.language_model.layers.4.mlp.experts.75.down_proj", "model.language_model.layers.4.mlp.experts.76.down_proj", "model.language_model.layers.4.mlp.experts.77.down_proj", "model.language_model.layers.4.mlp.experts.78.down_proj", "model.language_model.layers.4.mlp.experts.79.down_proj", "model.language_model.layers.4.mlp.experts.80.down_proj", "model.language_model.layers.4.mlp.experts.81.down_proj", "model.language_model.layers.4.mlp.experts.82.down_proj", "model.language_model.layers.4.mlp.experts.83.down_proj", "model.language_model.layers.4.mlp.experts.84.down_proj", "model.language_model.layers.4.mlp.experts.85.down_proj", "model.language_model.layers.4.mlp.experts.86.down_proj", "model.language_model.layers.4.mlp.experts.87.down_proj", "model.language_model.layers.4.mlp.experts.88.down_proj", "model.language_model.layers.4.mlp.experts.89.down_proj", "model.language_model.layers.4.mlp.experts.90.down_proj", "model.language_model.layers.4.mlp.experts.91.down_proj", "model.language_model.layers.4.mlp.experts.92.down_proj", "model.language_model.layers.4.mlp.experts.93.down_proj", "model.language_model.layers.4.mlp.experts.94.down_proj", "model.language_model.layers.4.mlp.experts.95.down_proj", "model.language_model.layers.4.mlp.experts.96.down_proj", "model.language_model.layers.4.mlp.experts.97.down_proj", "model.language_model.layers.4.mlp.experts.98.down_proj", "model.language_model.layers.4.mlp.experts.99.down_proj", "model.language_model.layers.4.mlp.experts.100.down_proj", "model.language_model.layers.4.mlp.experts.101.down_proj", "model.language_model.layers.4.mlp.experts.102.down_proj", "model.language_model.layers.4.mlp.experts.103.down_proj", "model.language_model.layers.4.mlp.experts.104.down_proj", "model.language_model.layers.4.mlp.experts.105.down_proj", "model.language_model.layers.4.mlp.experts.106.down_proj", "model.language_model.layers.4.mlp.experts.107.down_proj", "model.language_model.layers.4.mlp.experts.108.down_proj", "model.language_model.layers.4.mlp.experts.109.down_proj", "model.language_model.layers.4.mlp.experts.110.down_proj", "model.language_model.layers.4.mlp.experts.111.down_proj", "model.language_model.layers.4.mlp.experts.112.down_proj", "model.language_model.layers.4.mlp.experts.113.down_proj", "model.language_model.layers.4.mlp.experts.114.down_proj", "model.language_model.layers.4.mlp.experts.115.down_proj", "model.language_model.layers.4.mlp.experts.116.down_proj", "model.language_model.layers.4.mlp.experts.117.down_proj", "model.language_model.layers.4.mlp.experts.118.down_proj", "model.language_model.layers.4.mlp.experts.119.down_proj", "model.language_model.layers.4.mlp.experts.120.down_proj", "model.language_model.layers.4.mlp.experts.121.down_proj", "model.language_model.layers.4.mlp.experts.122.down_proj", "model.language_model.layers.4.mlp.experts.123.down_proj", "model.language_model.layers.4.mlp.experts.124.down_proj", "model.language_model.layers.4.mlp.experts.125.down_proj", "model.language_model.layers.4.mlp.experts.126.down_proj", "model.language_model.layers.4.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.002238314412534234, "dbits": 603979776 } ] }, { "idx": 10, "layers": [ "model.language_model.layers.5.self_attn.q_proj", "model.language_model.layers.5.self_attn.k_proj", "model.language_model.layers.5.self_attn.v_proj", "model.language_model.layers.5.self_attn.o_proj" ], "candidates": [ { "dkld": -0.004164687730371958, "dbits": 18874368 } ] }, { "idx": 11, "layers": [ "model.language_model.layers.5.mlp.experts.0.gate_proj", "model.language_model.layers.5.mlp.experts.1.gate_proj", "model.language_model.layers.5.mlp.experts.2.gate_proj", "model.language_model.layers.5.mlp.experts.3.gate_proj", "model.language_model.layers.5.mlp.experts.4.gate_proj", "model.language_model.layers.5.mlp.experts.5.gate_proj", "model.language_model.layers.5.mlp.experts.6.gate_proj", "model.language_model.layers.5.mlp.experts.7.gate_proj", "model.language_model.layers.5.mlp.experts.8.gate_proj", "model.language_model.layers.5.mlp.experts.9.gate_proj", "model.language_model.layers.5.mlp.experts.10.gate_proj", "model.language_model.layers.5.mlp.experts.11.gate_proj", "model.language_model.layers.5.mlp.experts.12.gate_proj", "model.language_model.layers.5.mlp.experts.13.gate_proj", "model.language_model.layers.5.mlp.experts.14.gate_proj", "model.language_model.layers.5.mlp.experts.15.gate_proj", "model.language_model.layers.5.mlp.experts.16.gate_proj", "model.language_model.layers.5.mlp.experts.17.gate_proj", "model.language_model.layers.5.mlp.experts.18.gate_proj", "model.language_model.layers.5.mlp.experts.19.gate_proj", "model.language_model.layers.5.mlp.experts.20.gate_proj", "model.language_model.layers.5.mlp.experts.21.gate_proj", "model.language_model.layers.5.mlp.experts.22.gate_proj", "model.language_model.layers.5.mlp.experts.23.gate_proj", "model.language_model.layers.5.mlp.experts.24.gate_proj", "model.language_model.layers.5.mlp.experts.25.gate_proj", "model.language_model.layers.5.mlp.experts.26.gate_proj", "model.language_model.layers.5.mlp.experts.27.gate_proj", "model.language_model.layers.5.mlp.experts.28.gate_proj", "model.language_model.layers.5.mlp.experts.29.gate_proj", "model.language_model.layers.5.mlp.experts.30.gate_proj", "model.language_model.layers.5.mlp.experts.31.gate_proj", "model.language_model.layers.5.mlp.experts.32.gate_proj", "model.language_model.layers.5.mlp.experts.33.gate_proj", "model.language_model.layers.5.mlp.experts.34.gate_proj", "model.language_model.layers.5.mlp.experts.35.gate_proj", "model.language_model.layers.5.mlp.experts.36.gate_proj", "model.language_model.layers.5.mlp.experts.37.gate_proj", "model.language_model.layers.5.mlp.experts.38.gate_proj", "model.language_model.layers.5.mlp.experts.39.gate_proj", "model.language_model.layers.5.mlp.experts.40.gate_proj", "model.language_model.layers.5.mlp.experts.41.gate_proj", "model.language_model.layers.5.mlp.experts.42.gate_proj", "model.language_model.layers.5.mlp.experts.43.gate_proj", "model.language_model.layers.5.mlp.experts.44.gate_proj", "model.language_model.layers.5.mlp.experts.45.gate_proj", "model.language_model.layers.5.mlp.experts.46.gate_proj", "model.language_model.layers.5.mlp.experts.47.gate_proj", "model.language_model.layers.5.mlp.experts.48.gate_proj", "model.language_model.layers.5.mlp.experts.49.gate_proj", "model.language_model.layers.5.mlp.experts.50.gate_proj", "model.language_model.layers.5.mlp.experts.51.gate_proj", "model.language_model.layers.5.mlp.experts.52.gate_proj", "model.language_model.layers.5.mlp.experts.53.gate_proj", "model.language_model.layers.5.mlp.experts.54.gate_proj", "model.language_model.layers.5.mlp.experts.55.gate_proj", "model.language_model.layers.5.mlp.experts.56.gate_proj", "model.language_model.layers.5.mlp.experts.57.gate_proj", "model.language_model.layers.5.mlp.experts.58.gate_proj", "model.language_model.layers.5.mlp.experts.59.gate_proj", "model.language_model.layers.5.mlp.experts.60.gate_proj", "model.language_model.layers.5.mlp.experts.61.gate_proj", "model.language_model.layers.5.mlp.experts.62.gate_proj", "model.language_model.layers.5.mlp.experts.63.gate_proj", "model.language_model.layers.5.mlp.experts.64.gate_proj", "model.language_model.layers.5.mlp.experts.65.gate_proj", "model.language_model.layers.5.mlp.experts.66.gate_proj", "model.language_model.layers.5.mlp.experts.67.gate_proj", "model.language_model.layers.5.mlp.experts.68.gate_proj", "model.language_model.layers.5.mlp.experts.69.gate_proj", "model.language_model.layers.5.mlp.experts.70.gate_proj", "model.language_model.layers.5.mlp.experts.71.gate_proj", "model.language_model.layers.5.mlp.experts.72.gate_proj", "model.language_model.layers.5.mlp.experts.73.gate_proj", "model.language_model.layers.5.mlp.experts.74.gate_proj", "model.language_model.layers.5.mlp.experts.75.gate_proj", "model.language_model.layers.5.mlp.experts.76.gate_proj", "model.language_model.layers.5.mlp.experts.77.gate_proj", "model.language_model.layers.5.mlp.experts.78.gate_proj", "model.language_model.layers.5.mlp.experts.79.gate_proj", "model.language_model.layers.5.mlp.experts.80.gate_proj", "model.language_model.layers.5.mlp.experts.81.gate_proj", "model.language_model.layers.5.mlp.experts.82.gate_proj", "model.language_model.layers.5.mlp.experts.83.gate_proj", "model.language_model.layers.5.mlp.experts.84.gate_proj", "model.language_model.layers.5.mlp.experts.85.gate_proj", "model.language_model.layers.5.mlp.experts.86.gate_proj", "model.language_model.layers.5.mlp.experts.87.gate_proj", "model.language_model.layers.5.mlp.experts.88.gate_proj", "model.language_model.layers.5.mlp.experts.89.gate_proj", "model.language_model.layers.5.mlp.experts.90.gate_proj", "model.language_model.layers.5.mlp.experts.91.gate_proj", "model.language_model.layers.5.mlp.experts.92.gate_proj", "model.language_model.layers.5.mlp.experts.93.gate_proj", "model.language_model.layers.5.mlp.experts.94.gate_proj", "model.language_model.layers.5.mlp.experts.95.gate_proj", "model.language_model.layers.5.mlp.experts.96.gate_proj", "model.language_model.layers.5.mlp.experts.97.gate_proj", "model.language_model.layers.5.mlp.experts.98.gate_proj", "model.language_model.layers.5.mlp.experts.99.gate_proj", "model.language_model.layers.5.mlp.experts.100.gate_proj", "model.language_model.layers.5.mlp.experts.101.gate_proj", "model.language_model.layers.5.mlp.experts.102.gate_proj", "model.language_model.layers.5.mlp.experts.103.gate_proj", "model.language_model.layers.5.mlp.experts.104.gate_proj", "model.language_model.layers.5.mlp.experts.105.gate_proj", "model.language_model.layers.5.mlp.experts.106.gate_proj", "model.language_model.layers.5.mlp.experts.107.gate_proj", "model.language_model.layers.5.mlp.experts.108.gate_proj", "model.language_model.layers.5.mlp.experts.109.gate_proj", "model.language_model.layers.5.mlp.experts.110.gate_proj", "model.language_model.layers.5.mlp.experts.111.gate_proj", "model.language_model.layers.5.mlp.experts.112.gate_proj", "model.language_model.layers.5.mlp.experts.113.gate_proj", "model.language_model.layers.5.mlp.experts.114.gate_proj", "model.language_model.layers.5.mlp.experts.115.gate_proj", "model.language_model.layers.5.mlp.experts.116.gate_proj", "model.language_model.layers.5.mlp.experts.117.gate_proj", "model.language_model.layers.5.mlp.experts.118.gate_proj", "model.language_model.layers.5.mlp.experts.119.gate_proj", "model.language_model.layers.5.mlp.experts.120.gate_proj", "model.language_model.layers.5.mlp.experts.121.gate_proj", "model.language_model.layers.5.mlp.experts.122.gate_proj", "model.language_model.layers.5.mlp.experts.123.gate_proj", "model.language_model.layers.5.mlp.experts.124.gate_proj", "model.language_model.layers.5.mlp.experts.125.gate_proj", "model.language_model.layers.5.mlp.experts.126.gate_proj", "model.language_model.layers.5.mlp.experts.127.gate_proj", "model.language_model.layers.5.mlp.experts.0.up_proj", "model.language_model.layers.5.mlp.experts.1.up_proj", "model.language_model.layers.5.mlp.experts.2.up_proj", "model.language_model.layers.5.mlp.experts.3.up_proj", "model.language_model.layers.5.mlp.experts.4.up_proj", "model.language_model.layers.5.mlp.experts.5.up_proj", "model.language_model.layers.5.mlp.experts.6.up_proj", "model.language_model.layers.5.mlp.experts.7.up_proj", "model.language_model.layers.5.mlp.experts.8.up_proj", "model.language_model.layers.5.mlp.experts.9.up_proj", "model.language_model.layers.5.mlp.experts.10.up_proj", "model.language_model.layers.5.mlp.experts.11.up_proj", "model.language_model.layers.5.mlp.experts.12.up_proj", "model.language_model.layers.5.mlp.experts.13.up_proj", "model.language_model.layers.5.mlp.experts.14.up_proj", "model.language_model.layers.5.mlp.experts.15.up_proj", "model.language_model.layers.5.mlp.experts.16.up_proj", "model.language_model.layers.5.mlp.experts.17.up_proj", "model.language_model.layers.5.mlp.experts.18.up_proj", "model.language_model.layers.5.mlp.experts.19.up_proj", "model.language_model.layers.5.mlp.experts.20.up_proj", "model.language_model.layers.5.mlp.experts.21.up_proj", "model.language_model.layers.5.mlp.experts.22.up_proj", "model.language_model.layers.5.mlp.experts.23.up_proj", "model.language_model.layers.5.mlp.experts.24.up_proj", "model.language_model.layers.5.mlp.experts.25.up_proj", "model.language_model.layers.5.mlp.experts.26.up_proj", "model.language_model.layers.5.mlp.experts.27.up_proj", "model.language_model.layers.5.mlp.experts.28.up_proj", "model.language_model.layers.5.mlp.experts.29.up_proj", "model.language_model.layers.5.mlp.experts.30.up_proj", "model.language_model.layers.5.mlp.experts.31.up_proj", "model.language_model.layers.5.mlp.experts.32.up_proj", "model.language_model.layers.5.mlp.experts.33.up_proj", "model.language_model.layers.5.mlp.experts.34.up_proj", "model.language_model.layers.5.mlp.experts.35.up_proj", "model.language_model.layers.5.mlp.experts.36.up_proj", "model.language_model.layers.5.mlp.experts.37.up_proj", "model.language_model.layers.5.mlp.experts.38.up_proj", "model.language_model.layers.5.mlp.experts.39.up_proj", "model.language_model.layers.5.mlp.experts.40.up_proj", "model.language_model.layers.5.mlp.experts.41.up_proj", "model.language_model.layers.5.mlp.experts.42.up_proj", "model.language_model.layers.5.mlp.experts.43.up_proj", "model.language_model.layers.5.mlp.experts.44.up_proj", "model.language_model.layers.5.mlp.experts.45.up_proj", "model.language_model.layers.5.mlp.experts.46.up_proj", "model.language_model.layers.5.mlp.experts.47.up_proj", "model.language_model.layers.5.mlp.experts.48.up_proj", "model.language_model.layers.5.mlp.experts.49.up_proj", "model.language_model.layers.5.mlp.experts.50.up_proj", "model.language_model.layers.5.mlp.experts.51.up_proj", "model.language_model.layers.5.mlp.experts.52.up_proj", "model.language_model.layers.5.mlp.experts.53.up_proj", "model.language_model.layers.5.mlp.experts.54.up_proj", "model.language_model.layers.5.mlp.experts.55.up_proj", "model.language_model.layers.5.mlp.experts.56.up_proj", "model.language_model.layers.5.mlp.experts.57.up_proj", "model.language_model.layers.5.mlp.experts.58.up_proj", "model.language_model.layers.5.mlp.experts.59.up_proj", "model.language_model.layers.5.mlp.experts.60.up_proj", "model.language_model.layers.5.mlp.experts.61.up_proj", "model.language_model.layers.5.mlp.experts.62.up_proj", "model.language_model.layers.5.mlp.experts.63.up_proj", "model.language_model.layers.5.mlp.experts.64.up_proj", "model.language_model.layers.5.mlp.experts.65.up_proj", "model.language_model.layers.5.mlp.experts.66.up_proj", "model.language_model.layers.5.mlp.experts.67.up_proj", "model.language_model.layers.5.mlp.experts.68.up_proj", "model.language_model.layers.5.mlp.experts.69.up_proj", "model.language_model.layers.5.mlp.experts.70.up_proj", "model.language_model.layers.5.mlp.experts.71.up_proj", "model.language_model.layers.5.mlp.experts.72.up_proj", "model.language_model.layers.5.mlp.experts.73.up_proj", "model.language_model.layers.5.mlp.experts.74.up_proj", "model.language_model.layers.5.mlp.experts.75.up_proj", "model.language_model.layers.5.mlp.experts.76.up_proj", "model.language_model.layers.5.mlp.experts.77.up_proj", "model.language_model.layers.5.mlp.experts.78.up_proj", "model.language_model.layers.5.mlp.experts.79.up_proj", "model.language_model.layers.5.mlp.experts.80.up_proj", "model.language_model.layers.5.mlp.experts.81.up_proj", "model.language_model.layers.5.mlp.experts.82.up_proj", "model.language_model.layers.5.mlp.experts.83.up_proj", "model.language_model.layers.5.mlp.experts.84.up_proj", "model.language_model.layers.5.mlp.experts.85.up_proj", "model.language_model.layers.5.mlp.experts.86.up_proj", "model.language_model.layers.5.mlp.experts.87.up_proj", "model.language_model.layers.5.mlp.experts.88.up_proj", "model.language_model.layers.5.mlp.experts.89.up_proj", "model.language_model.layers.5.mlp.experts.90.up_proj", "model.language_model.layers.5.mlp.experts.91.up_proj", "model.language_model.layers.5.mlp.experts.92.up_proj", "model.language_model.layers.5.mlp.experts.93.up_proj", "model.language_model.layers.5.mlp.experts.94.up_proj", "model.language_model.layers.5.mlp.experts.95.up_proj", "model.language_model.layers.5.mlp.experts.96.up_proj", "model.language_model.layers.5.mlp.experts.97.up_proj", "model.language_model.layers.5.mlp.experts.98.up_proj", "model.language_model.layers.5.mlp.experts.99.up_proj", "model.language_model.layers.5.mlp.experts.100.up_proj", "model.language_model.layers.5.mlp.experts.101.up_proj", "model.language_model.layers.5.mlp.experts.102.up_proj", "model.language_model.layers.5.mlp.experts.103.up_proj", "model.language_model.layers.5.mlp.experts.104.up_proj", "model.language_model.layers.5.mlp.experts.105.up_proj", "model.language_model.layers.5.mlp.experts.106.up_proj", "model.language_model.layers.5.mlp.experts.107.up_proj", "model.language_model.layers.5.mlp.experts.108.up_proj", "model.language_model.layers.5.mlp.experts.109.up_proj", "model.language_model.layers.5.mlp.experts.110.up_proj", "model.language_model.layers.5.mlp.experts.111.up_proj", "model.language_model.layers.5.mlp.experts.112.up_proj", "model.language_model.layers.5.mlp.experts.113.up_proj", "model.language_model.layers.5.mlp.experts.114.up_proj", "model.language_model.layers.5.mlp.experts.115.up_proj", "model.language_model.layers.5.mlp.experts.116.up_proj", "model.language_model.layers.5.mlp.experts.117.up_proj", "model.language_model.layers.5.mlp.experts.118.up_proj", "model.language_model.layers.5.mlp.experts.119.up_proj", "model.language_model.layers.5.mlp.experts.120.up_proj", "model.language_model.layers.5.mlp.experts.121.up_proj", "model.language_model.layers.5.mlp.experts.122.up_proj", "model.language_model.layers.5.mlp.experts.123.up_proj", "model.language_model.layers.5.mlp.experts.124.up_proj", "model.language_model.layers.5.mlp.experts.125.up_proj", "model.language_model.layers.5.mlp.experts.126.up_proj", "model.language_model.layers.5.mlp.experts.127.up_proj", "model.language_model.layers.5.mlp.experts.0.down_proj", "model.language_model.layers.5.mlp.experts.1.down_proj", "model.language_model.layers.5.mlp.experts.2.down_proj", "model.language_model.layers.5.mlp.experts.3.down_proj", "model.language_model.layers.5.mlp.experts.4.down_proj", "model.language_model.layers.5.mlp.experts.5.down_proj", "model.language_model.layers.5.mlp.experts.6.down_proj", "model.language_model.layers.5.mlp.experts.7.down_proj", "model.language_model.layers.5.mlp.experts.8.down_proj", "model.language_model.layers.5.mlp.experts.9.down_proj", "model.language_model.layers.5.mlp.experts.10.down_proj", "model.language_model.layers.5.mlp.experts.11.down_proj", "model.language_model.layers.5.mlp.experts.12.down_proj", "model.language_model.layers.5.mlp.experts.13.down_proj", "model.language_model.layers.5.mlp.experts.14.down_proj", "model.language_model.layers.5.mlp.experts.15.down_proj", "model.language_model.layers.5.mlp.experts.16.down_proj", "model.language_model.layers.5.mlp.experts.17.down_proj", "model.language_model.layers.5.mlp.experts.18.down_proj", "model.language_model.layers.5.mlp.experts.19.down_proj", "model.language_model.layers.5.mlp.experts.20.down_proj", "model.language_model.layers.5.mlp.experts.21.down_proj", "model.language_model.layers.5.mlp.experts.22.down_proj", "model.language_model.layers.5.mlp.experts.23.down_proj", "model.language_model.layers.5.mlp.experts.24.down_proj", "model.language_model.layers.5.mlp.experts.25.down_proj", "model.language_model.layers.5.mlp.experts.26.down_proj", "model.language_model.layers.5.mlp.experts.27.down_proj", "model.language_model.layers.5.mlp.experts.28.down_proj", "model.language_model.layers.5.mlp.experts.29.down_proj", "model.language_model.layers.5.mlp.experts.30.down_proj", "model.language_model.layers.5.mlp.experts.31.down_proj", "model.language_model.layers.5.mlp.experts.32.down_proj", "model.language_model.layers.5.mlp.experts.33.down_proj", "model.language_model.layers.5.mlp.experts.34.down_proj", "model.language_model.layers.5.mlp.experts.35.down_proj", "model.language_model.layers.5.mlp.experts.36.down_proj", "model.language_model.layers.5.mlp.experts.37.down_proj", "model.language_model.layers.5.mlp.experts.38.down_proj", "model.language_model.layers.5.mlp.experts.39.down_proj", "model.language_model.layers.5.mlp.experts.40.down_proj", "model.language_model.layers.5.mlp.experts.41.down_proj", "model.language_model.layers.5.mlp.experts.42.down_proj", "model.language_model.layers.5.mlp.experts.43.down_proj", "model.language_model.layers.5.mlp.experts.44.down_proj", "model.language_model.layers.5.mlp.experts.45.down_proj", "model.language_model.layers.5.mlp.experts.46.down_proj", "model.language_model.layers.5.mlp.experts.47.down_proj", "model.language_model.layers.5.mlp.experts.48.down_proj", "model.language_model.layers.5.mlp.experts.49.down_proj", "model.language_model.layers.5.mlp.experts.50.down_proj", "model.language_model.layers.5.mlp.experts.51.down_proj", "model.language_model.layers.5.mlp.experts.52.down_proj", "model.language_model.layers.5.mlp.experts.53.down_proj", "model.language_model.layers.5.mlp.experts.54.down_proj", "model.language_model.layers.5.mlp.experts.55.down_proj", "model.language_model.layers.5.mlp.experts.56.down_proj", "model.language_model.layers.5.mlp.experts.57.down_proj", "model.language_model.layers.5.mlp.experts.58.down_proj", "model.language_model.layers.5.mlp.experts.59.down_proj", "model.language_model.layers.5.mlp.experts.60.down_proj", "model.language_model.layers.5.mlp.experts.61.down_proj", "model.language_model.layers.5.mlp.experts.62.down_proj", "model.language_model.layers.5.mlp.experts.63.down_proj", "model.language_model.layers.5.mlp.experts.64.down_proj", "model.language_model.layers.5.mlp.experts.65.down_proj", "model.language_model.layers.5.mlp.experts.66.down_proj", "model.language_model.layers.5.mlp.experts.67.down_proj", "model.language_model.layers.5.mlp.experts.68.down_proj", "model.language_model.layers.5.mlp.experts.69.down_proj", "model.language_model.layers.5.mlp.experts.70.down_proj", "model.language_model.layers.5.mlp.experts.71.down_proj", "model.language_model.layers.5.mlp.experts.72.down_proj", "model.language_model.layers.5.mlp.experts.73.down_proj", "model.language_model.layers.5.mlp.experts.74.down_proj", "model.language_model.layers.5.mlp.experts.75.down_proj", "model.language_model.layers.5.mlp.experts.76.down_proj", "model.language_model.layers.5.mlp.experts.77.down_proj", "model.language_model.layers.5.mlp.experts.78.down_proj", "model.language_model.layers.5.mlp.experts.79.down_proj", "model.language_model.layers.5.mlp.experts.80.down_proj", "model.language_model.layers.5.mlp.experts.81.down_proj", "model.language_model.layers.5.mlp.experts.82.down_proj", "model.language_model.layers.5.mlp.experts.83.down_proj", "model.language_model.layers.5.mlp.experts.84.down_proj", "model.language_model.layers.5.mlp.experts.85.down_proj", "model.language_model.layers.5.mlp.experts.86.down_proj", "model.language_model.layers.5.mlp.experts.87.down_proj", "model.language_model.layers.5.mlp.experts.88.down_proj", "model.language_model.layers.5.mlp.experts.89.down_proj", "model.language_model.layers.5.mlp.experts.90.down_proj", "model.language_model.layers.5.mlp.experts.91.down_proj", "model.language_model.layers.5.mlp.experts.92.down_proj", "model.language_model.layers.5.mlp.experts.93.down_proj", "model.language_model.layers.5.mlp.experts.94.down_proj", "model.language_model.layers.5.mlp.experts.95.down_proj", "model.language_model.layers.5.mlp.experts.96.down_proj", "model.language_model.layers.5.mlp.experts.97.down_proj", "model.language_model.layers.5.mlp.experts.98.down_proj", "model.language_model.layers.5.mlp.experts.99.down_proj", "model.language_model.layers.5.mlp.experts.100.down_proj", "model.language_model.layers.5.mlp.experts.101.down_proj", "model.language_model.layers.5.mlp.experts.102.down_proj", "model.language_model.layers.5.mlp.experts.103.down_proj", "model.language_model.layers.5.mlp.experts.104.down_proj", "model.language_model.layers.5.mlp.experts.105.down_proj", "model.language_model.layers.5.mlp.experts.106.down_proj", "model.language_model.layers.5.mlp.experts.107.down_proj", "model.language_model.layers.5.mlp.experts.108.down_proj", "model.language_model.layers.5.mlp.experts.109.down_proj", "model.language_model.layers.5.mlp.experts.110.down_proj", "model.language_model.layers.5.mlp.experts.111.down_proj", "model.language_model.layers.5.mlp.experts.112.down_proj", "model.language_model.layers.5.mlp.experts.113.down_proj", "model.language_model.layers.5.mlp.experts.114.down_proj", "model.language_model.layers.5.mlp.experts.115.down_proj", "model.language_model.layers.5.mlp.experts.116.down_proj", "model.language_model.layers.5.mlp.experts.117.down_proj", "model.language_model.layers.5.mlp.experts.118.down_proj", "model.language_model.layers.5.mlp.experts.119.down_proj", "model.language_model.layers.5.mlp.experts.120.down_proj", "model.language_model.layers.5.mlp.experts.121.down_proj", "model.language_model.layers.5.mlp.experts.122.down_proj", "model.language_model.layers.5.mlp.experts.123.down_proj", "model.language_model.layers.5.mlp.experts.124.down_proj", "model.language_model.layers.5.mlp.experts.125.down_proj", "model.language_model.layers.5.mlp.experts.126.down_proj", "model.language_model.layers.5.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.011422762088477617, "dbits": 603979776 } ] }, { "idx": 12, "layers": [ "model.language_model.layers.6.self_attn.q_proj", "model.language_model.layers.6.self_attn.k_proj", "model.language_model.layers.6.self_attn.v_proj", "model.language_model.layers.6.self_attn.o_proj" ], "candidates": [ { "dkld": 0.0029464741237461567, "dbits": 18874368 } ] }, { "idx": 13, "layers": [ "model.language_model.layers.6.mlp.experts.0.gate_proj", "model.language_model.layers.6.mlp.experts.1.gate_proj", "model.language_model.layers.6.mlp.experts.2.gate_proj", "model.language_model.layers.6.mlp.experts.3.gate_proj", "model.language_model.layers.6.mlp.experts.4.gate_proj", "model.language_model.layers.6.mlp.experts.5.gate_proj", "model.language_model.layers.6.mlp.experts.6.gate_proj", "model.language_model.layers.6.mlp.experts.7.gate_proj", "model.language_model.layers.6.mlp.experts.8.gate_proj", "model.language_model.layers.6.mlp.experts.9.gate_proj", "model.language_model.layers.6.mlp.experts.10.gate_proj", "model.language_model.layers.6.mlp.experts.11.gate_proj", "model.language_model.layers.6.mlp.experts.12.gate_proj", "model.language_model.layers.6.mlp.experts.13.gate_proj", "model.language_model.layers.6.mlp.experts.14.gate_proj", "model.language_model.layers.6.mlp.experts.15.gate_proj", "model.language_model.layers.6.mlp.experts.16.gate_proj", "model.language_model.layers.6.mlp.experts.17.gate_proj", "model.language_model.layers.6.mlp.experts.18.gate_proj", "model.language_model.layers.6.mlp.experts.19.gate_proj", "model.language_model.layers.6.mlp.experts.20.gate_proj", "model.language_model.layers.6.mlp.experts.21.gate_proj", "model.language_model.layers.6.mlp.experts.22.gate_proj", "model.language_model.layers.6.mlp.experts.23.gate_proj", "model.language_model.layers.6.mlp.experts.24.gate_proj", "model.language_model.layers.6.mlp.experts.25.gate_proj", "model.language_model.layers.6.mlp.experts.26.gate_proj", "model.language_model.layers.6.mlp.experts.27.gate_proj", "model.language_model.layers.6.mlp.experts.28.gate_proj", "model.language_model.layers.6.mlp.experts.29.gate_proj", "model.language_model.layers.6.mlp.experts.30.gate_proj", "model.language_model.layers.6.mlp.experts.31.gate_proj", "model.language_model.layers.6.mlp.experts.32.gate_proj", "model.language_model.layers.6.mlp.experts.33.gate_proj", "model.language_model.layers.6.mlp.experts.34.gate_proj", "model.language_model.layers.6.mlp.experts.35.gate_proj", "model.language_model.layers.6.mlp.experts.36.gate_proj", "model.language_model.layers.6.mlp.experts.37.gate_proj", "model.language_model.layers.6.mlp.experts.38.gate_proj", "model.language_model.layers.6.mlp.experts.39.gate_proj", "model.language_model.layers.6.mlp.experts.40.gate_proj", "model.language_model.layers.6.mlp.experts.41.gate_proj", "model.language_model.layers.6.mlp.experts.42.gate_proj", "model.language_model.layers.6.mlp.experts.43.gate_proj", "model.language_model.layers.6.mlp.experts.44.gate_proj", "model.language_model.layers.6.mlp.experts.45.gate_proj", "model.language_model.layers.6.mlp.experts.46.gate_proj", "model.language_model.layers.6.mlp.experts.47.gate_proj", "model.language_model.layers.6.mlp.experts.48.gate_proj", "model.language_model.layers.6.mlp.experts.49.gate_proj", "model.language_model.layers.6.mlp.experts.50.gate_proj", "model.language_model.layers.6.mlp.experts.51.gate_proj", "model.language_model.layers.6.mlp.experts.52.gate_proj", "model.language_model.layers.6.mlp.experts.53.gate_proj", "model.language_model.layers.6.mlp.experts.54.gate_proj", "model.language_model.layers.6.mlp.experts.55.gate_proj", "model.language_model.layers.6.mlp.experts.56.gate_proj", "model.language_model.layers.6.mlp.experts.57.gate_proj", "model.language_model.layers.6.mlp.experts.58.gate_proj", "model.language_model.layers.6.mlp.experts.59.gate_proj", "model.language_model.layers.6.mlp.experts.60.gate_proj", "model.language_model.layers.6.mlp.experts.61.gate_proj", "model.language_model.layers.6.mlp.experts.62.gate_proj", "model.language_model.layers.6.mlp.experts.63.gate_proj", "model.language_model.layers.6.mlp.experts.64.gate_proj", "model.language_model.layers.6.mlp.experts.65.gate_proj", "model.language_model.layers.6.mlp.experts.66.gate_proj", "model.language_model.layers.6.mlp.experts.67.gate_proj", "model.language_model.layers.6.mlp.experts.68.gate_proj", "model.language_model.layers.6.mlp.experts.69.gate_proj", "model.language_model.layers.6.mlp.experts.70.gate_proj", "model.language_model.layers.6.mlp.experts.71.gate_proj", "model.language_model.layers.6.mlp.experts.72.gate_proj", "model.language_model.layers.6.mlp.experts.73.gate_proj", "model.language_model.layers.6.mlp.experts.74.gate_proj", "model.language_model.layers.6.mlp.experts.75.gate_proj", "model.language_model.layers.6.mlp.experts.76.gate_proj", "model.language_model.layers.6.mlp.experts.77.gate_proj", "model.language_model.layers.6.mlp.experts.78.gate_proj", "model.language_model.layers.6.mlp.experts.79.gate_proj", "model.language_model.layers.6.mlp.experts.80.gate_proj", "model.language_model.layers.6.mlp.experts.81.gate_proj", "model.language_model.layers.6.mlp.experts.82.gate_proj", "model.language_model.layers.6.mlp.experts.83.gate_proj", "model.language_model.layers.6.mlp.experts.84.gate_proj", "model.language_model.layers.6.mlp.experts.85.gate_proj", "model.language_model.layers.6.mlp.experts.86.gate_proj", "model.language_model.layers.6.mlp.experts.87.gate_proj", "model.language_model.layers.6.mlp.experts.88.gate_proj", "model.language_model.layers.6.mlp.experts.89.gate_proj", "model.language_model.layers.6.mlp.experts.90.gate_proj", "model.language_model.layers.6.mlp.experts.91.gate_proj", "model.language_model.layers.6.mlp.experts.92.gate_proj", "model.language_model.layers.6.mlp.experts.93.gate_proj", "model.language_model.layers.6.mlp.experts.94.gate_proj", "model.language_model.layers.6.mlp.experts.95.gate_proj", "model.language_model.layers.6.mlp.experts.96.gate_proj", "model.language_model.layers.6.mlp.experts.97.gate_proj", "model.language_model.layers.6.mlp.experts.98.gate_proj", "model.language_model.layers.6.mlp.experts.99.gate_proj", "model.language_model.layers.6.mlp.experts.100.gate_proj", "model.language_model.layers.6.mlp.experts.101.gate_proj", "model.language_model.layers.6.mlp.experts.102.gate_proj", "model.language_model.layers.6.mlp.experts.103.gate_proj", "model.language_model.layers.6.mlp.experts.104.gate_proj", "model.language_model.layers.6.mlp.experts.105.gate_proj", "model.language_model.layers.6.mlp.experts.106.gate_proj", "model.language_model.layers.6.mlp.experts.107.gate_proj", "model.language_model.layers.6.mlp.experts.108.gate_proj", "model.language_model.layers.6.mlp.experts.109.gate_proj", "model.language_model.layers.6.mlp.experts.110.gate_proj", "model.language_model.layers.6.mlp.experts.111.gate_proj", "model.language_model.layers.6.mlp.experts.112.gate_proj", "model.language_model.layers.6.mlp.experts.113.gate_proj", "model.language_model.layers.6.mlp.experts.114.gate_proj", "model.language_model.layers.6.mlp.experts.115.gate_proj", "model.language_model.layers.6.mlp.experts.116.gate_proj", "model.language_model.layers.6.mlp.experts.117.gate_proj", "model.language_model.layers.6.mlp.experts.118.gate_proj", "model.language_model.layers.6.mlp.experts.119.gate_proj", "model.language_model.layers.6.mlp.experts.120.gate_proj", "model.language_model.layers.6.mlp.experts.121.gate_proj", "model.language_model.layers.6.mlp.experts.122.gate_proj", "model.language_model.layers.6.mlp.experts.123.gate_proj", "model.language_model.layers.6.mlp.experts.124.gate_proj", "model.language_model.layers.6.mlp.experts.125.gate_proj", "model.language_model.layers.6.mlp.experts.126.gate_proj", "model.language_model.layers.6.mlp.experts.127.gate_proj", "model.language_model.layers.6.mlp.experts.0.up_proj", "model.language_model.layers.6.mlp.experts.1.up_proj", "model.language_model.layers.6.mlp.experts.2.up_proj", "model.language_model.layers.6.mlp.experts.3.up_proj", "model.language_model.layers.6.mlp.experts.4.up_proj", "model.language_model.layers.6.mlp.experts.5.up_proj", "model.language_model.layers.6.mlp.experts.6.up_proj", "model.language_model.layers.6.mlp.experts.7.up_proj", "model.language_model.layers.6.mlp.experts.8.up_proj", "model.language_model.layers.6.mlp.experts.9.up_proj", "model.language_model.layers.6.mlp.experts.10.up_proj", "model.language_model.layers.6.mlp.experts.11.up_proj", "model.language_model.layers.6.mlp.experts.12.up_proj", "model.language_model.layers.6.mlp.experts.13.up_proj", "model.language_model.layers.6.mlp.experts.14.up_proj", "model.language_model.layers.6.mlp.experts.15.up_proj", "model.language_model.layers.6.mlp.experts.16.up_proj", "model.language_model.layers.6.mlp.experts.17.up_proj", "model.language_model.layers.6.mlp.experts.18.up_proj", "model.language_model.layers.6.mlp.experts.19.up_proj", "model.language_model.layers.6.mlp.experts.20.up_proj", "model.language_model.layers.6.mlp.experts.21.up_proj", "model.language_model.layers.6.mlp.experts.22.up_proj", "model.language_model.layers.6.mlp.experts.23.up_proj", "model.language_model.layers.6.mlp.experts.24.up_proj", "model.language_model.layers.6.mlp.experts.25.up_proj", "model.language_model.layers.6.mlp.experts.26.up_proj", "model.language_model.layers.6.mlp.experts.27.up_proj", "model.language_model.layers.6.mlp.experts.28.up_proj", "model.language_model.layers.6.mlp.experts.29.up_proj", "model.language_model.layers.6.mlp.experts.30.up_proj", "model.language_model.layers.6.mlp.experts.31.up_proj", "model.language_model.layers.6.mlp.experts.32.up_proj", "model.language_model.layers.6.mlp.experts.33.up_proj", "model.language_model.layers.6.mlp.experts.34.up_proj", "model.language_model.layers.6.mlp.experts.35.up_proj", "model.language_model.layers.6.mlp.experts.36.up_proj", "model.language_model.layers.6.mlp.experts.37.up_proj", "model.language_model.layers.6.mlp.experts.38.up_proj", "model.language_model.layers.6.mlp.experts.39.up_proj", "model.language_model.layers.6.mlp.experts.40.up_proj", "model.language_model.layers.6.mlp.experts.41.up_proj", "model.language_model.layers.6.mlp.experts.42.up_proj", "model.language_model.layers.6.mlp.experts.43.up_proj", "model.language_model.layers.6.mlp.experts.44.up_proj", "model.language_model.layers.6.mlp.experts.45.up_proj", "model.language_model.layers.6.mlp.experts.46.up_proj", "model.language_model.layers.6.mlp.experts.47.up_proj", "model.language_model.layers.6.mlp.experts.48.up_proj", "model.language_model.layers.6.mlp.experts.49.up_proj", "model.language_model.layers.6.mlp.experts.50.up_proj", "model.language_model.layers.6.mlp.experts.51.up_proj", "model.language_model.layers.6.mlp.experts.52.up_proj", "model.language_model.layers.6.mlp.experts.53.up_proj", "model.language_model.layers.6.mlp.experts.54.up_proj", "model.language_model.layers.6.mlp.experts.55.up_proj", "model.language_model.layers.6.mlp.experts.56.up_proj", "model.language_model.layers.6.mlp.experts.57.up_proj", "model.language_model.layers.6.mlp.experts.58.up_proj", "model.language_model.layers.6.mlp.experts.59.up_proj", "model.language_model.layers.6.mlp.experts.60.up_proj", "model.language_model.layers.6.mlp.experts.61.up_proj", "model.language_model.layers.6.mlp.experts.62.up_proj", "model.language_model.layers.6.mlp.experts.63.up_proj", "model.language_model.layers.6.mlp.experts.64.up_proj", "model.language_model.layers.6.mlp.experts.65.up_proj", "model.language_model.layers.6.mlp.experts.66.up_proj", "model.language_model.layers.6.mlp.experts.67.up_proj", "model.language_model.layers.6.mlp.experts.68.up_proj", "model.language_model.layers.6.mlp.experts.69.up_proj", "model.language_model.layers.6.mlp.experts.70.up_proj", "model.language_model.layers.6.mlp.experts.71.up_proj", "model.language_model.layers.6.mlp.experts.72.up_proj", "model.language_model.layers.6.mlp.experts.73.up_proj", "model.language_model.layers.6.mlp.experts.74.up_proj", "model.language_model.layers.6.mlp.experts.75.up_proj", "model.language_model.layers.6.mlp.experts.76.up_proj", "model.language_model.layers.6.mlp.experts.77.up_proj", "model.language_model.layers.6.mlp.experts.78.up_proj", "model.language_model.layers.6.mlp.experts.79.up_proj", "model.language_model.layers.6.mlp.experts.80.up_proj", "model.language_model.layers.6.mlp.experts.81.up_proj", "model.language_model.layers.6.mlp.experts.82.up_proj", "model.language_model.layers.6.mlp.experts.83.up_proj", "model.language_model.layers.6.mlp.experts.84.up_proj", "model.language_model.layers.6.mlp.experts.85.up_proj", "model.language_model.layers.6.mlp.experts.86.up_proj", "model.language_model.layers.6.mlp.experts.87.up_proj", "model.language_model.layers.6.mlp.experts.88.up_proj", "model.language_model.layers.6.mlp.experts.89.up_proj", "model.language_model.layers.6.mlp.experts.90.up_proj", "model.language_model.layers.6.mlp.experts.91.up_proj", "model.language_model.layers.6.mlp.experts.92.up_proj", "model.language_model.layers.6.mlp.experts.93.up_proj", "model.language_model.layers.6.mlp.experts.94.up_proj", "model.language_model.layers.6.mlp.experts.95.up_proj", "model.language_model.layers.6.mlp.experts.96.up_proj", "model.language_model.layers.6.mlp.experts.97.up_proj", "model.language_model.layers.6.mlp.experts.98.up_proj", "model.language_model.layers.6.mlp.experts.99.up_proj", "model.language_model.layers.6.mlp.experts.100.up_proj", "model.language_model.layers.6.mlp.experts.101.up_proj", "model.language_model.layers.6.mlp.experts.102.up_proj", "model.language_model.layers.6.mlp.experts.103.up_proj", "model.language_model.layers.6.mlp.experts.104.up_proj", "model.language_model.layers.6.mlp.experts.105.up_proj", "model.language_model.layers.6.mlp.experts.106.up_proj", "model.language_model.layers.6.mlp.experts.107.up_proj", "model.language_model.layers.6.mlp.experts.108.up_proj", "model.language_model.layers.6.mlp.experts.109.up_proj", "model.language_model.layers.6.mlp.experts.110.up_proj", "model.language_model.layers.6.mlp.experts.111.up_proj", "model.language_model.layers.6.mlp.experts.112.up_proj", "model.language_model.layers.6.mlp.experts.113.up_proj", "model.language_model.layers.6.mlp.experts.114.up_proj", "model.language_model.layers.6.mlp.experts.115.up_proj", "model.language_model.layers.6.mlp.experts.116.up_proj", "model.language_model.layers.6.mlp.experts.117.up_proj", "model.language_model.layers.6.mlp.experts.118.up_proj", "model.language_model.layers.6.mlp.experts.119.up_proj", "model.language_model.layers.6.mlp.experts.120.up_proj", "model.language_model.layers.6.mlp.experts.121.up_proj", "model.language_model.layers.6.mlp.experts.122.up_proj", "model.language_model.layers.6.mlp.experts.123.up_proj", "model.language_model.layers.6.mlp.experts.124.up_proj", "model.language_model.layers.6.mlp.experts.125.up_proj", "model.language_model.layers.6.mlp.experts.126.up_proj", "model.language_model.layers.6.mlp.experts.127.up_proj", "model.language_model.layers.6.mlp.experts.0.down_proj", "model.language_model.layers.6.mlp.experts.1.down_proj", "model.language_model.layers.6.mlp.experts.2.down_proj", "model.language_model.layers.6.mlp.experts.3.down_proj", "model.language_model.layers.6.mlp.experts.4.down_proj", "model.language_model.layers.6.mlp.experts.5.down_proj", "model.language_model.layers.6.mlp.experts.6.down_proj", "model.language_model.layers.6.mlp.experts.7.down_proj", "model.language_model.layers.6.mlp.experts.8.down_proj", "model.language_model.layers.6.mlp.experts.9.down_proj", "model.language_model.layers.6.mlp.experts.10.down_proj", "model.language_model.layers.6.mlp.experts.11.down_proj", "model.language_model.layers.6.mlp.experts.12.down_proj", "model.language_model.layers.6.mlp.experts.13.down_proj", "model.language_model.layers.6.mlp.experts.14.down_proj", "model.language_model.layers.6.mlp.experts.15.down_proj", "model.language_model.layers.6.mlp.experts.16.down_proj", "model.language_model.layers.6.mlp.experts.17.down_proj", "model.language_model.layers.6.mlp.experts.18.down_proj", "model.language_model.layers.6.mlp.experts.19.down_proj", "model.language_model.layers.6.mlp.experts.20.down_proj", "model.language_model.layers.6.mlp.experts.21.down_proj", "model.language_model.layers.6.mlp.experts.22.down_proj", "model.language_model.layers.6.mlp.experts.23.down_proj", "model.language_model.layers.6.mlp.experts.24.down_proj", "model.language_model.layers.6.mlp.experts.25.down_proj", "model.language_model.layers.6.mlp.experts.26.down_proj", "model.language_model.layers.6.mlp.experts.27.down_proj", "model.language_model.layers.6.mlp.experts.28.down_proj", "model.language_model.layers.6.mlp.experts.29.down_proj", "model.language_model.layers.6.mlp.experts.30.down_proj", "model.language_model.layers.6.mlp.experts.31.down_proj", "model.language_model.layers.6.mlp.experts.32.down_proj", "model.language_model.layers.6.mlp.experts.33.down_proj", "model.language_model.layers.6.mlp.experts.34.down_proj", "model.language_model.layers.6.mlp.experts.35.down_proj", "model.language_model.layers.6.mlp.experts.36.down_proj", "model.language_model.layers.6.mlp.experts.37.down_proj", "model.language_model.layers.6.mlp.experts.38.down_proj", "model.language_model.layers.6.mlp.experts.39.down_proj", "model.language_model.layers.6.mlp.experts.40.down_proj", "model.language_model.layers.6.mlp.experts.41.down_proj", "model.language_model.layers.6.mlp.experts.42.down_proj", "model.language_model.layers.6.mlp.experts.43.down_proj", "model.language_model.layers.6.mlp.experts.44.down_proj", "model.language_model.layers.6.mlp.experts.45.down_proj", "model.language_model.layers.6.mlp.experts.46.down_proj", "model.language_model.layers.6.mlp.experts.47.down_proj", "model.language_model.layers.6.mlp.experts.48.down_proj", "model.language_model.layers.6.mlp.experts.49.down_proj", "model.language_model.layers.6.mlp.experts.50.down_proj", "model.language_model.layers.6.mlp.experts.51.down_proj", "model.language_model.layers.6.mlp.experts.52.down_proj", "model.language_model.layers.6.mlp.experts.53.down_proj", "model.language_model.layers.6.mlp.experts.54.down_proj", "model.language_model.layers.6.mlp.experts.55.down_proj", "model.language_model.layers.6.mlp.experts.56.down_proj", "model.language_model.layers.6.mlp.experts.57.down_proj", "model.language_model.layers.6.mlp.experts.58.down_proj", "model.language_model.layers.6.mlp.experts.59.down_proj", "model.language_model.layers.6.mlp.experts.60.down_proj", "model.language_model.layers.6.mlp.experts.61.down_proj", "model.language_model.layers.6.mlp.experts.62.down_proj", "model.language_model.layers.6.mlp.experts.63.down_proj", "model.language_model.layers.6.mlp.experts.64.down_proj", "model.language_model.layers.6.mlp.experts.65.down_proj", "model.language_model.layers.6.mlp.experts.66.down_proj", "model.language_model.layers.6.mlp.experts.67.down_proj", "model.language_model.layers.6.mlp.experts.68.down_proj", "model.language_model.layers.6.mlp.experts.69.down_proj", "model.language_model.layers.6.mlp.experts.70.down_proj", "model.language_model.layers.6.mlp.experts.71.down_proj", "model.language_model.layers.6.mlp.experts.72.down_proj", "model.language_model.layers.6.mlp.experts.73.down_proj", "model.language_model.layers.6.mlp.experts.74.down_proj", "model.language_model.layers.6.mlp.experts.75.down_proj", "model.language_model.layers.6.mlp.experts.76.down_proj", "model.language_model.layers.6.mlp.experts.77.down_proj", "model.language_model.layers.6.mlp.experts.78.down_proj", "model.language_model.layers.6.mlp.experts.79.down_proj", "model.language_model.layers.6.mlp.experts.80.down_proj", "model.language_model.layers.6.mlp.experts.81.down_proj", "model.language_model.layers.6.mlp.experts.82.down_proj", "model.language_model.layers.6.mlp.experts.83.down_proj", "model.language_model.layers.6.mlp.experts.84.down_proj", "model.language_model.layers.6.mlp.experts.85.down_proj", "model.language_model.layers.6.mlp.experts.86.down_proj", "model.language_model.layers.6.mlp.experts.87.down_proj", "model.language_model.layers.6.mlp.experts.88.down_proj", "model.language_model.layers.6.mlp.experts.89.down_proj", "model.language_model.layers.6.mlp.experts.90.down_proj", "model.language_model.layers.6.mlp.experts.91.down_proj", "model.language_model.layers.6.mlp.experts.92.down_proj", "model.language_model.layers.6.mlp.experts.93.down_proj", "model.language_model.layers.6.mlp.experts.94.down_proj", "model.language_model.layers.6.mlp.experts.95.down_proj", "model.language_model.layers.6.mlp.experts.96.down_proj", "model.language_model.layers.6.mlp.experts.97.down_proj", "model.language_model.layers.6.mlp.experts.98.down_proj", "model.language_model.layers.6.mlp.experts.99.down_proj", "model.language_model.layers.6.mlp.experts.100.down_proj", "model.language_model.layers.6.mlp.experts.101.down_proj", "model.language_model.layers.6.mlp.experts.102.down_proj", "model.language_model.layers.6.mlp.experts.103.down_proj", "model.language_model.layers.6.mlp.experts.104.down_proj", "model.language_model.layers.6.mlp.experts.105.down_proj", "model.language_model.layers.6.mlp.experts.106.down_proj", "model.language_model.layers.6.mlp.experts.107.down_proj", "model.language_model.layers.6.mlp.experts.108.down_proj", "model.language_model.layers.6.mlp.experts.109.down_proj", "model.language_model.layers.6.mlp.experts.110.down_proj", "model.language_model.layers.6.mlp.experts.111.down_proj", "model.language_model.layers.6.mlp.experts.112.down_proj", "model.language_model.layers.6.mlp.experts.113.down_proj", "model.language_model.layers.6.mlp.experts.114.down_proj", "model.language_model.layers.6.mlp.experts.115.down_proj", "model.language_model.layers.6.mlp.experts.116.down_proj", "model.language_model.layers.6.mlp.experts.117.down_proj", "model.language_model.layers.6.mlp.experts.118.down_proj", "model.language_model.layers.6.mlp.experts.119.down_proj", "model.language_model.layers.6.mlp.experts.120.down_proj", "model.language_model.layers.6.mlp.experts.121.down_proj", "model.language_model.layers.6.mlp.experts.122.down_proj", "model.language_model.layers.6.mlp.experts.123.down_proj", "model.language_model.layers.6.mlp.experts.124.down_proj", "model.language_model.layers.6.mlp.experts.125.down_proj", "model.language_model.layers.6.mlp.experts.126.down_proj", "model.language_model.layers.6.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.002222500089555976, "dbits": 603979776 } ] }, { "idx": 14, "layers": [ "model.language_model.layers.7.self_attn.q_proj", "model.language_model.layers.7.self_attn.k_proj", "model.language_model.layers.7.self_attn.v_proj", "model.language_model.layers.7.self_attn.o_proj" ], "candidates": [ { "dkld": 0.00012396723031997403, "dbits": 18874368 } ] }, { "idx": 15, "layers": [ "model.language_model.layers.7.mlp.experts.0.gate_proj", "model.language_model.layers.7.mlp.experts.1.gate_proj", "model.language_model.layers.7.mlp.experts.2.gate_proj", "model.language_model.layers.7.mlp.experts.3.gate_proj", "model.language_model.layers.7.mlp.experts.4.gate_proj", "model.language_model.layers.7.mlp.experts.5.gate_proj", "model.language_model.layers.7.mlp.experts.6.gate_proj", "model.language_model.layers.7.mlp.experts.7.gate_proj", "model.language_model.layers.7.mlp.experts.8.gate_proj", "model.language_model.layers.7.mlp.experts.9.gate_proj", "model.language_model.layers.7.mlp.experts.10.gate_proj", "model.language_model.layers.7.mlp.experts.11.gate_proj", "model.language_model.layers.7.mlp.experts.12.gate_proj", "model.language_model.layers.7.mlp.experts.13.gate_proj", "model.language_model.layers.7.mlp.experts.14.gate_proj", "model.language_model.layers.7.mlp.experts.15.gate_proj", "model.language_model.layers.7.mlp.experts.16.gate_proj", "model.language_model.layers.7.mlp.experts.17.gate_proj", "model.language_model.layers.7.mlp.experts.18.gate_proj", "model.language_model.layers.7.mlp.experts.19.gate_proj", "model.language_model.layers.7.mlp.experts.20.gate_proj", "model.language_model.layers.7.mlp.experts.21.gate_proj", "model.language_model.layers.7.mlp.experts.22.gate_proj", "model.language_model.layers.7.mlp.experts.23.gate_proj", "model.language_model.layers.7.mlp.experts.24.gate_proj", "model.language_model.layers.7.mlp.experts.25.gate_proj", "model.language_model.layers.7.mlp.experts.26.gate_proj", "model.language_model.layers.7.mlp.experts.27.gate_proj", "model.language_model.layers.7.mlp.experts.28.gate_proj", "model.language_model.layers.7.mlp.experts.29.gate_proj", "model.language_model.layers.7.mlp.experts.30.gate_proj", "model.language_model.layers.7.mlp.experts.31.gate_proj", "model.language_model.layers.7.mlp.experts.32.gate_proj", "model.language_model.layers.7.mlp.experts.33.gate_proj", "model.language_model.layers.7.mlp.experts.34.gate_proj", "model.language_model.layers.7.mlp.experts.35.gate_proj", "model.language_model.layers.7.mlp.experts.36.gate_proj", "model.language_model.layers.7.mlp.experts.37.gate_proj", "model.language_model.layers.7.mlp.experts.38.gate_proj", "model.language_model.layers.7.mlp.experts.39.gate_proj", "model.language_model.layers.7.mlp.experts.40.gate_proj", "model.language_model.layers.7.mlp.experts.41.gate_proj", "model.language_model.layers.7.mlp.experts.42.gate_proj", "model.language_model.layers.7.mlp.experts.43.gate_proj", "model.language_model.layers.7.mlp.experts.44.gate_proj", "model.language_model.layers.7.mlp.experts.45.gate_proj", "model.language_model.layers.7.mlp.experts.46.gate_proj", "model.language_model.layers.7.mlp.experts.47.gate_proj", "model.language_model.layers.7.mlp.experts.48.gate_proj", "model.language_model.layers.7.mlp.experts.49.gate_proj", "model.language_model.layers.7.mlp.experts.50.gate_proj", "model.language_model.layers.7.mlp.experts.51.gate_proj", "model.language_model.layers.7.mlp.experts.52.gate_proj", "model.language_model.layers.7.mlp.experts.53.gate_proj", "model.language_model.layers.7.mlp.experts.54.gate_proj", "model.language_model.layers.7.mlp.experts.55.gate_proj", "model.language_model.layers.7.mlp.experts.56.gate_proj", "model.language_model.layers.7.mlp.experts.57.gate_proj", "model.language_model.layers.7.mlp.experts.58.gate_proj", "model.language_model.layers.7.mlp.experts.59.gate_proj", "model.language_model.layers.7.mlp.experts.60.gate_proj", "model.language_model.layers.7.mlp.experts.61.gate_proj", "model.language_model.layers.7.mlp.experts.62.gate_proj", "model.language_model.layers.7.mlp.experts.63.gate_proj", "model.language_model.layers.7.mlp.experts.64.gate_proj", "model.language_model.layers.7.mlp.experts.65.gate_proj", "model.language_model.layers.7.mlp.experts.66.gate_proj", "model.language_model.layers.7.mlp.experts.67.gate_proj", "model.language_model.layers.7.mlp.experts.68.gate_proj", "model.language_model.layers.7.mlp.experts.69.gate_proj", "model.language_model.layers.7.mlp.experts.70.gate_proj", "model.language_model.layers.7.mlp.experts.71.gate_proj", "model.language_model.layers.7.mlp.experts.72.gate_proj", "model.language_model.layers.7.mlp.experts.73.gate_proj", "model.language_model.layers.7.mlp.experts.74.gate_proj", "model.language_model.layers.7.mlp.experts.75.gate_proj", "model.language_model.layers.7.mlp.experts.76.gate_proj", "model.language_model.layers.7.mlp.experts.77.gate_proj", "model.language_model.layers.7.mlp.experts.78.gate_proj", "model.language_model.layers.7.mlp.experts.79.gate_proj", "model.language_model.layers.7.mlp.experts.80.gate_proj", "model.language_model.layers.7.mlp.experts.81.gate_proj", "model.language_model.layers.7.mlp.experts.82.gate_proj", "model.language_model.layers.7.mlp.experts.83.gate_proj", "model.language_model.layers.7.mlp.experts.84.gate_proj", "model.language_model.layers.7.mlp.experts.85.gate_proj", "model.language_model.layers.7.mlp.experts.86.gate_proj", "model.language_model.layers.7.mlp.experts.87.gate_proj", "model.language_model.layers.7.mlp.experts.88.gate_proj", "model.language_model.layers.7.mlp.experts.89.gate_proj", "model.language_model.layers.7.mlp.experts.90.gate_proj", "model.language_model.layers.7.mlp.experts.91.gate_proj", "model.language_model.layers.7.mlp.experts.92.gate_proj", "model.language_model.layers.7.mlp.experts.93.gate_proj", "model.language_model.layers.7.mlp.experts.94.gate_proj", "model.language_model.layers.7.mlp.experts.95.gate_proj", "model.language_model.layers.7.mlp.experts.96.gate_proj", "model.language_model.layers.7.mlp.experts.97.gate_proj", "model.language_model.layers.7.mlp.experts.98.gate_proj", "model.language_model.layers.7.mlp.experts.99.gate_proj", "model.language_model.layers.7.mlp.experts.100.gate_proj", "model.language_model.layers.7.mlp.experts.101.gate_proj", "model.language_model.layers.7.mlp.experts.102.gate_proj", "model.language_model.layers.7.mlp.experts.103.gate_proj", "model.language_model.layers.7.mlp.experts.104.gate_proj", "model.language_model.layers.7.mlp.experts.105.gate_proj", "model.language_model.layers.7.mlp.experts.106.gate_proj", "model.language_model.layers.7.mlp.experts.107.gate_proj", "model.language_model.layers.7.mlp.experts.108.gate_proj", "model.language_model.layers.7.mlp.experts.109.gate_proj", "model.language_model.layers.7.mlp.experts.110.gate_proj", "model.language_model.layers.7.mlp.experts.111.gate_proj", "model.language_model.layers.7.mlp.experts.112.gate_proj", "model.language_model.layers.7.mlp.experts.113.gate_proj", "model.language_model.layers.7.mlp.experts.114.gate_proj", "model.language_model.layers.7.mlp.experts.115.gate_proj", "model.language_model.layers.7.mlp.experts.116.gate_proj", "model.language_model.layers.7.mlp.experts.117.gate_proj", "model.language_model.layers.7.mlp.experts.118.gate_proj", "model.language_model.layers.7.mlp.experts.119.gate_proj", "model.language_model.layers.7.mlp.experts.120.gate_proj", "model.language_model.layers.7.mlp.experts.121.gate_proj", "model.language_model.layers.7.mlp.experts.122.gate_proj", "model.language_model.layers.7.mlp.experts.123.gate_proj", "model.language_model.layers.7.mlp.experts.124.gate_proj", "model.language_model.layers.7.mlp.experts.125.gate_proj", "model.language_model.layers.7.mlp.experts.126.gate_proj", "model.language_model.layers.7.mlp.experts.127.gate_proj", "model.language_model.layers.7.mlp.experts.0.up_proj", "model.language_model.layers.7.mlp.experts.1.up_proj", "model.language_model.layers.7.mlp.experts.2.up_proj", "model.language_model.layers.7.mlp.experts.3.up_proj", "model.language_model.layers.7.mlp.experts.4.up_proj", "model.language_model.layers.7.mlp.experts.5.up_proj", "model.language_model.layers.7.mlp.experts.6.up_proj", "model.language_model.layers.7.mlp.experts.7.up_proj", "model.language_model.layers.7.mlp.experts.8.up_proj", "model.language_model.layers.7.mlp.experts.9.up_proj", "model.language_model.layers.7.mlp.experts.10.up_proj", "model.language_model.layers.7.mlp.experts.11.up_proj", "model.language_model.layers.7.mlp.experts.12.up_proj", "model.language_model.layers.7.mlp.experts.13.up_proj", "model.language_model.layers.7.mlp.experts.14.up_proj", "model.language_model.layers.7.mlp.experts.15.up_proj", "model.language_model.layers.7.mlp.experts.16.up_proj", "model.language_model.layers.7.mlp.experts.17.up_proj", "model.language_model.layers.7.mlp.experts.18.up_proj", "model.language_model.layers.7.mlp.experts.19.up_proj", "model.language_model.layers.7.mlp.experts.20.up_proj", "model.language_model.layers.7.mlp.experts.21.up_proj", "model.language_model.layers.7.mlp.experts.22.up_proj", "model.language_model.layers.7.mlp.experts.23.up_proj", "model.language_model.layers.7.mlp.experts.24.up_proj", "model.language_model.layers.7.mlp.experts.25.up_proj", "model.language_model.layers.7.mlp.experts.26.up_proj", "model.language_model.layers.7.mlp.experts.27.up_proj", "model.language_model.layers.7.mlp.experts.28.up_proj", "model.language_model.layers.7.mlp.experts.29.up_proj", "model.language_model.layers.7.mlp.experts.30.up_proj", "model.language_model.layers.7.mlp.experts.31.up_proj", "model.language_model.layers.7.mlp.experts.32.up_proj", "model.language_model.layers.7.mlp.experts.33.up_proj", "model.language_model.layers.7.mlp.experts.34.up_proj", "model.language_model.layers.7.mlp.experts.35.up_proj", "model.language_model.layers.7.mlp.experts.36.up_proj", "model.language_model.layers.7.mlp.experts.37.up_proj", "model.language_model.layers.7.mlp.experts.38.up_proj", "model.language_model.layers.7.mlp.experts.39.up_proj", "model.language_model.layers.7.mlp.experts.40.up_proj", "model.language_model.layers.7.mlp.experts.41.up_proj", "model.language_model.layers.7.mlp.experts.42.up_proj", "model.language_model.layers.7.mlp.experts.43.up_proj", "model.language_model.layers.7.mlp.experts.44.up_proj", "model.language_model.layers.7.mlp.experts.45.up_proj", "model.language_model.layers.7.mlp.experts.46.up_proj", "model.language_model.layers.7.mlp.experts.47.up_proj", "model.language_model.layers.7.mlp.experts.48.up_proj", "model.language_model.layers.7.mlp.experts.49.up_proj", "model.language_model.layers.7.mlp.experts.50.up_proj", "model.language_model.layers.7.mlp.experts.51.up_proj", "model.language_model.layers.7.mlp.experts.52.up_proj", "model.language_model.layers.7.mlp.experts.53.up_proj", "model.language_model.layers.7.mlp.experts.54.up_proj", "model.language_model.layers.7.mlp.experts.55.up_proj", "model.language_model.layers.7.mlp.experts.56.up_proj", "model.language_model.layers.7.mlp.experts.57.up_proj", "model.language_model.layers.7.mlp.experts.58.up_proj", "model.language_model.layers.7.mlp.experts.59.up_proj", "model.language_model.layers.7.mlp.experts.60.up_proj", "model.language_model.layers.7.mlp.experts.61.up_proj", "model.language_model.layers.7.mlp.experts.62.up_proj", "model.language_model.layers.7.mlp.experts.63.up_proj", "model.language_model.layers.7.mlp.experts.64.up_proj", "model.language_model.layers.7.mlp.experts.65.up_proj", "model.language_model.layers.7.mlp.experts.66.up_proj", "model.language_model.layers.7.mlp.experts.67.up_proj", "model.language_model.layers.7.mlp.experts.68.up_proj", "model.language_model.layers.7.mlp.experts.69.up_proj", "model.language_model.layers.7.mlp.experts.70.up_proj", "model.language_model.layers.7.mlp.experts.71.up_proj", "model.language_model.layers.7.mlp.experts.72.up_proj", "model.language_model.layers.7.mlp.experts.73.up_proj", "model.language_model.layers.7.mlp.experts.74.up_proj", "model.language_model.layers.7.mlp.experts.75.up_proj", "model.language_model.layers.7.mlp.experts.76.up_proj", "model.language_model.layers.7.mlp.experts.77.up_proj", "model.language_model.layers.7.mlp.experts.78.up_proj", "model.language_model.layers.7.mlp.experts.79.up_proj", "model.language_model.layers.7.mlp.experts.80.up_proj", "model.language_model.layers.7.mlp.experts.81.up_proj", "model.language_model.layers.7.mlp.experts.82.up_proj", "model.language_model.layers.7.mlp.experts.83.up_proj", "model.language_model.layers.7.mlp.experts.84.up_proj", "model.language_model.layers.7.mlp.experts.85.up_proj", "model.language_model.layers.7.mlp.experts.86.up_proj", "model.language_model.layers.7.mlp.experts.87.up_proj", "model.language_model.layers.7.mlp.experts.88.up_proj", "model.language_model.layers.7.mlp.experts.89.up_proj", "model.language_model.layers.7.mlp.experts.90.up_proj", "model.language_model.layers.7.mlp.experts.91.up_proj", "model.language_model.layers.7.mlp.experts.92.up_proj", "model.language_model.layers.7.mlp.experts.93.up_proj", "model.language_model.layers.7.mlp.experts.94.up_proj", "model.language_model.layers.7.mlp.experts.95.up_proj", "model.language_model.layers.7.mlp.experts.96.up_proj", "model.language_model.layers.7.mlp.experts.97.up_proj", "model.language_model.layers.7.mlp.experts.98.up_proj", "model.language_model.layers.7.mlp.experts.99.up_proj", "model.language_model.layers.7.mlp.experts.100.up_proj", "model.language_model.layers.7.mlp.experts.101.up_proj", "model.language_model.layers.7.mlp.experts.102.up_proj", "model.language_model.layers.7.mlp.experts.103.up_proj", "model.language_model.layers.7.mlp.experts.104.up_proj", "model.language_model.layers.7.mlp.experts.105.up_proj", "model.language_model.layers.7.mlp.experts.106.up_proj", "model.language_model.layers.7.mlp.experts.107.up_proj", "model.language_model.layers.7.mlp.experts.108.up_proj", "model.language_model.layers.7.mlp.experts.109.up_proj", "model.language_model.layers.7.mlp.experts.110.up_proj", "model.language_model.layers.7.mlp.experts.111.up_proj", "model.language_model.layers.7.mlp.experts.112.up_proj", "model.language_model.layers.7.mlp.experts.113.up_proj", "model.language_model.layers.7.mlp.experts.114.up_proj", "model.language_model.layers.7.mlp.experts.115.up_proj", "model.language_model.layers.7.mlp.experts.116.up_proj", "model.language_model.layers.7.mlp.experts.117.up_proj", "model.language_model.layers.7.mlp.experts.118.up_proj", "model.language_model.layers.7.mlp.experts.119.up_proj", "model.language_model.layers.7.mlp.experts.120.up_proj", "model.language_model.layers.7.mlp.experts.121.up_proj", "model.language_model.layers.7.mlp.experts.122.up_proj", "model.language_model.layers.7.mlp.experts.123.up_proj", "model.language_model.layers.7.mlp.experts.124.up_proj", "model.language_model.layers.7.mlp.experts.125.up_proj", "model.language_model.layers.7.mlp.experts.126.up_proj", "model.language_model.layers.7.mlp.experts.127.up_proj", "model.language_model.layers.7.mlp.experts.0.down_proj", "model.language_model.layers.7.mlp.experts.1.down_proj", "model.language_model.layers.7.mlp.experts.2.down_proj", "model.language_model.layers.7.mlp.experts.3.down_proj", "model.language_model.layers.7.mlp.experts.4.down_proj", "model.language_model.layers.7.mlp.experts.5.down_proj", "model.language_model.layers.7.mlp.experts.6.down_proj", "model.language_model.layers.7.mlp.experts.7.down_proj", "model.language_model.layers.7.mlp.experts.8.down_proj", "model.language_model.layers.7.mlp.experts.9.down_proj", "model.language_model.layers.7.mlp.experts.10.down_proj", "model.language_model.layers.7.mlp.experts.11.down_proj", "model.language_model.layers.7.mlp.experts.12.down_proj", "model.language_model.layers.7.mlp.experts.13.down_proj", "model.language_model.layers.7.mlp.experts.14.down_proj", "model.language_model.layers.7.mlp.experts.15.down_proj", "model.language_model.layers.7.mlp.experts.16.down_proj", "model.language_model.layers.7.mlp.experts.17.down_proj", "model.language_model.layers.7.mlp.experts.18.down_proj", "model.language_model.layers.7.mlp.experts.19.down_proj", "model.language_model.layers.7.mlp.experts.20.down_proj", "model.language_model.layers.7.mlp.experts.21.down_proj", "model.language_model.layers.7.mlp.experts.22.down_proj", "model.language_model.layers.7.mlp.experts.23.down_proj", "model.language_model.layers.7.mlp.experts.24.down_proj", "model.language_model.layers.7.mlp.experts.25.down_proj", "model.language_model.layers.7.mlp.experts.26.down_proj", "model.language_model.layers.7.mlp.experts.27.down_proj", "model.language_model.layers.7.mlp.experts.28.down_proj", "model.language_model.layers.7.mlp.experts.29.down_proj", "model.language_model.layers.7.mlp.experts.30.down_proj", "model.language_model.layers.7.mlp.experts.31.down_proj", "model.language_model.layers.7.mlp.experts.32.down_proj", "model.language_model.layers.7.mlp.experts.33.down_proj", "model.language_model.layers.7.mlp.experts.34.down_proj", "model.language_model.layers.7.mlp.experts.35.down_proj", "model.language_model.layers.7.mlp.experts.36.down_proj", "model.language_model.layers.7.mlp.experts.37.down_proj", "model.language_model.layers.7.mlp.experts.38.down_proj", "model.language_model.layers.7.mlp.experts.39.down_proj", "model.language_model.layers.7.mlp.experts.40.down_proj", "model.language_model.layers.7.mlp.experts.41.down_proj", "model.language_model.layers.7.mlp.experts.42.down_proj", "model.language_model.layers.7.mlp.experts.43.down_proj", "model.language_model.layers.7.mlp.experts.44.down_proj", "model.language_model.layers.7.mlp.experts.45.down_proj", "model.language_model.layers.7.mlp.experts.46.down_proj", "model.language_model.layers.7.mlp.experts.47.down_proj", "model.language_model.layers.7.mlp.experts.48.down_proj", "model.language_model.layers.7.mlp.experts.49.down_proj", "model.language_model.layers.7.mlp.experts.50.down_proj", "model.language_model.layers.7.mlp.experts.51.down_proj", "model.language_model.layers.7.mlp.experts.52.down_proj", "model.language_model.layers.7.mlp.experts.53.down_proj", "model.language_model.layers.7.mlp.experts.54.down_proj", "model.language_model.layers.7.mlp.experts.55.down_proj", "model.language_model.layers.7.mlp.experts.56.down_proj", "model.language_model.layers.7.mlp.experts.57.down_proj", "model.language_model.layers.7.mlp.experts.58.down_proj", "model.language_model.layers.7.mlp.experts.59.down_proj", "model.language_model.layers.7.mlp.experts.60.down_proj", "model.language_model.layers.7.mlp.experts.61.down_proj", "model.language_model.layers.7.mlp.experts.62.down_proj", "model.language_model.layers.7.mlp.experts.63.down_proj", "model.language_model.layers.7.mlp.experts.64.down_proj", "model.language_model.layers.7.mlp.experts.65.down_proj", "model.language_model.layers.7.mlp.experts.66.down_proj", "model.language_model.layers.7.mlp.experts.67.down_proj", "model.language_model.layers.7.mlp.experts.68.down_proj", "model.language_model.layers.7.mlp.experts.69.down_proj", "model.language_model.layers.7.mlp.experts.70.down_proj", "model.language_model.layers.7.mlp.experts.71.down_proj", "model.language_model.layers.7.mlp.experts.72.down_proj", "model.language_model.layers.7.mlp.experts.73.down_proj", "model.language_model.layers.7.mlp.experts.74.down_proj", "model.language_model.layers.7.mlp.experts.75.down_proj", "model.language_model.layers.7.mlp.experts.76.down_proj", "model.language_model.layers.7.mlp.experts.77.down_proj", "model.language_model.layers.7.mlp.experts.78.down_proj", "model.language_model.layers.7.mlp.experts.79.down_proj", "model.language_model.layers.7.mlp.experts.80.down_proj", "model.language_model.layers.7.mlp.experts.81.down_proj", "model.language_model.layers.7.mlp.experts.82.down_proj", "model.language_model.layers.7.mlp.experts.83.down_proj", "model.language_model.layers.7.mlp.experts.84.down_proj", "model.language_model.layers.7.mlp.experts.85.down_proj", "model.language_model.layers.7.mlp.experts.86.down_proj", "model.language_model.layers.7.mlp.experts.87.down_proj", "model.language_model.layers.7.mlp.experts.88.down_proj", "model.language_model.layers.7.mlp.experts.89.down_proj", "model.language_model.layers.7.mlp.experts.90.down_proj", "model.language_model.layers.7.mlp.experts.91.down_proj", "model.language_model.layers.7.mlp.experts.92.down_proj", "model.language_model.layers.7.mlp.experts.93.down_proj", "model.language_model.layers.7.mlp.experts.94.down_proj", "model.language_model.layers.7.mlp.experts.95.down_proj", "model.language_model.layers.7.mlp.experts.96.down_proj", "model.language_model.layers.7.mlp.experts.97.down_proj", "model.language_model.layers.7.mlp.experts.98.down_proj", "model.language_model.layers.7.mlp.experts.99.down_proj", "model.language_model.layers.7.mlp.experts.100.down_proj", "model.language_model.layers.7.mlp.experts.101.down_proj", "model.language_model.layers.7.mlp.experts.102.down_proj", "model.language_model.layers.7.mlp.experts.103.down_proj", "model.language_model.layers.7.mlp.experts.104.down_proj", "model.language_model.layers.7.mlp.experts.105.down_proj", "model.language_model.layers.7.mlp.experts.106.down_proj", "model.language_model.layers.7.mlp.experts.107.down_proj", "model.language_model.layers.7.mlp.experts.108.down_proj", "model.language_model.layers.7.mlp.experts.109.down_proj", "model.language_model.layers.7.mlp.experts.110.down_proj", "model.language_model.layers.7.mlp.experts.111.down_proj", "model.language_model.layers.7.mlp.experts.112.down_proj", "model.language_model.layers.7.mlp.experts.113.down_proj", "model.language_model.layers.7.mlp.experts.114.down_proj", "model.language_model.layers.7.mlp.experts.115.down_proj", "model.language_model.layers.7.mlp.experts.116.down_proj", "model.language_model.layers.7.mlp.experts.117.down_proj", "model.language_model.layers.7.mlp.experts.118.down_proj", "model.language_model.layers.7.mlp.experts.119.down_proj", "model.language_model.layers.7.mlp.experts.120.down_proj", "model.language_model.layers.7.mlp.experts.121.down_proj", "model.language_model.layers.7.mlp.experts.122.down_proj", "model.language_model.layers.7.mlp.experts.123.down_proj", "model.language_model.layers.7.mlp.experts.124.down_proj", "model.language_model.layers.7.mlp.experts.125.down_proj", "model.language_model.layers.7.mlp.experts.126.down_proj", "model.language_model.layers.7.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.0018762143328785869, "dbits": 603979776 } ] }, { "idx": 16, "layers": [ "model.language_model.layers.8.self_attn.q_proj", "model.language_model.layers.8.self_attn.k_proj", "model.language_model.layers.8.self_attn.v_proj", "model.language_model.layers.8.self_attn.o_proj" ], "candidates": [ { "dkld": -0.001417303644120696, "dbits": 18874368 } ] }, { "idx": 17, "layers": [ "model.language_model.layers.8.mlp.experts.0.gate_proj", "model.language_model.layers.8.mlp.experts.1.gate_proj", "model.language_model.layers.8.mlp.experts.2.gate_proj", "model.language_model.layers.8.mlp.experts.3.gate_proj", "model.language_model.layers.8.mlp.experts.4.gate_proj", "model.language_model.layers.8.mlp.experts.5.gate_proj", "model.language_model.layers.8.mlp.experts.6.gate_proj", "model.language_model.layers.8.mlp.experts.7.gate_proj", "model.language_model.layers.8.mlp.experts.8.gate_proj", "model.language_model.layers.8.mlp.experts.9.gate_proj", "model.language_model.layers.8.mlp.experts.10.gate_proj", "model.language_model.layers.8.mlp.experts.11.gate_proj", "model.language_model.layers.8.mlp.experts.12.gate_proj", "model.language_model.layers.8.mlp.experts.13.gate_proj", "model.language_model.layers.8.mlp.experts.14.gate_proj", "model.language_model.layers.8.mlp.experts.15.gate_proj", "model.language_model.layers.8.mlp.experts.16.gate_proj", "model.language_model.layers.8.mlp.experts.17.gate_proj", "model.language_model.layers.8.mlp.experts.18.gate_proj", "model.language_model.layers.8.mlp.experts.19.gate_proj", "model.language_model.layers.8.mlp.experts.20.gate_proj", "model.language_model.layers.8.mlp.experts.21.gate_proj", "model.language_model.layers.8.mlp.experts.22.gate_proj", "model.language_model.layers.8.mlp.experts.23.gate_proj", "model.language_model.layers.8.mlp.experts.24.gate_proj", "model.language_model.layers.8.mlp.experts.25.gate_proj", "model.language_model.layers.8.mlp.experts.26.gate_proj", "model.language_model.layers.8.mlp.experts.27.gate_proj", "model.language_model.layers.8.mlp.experts.28.gate_proj", "model.language_model.layers.8.mlp.experts.29.gate_proj", "model.language_model.layers.8.mlp.experts.30.gate_proj", "model.language_model.layers.8.mlp.experts.31.gate_proj", "model.language_model.layers.8.mlp.experts.32.gate_proj", "model.language_model.layers.8.mlp.experts.33.gate_proj", "model.language_model.layers.8.mlp.experts.34.gate_proj", "model.language_model.layers.8.mlp.experts.35.gate_proj", "model.language_model.layers.8.mlp.experts.36.gate_proj", "model.language_model.layers.8.mlp.experts.37.gate_proj", "model.language_model.layers.8.mlp.experts.38.gate_proj", "model.language_model.layers.8.mlp.experts.39.gate_proj", "model.language_model.layers.8.mlp.experts.40.gate_proj", "model.language_model.layers.8.mlp.experts.41.gate_proj", "model.language_model.layers.8.mlp.experts.42.gate_proj", "model.language_model.layers.8.mlp.experts.43.gate_proj", "model.language_model.layers.8.mlp.experts.44.gate_proj", "model.language_model.layers.8.mlp.experts.45.gate_proj", "model.language_model.layers.8.mlp.experts.46.gate_proj", "model.language_model.layers.8.mlp.experts.47.gate_proj", "model.language_model.layers.8.mlp.experts.48.gate_proj", "model.language_model.layers.8.mlp.experts.49.gate_proj", "model.language_model.layers.8.mlp.experts.50.gate_proj", "model.language_model.layers.8.mlp.experts.51.gate_proj", "model.language_model.layers.8.mlp.experts.52.gate_proj", "model.language_model.layers.8.mlp.experts.53.gate_proj", "model.language_model.layers.8.mlp.experts.54.gate_proj", "model.language_model.layers.8.mlp.experts.55.gate_proj", "model.language_model.layers.8.mlp.experts.56.gate_proj", "model.language_model.layers.8.mlp.experts.57.gate_proj", "model.language_model.layers.8.mlp.experts.58.gate_proj", "model.language_model.layers.8.mlp.experts.59.gate_proj", "model.language_model.layers.8.mlp.experts.60.gate_proj", "model.language_model.layers.8.mlp.experts.61.gate_proj", "model.language_model.layers.8.mlp.experts.62.gate_proj", "model.language_model.layers.8.mlp.experts.63.gate_proj", "model.language_model.layers.8.mlp.experts.64.gate_proj", "model.language_model.layers.8.mlp.experts.65.gate_proj", "model.language_model.layers.8.mlp.experts.66.gate_proj", "model.language_model.layers.8.mlp.experts.67.gate_proj", "model.language_model.layers.8.mlp.experts.68.gate_proj", "model.language_model.layers.8.mlp.experts.69.gate_proj", "model.language_model.layers.8.mlp.experts.70.gate_proj", "model.language_model.layers.8.mlp.experts.71.gate_proj", "model.language_model.layers.8.mlp.experts.72.gate_proj", "model.language_model.layers.8.mlp.experts.73.gate_proj", "model.language_model.layers.8.mlp.experts.74.gate_proj", "model.language_model.layers.8.mlp.experts.75.gate_proj", "model.language_model.layers.8.mlp.experts.76.gate_proj", "model.language_model.layers.8.mlp.experts.77.gate_proj", "model.language_model.layers.8.mlp.experts.78.gate_proj", "model.language_model.layers.8.mlp.experts.79.gate_proj", "model.language_model.layers.8.mlp.experts.80.gate_proj", "model.language_model.layers.8.mlp.experts.81.gate_proj", "model.language_model.layers.8.mlp.experts.82.gate_proj", "model.language_model.layers.8.mlp.experts.83.gate_proj", "model.language_model.layers.8.mlp.experts.84.gate_proj", "model.language_model.layers.8.mlp.experts.85.gate_proj", "model.language_model.layers.8.mlp.experts.86.gate_proj", "model.language_model.layers.8.mlp.experts.87.gate_proj", "model.language_model.layers.8.mlp.experts.88.gate_proj", "model.language_model.layers.8.mlp.experts.89.gate_proj", "model.language_model.layers.8.mlp.experts.90.gate_proj", "model.language_model.layers.8.mlp.experts.91.gate_proj", "model.language_model.layers.8.mlp.experts.92.gate_proj", "model.language_model.layers.8.mlp.experts.93.gate_proj", "model.language_model.layers.8.mlp.experts.94.gate_proj", "model.language_model.layers.8.mlp.experts.95.gate_proj", "model.language_model.layers.8.mlp.experts.96.gate_proj", "model.language_model.layers.8.mlp.experts.97.gate_proj", "model.language_model.layers.8.mlp.experts.98.gate_proj", "model.language_model.layers.8.mlp.experts.99.gate_proj", "model.language_model.layers.8.mlp.experts.100.gate_proj", "model.language_model.layers.8.mlp.experts.101.gate_proj", "model.language_model.layers.8.mlp.experts.102.gate_proj", "model.language_model.layers.8.mlp.experts.103.gate_proj", "model.language_model.layers.8.mlp.experts.104.gate_proj", "model.language_model.layers.8.mlp.experts.105.gate_proj", "model.language_model.layers.8.mlp.experts.106.gate_proj", "model.language_model.layers.8.mlp.experts.107.gate_proj", "model.language_model.layers.8.mlp.experts.108.gate_proj", "model.language_model.layers.8.mlp.experts.109.gate_proj", "model.language_model.layers.8.mlp.experts.110.gate_proj", "model.language_model.layers.8.mlp.experts.111.gate_proj", "model.language_model.layers.8.mlp.experts.112.gate_proj", "model.language_model.layers.8.mlp.experts.113.gate_proj", "model.language_model.layers.8.mlp.experts.114.gate_proj", "model.language_model.layers.8.mlp.experts.115.gate_proj", "model.language_model.layers.8.mlp.experts.116.gate_proj", "model.language_model.layers.8.mlp.experts.117.gate_proj", "model.language_model.layers.8.mlp.experts.118.gate_proj", "model.language_model.layers.8.mlp.experts.119.gate_proj", "model.language_model.layers.8.mlp.experts.120.gate_proj", "model.language_model.layers.8.mlp.experts.121.gate_proj", "model.language_model.layers.8.mlp.experts.122.gate_proj", "model.language_model.layers.8.mlp.experts.123.gate_proj", "model.language_model.layers.8.mlp.experts.124.gate_proj", "model.language_model.layers.8.mlp.experts.125.gate_proj", "model.language_model.layers.8.mlp.experts.126.gate_proj", "model.language_model.layers.8.mlp.experts.127.gate_proj", "model.language_model.layers.8.mlp.experts.0.up_proj", "model.language_model.layers.8.mlp.experts.1.up_proj", "model.language_model.layers.8.mlp.experts.2.up_proj", "model.language_model.layers.8.mlp.experts.3.up_proj", "model.language_model.layers.8.mlp.experts.4.up_proj", "model.language_model.layers.8.mlp.experts.5.up_proj", "model.language_model.layers.8.mlp.experts.6.up_proj", "model.language_model.layers.8.mlp.experts.7.up_proj", "model.language_model.layers.8.mlp.experts.8.up_proj", "model.language_model.layers.8.mlp.experts.9.up_proj", "model.language_model.layers.8.mlp.experts.10.up_proj", "model.language_model.layers.8.mlp.experts.11.up_proj", "model.language_model.layers.8.mlp.experts.12.up_proj", "model.language_model.layers.8.mlp.experts.13.up_proj", "model.language_model.layers.8.mlp.experts.14.up_proj", "model.language_model.layers.8.mlp.experts.15.up_proj", "model.language_model.layers.8.mlp.experts.16.up_proj", "model.language_model.layers.8.mlp.experts.17.up_proj", "model.language_model.layers.8.mlp.experts.18.up_proj", "model.language_model.layers.8.mlp.experts.19.up_proj", "model.language_model.layers.8.mlp.experts.20.up_proj", "model.language_model.layers.8.mlp.experts.21.up_proj", "model.language_model.layers.8.mlp.experts.22.up_proj", "model.language_model.layers.8.mlp.experts.23.up_proj", "model.language_model.layers.8.mlp.experts.24.up_proj", "model.language_model.layers.8.mlp.experts.25.up_proj", "model.language_model.layers.8.mlp.experts.26.up_proj", "model.language_model.layers.8.mlp.experts.27.up_proj", "model.language_model.layers.8.mlp.experts.28.up_proj", "model.language_model.layers.8.mlp.experts.29.up_proj", "model.language_model.layers.8.mlp.experts.30.up_proj", "model.language_model.layers.8.mlp.experts.31.up_proj", "model.language_model.layers.8.mlp.experts.32.up_proj", "model.language_model.layers.8.mlp.experts.33.up_proj", "model.language_model.layers.8.mlp.experts.34.up_proj", "model.language_model.layers.8.mlp.experts.35.up_proj", "model.language_model.layers.8.mlp.experts.36.up_proj", "model.language_model.layers.8.mlp.experts.37.up_proj", "model.language_model.layers.8.mlp.experts.38.up_proj", "model.language_model.layers.8.mlp.experts.39.up_proj", "model.language_model.layers.8.mlp.experts.40.up_proj", "model.language_model.layers.8.mlp.experts.41.up_proj", "model.language_model.layers.8.mlp.experts.42.up_proj", "model.language_model.layers.8.mlp.experts.43.up_proj", "model.language_model.layers.8.mlp.experts.44.up_proj", "model.language_model.layers.8.mlp.experts.45.up_proj", "model.language_model.layers.8.mlp.experts.46.up_proj", "model.language_model.layers.8.mlp.experts.47.up_proj", "model.language_model.layers.8.mlp.experts.48.up_proj", "model.language_model.layers.8.mlp.experts.49.up_proj", "model.language_model.layers.8.mlp.experts.50.up_proj", "model.language_model.layers.8.mlp.experts.51.up_proj", "model.language_model.layers.8.mlp.experts.52.up_proj", "model.language_model.layers.8.mlp.experts.53.up_proj", "model.language_model.layers.8.mlp.experts.54.up_proj", "model.language_model.layers.8.mlp.experts.55.up_proj", "model.language_model.layers.8.mlp.experts.56.up_proj", "model.language_model.layers.8.mlp.experts.57.up_proj", "model.language_model.layers.8.mlp.experts.58.up_proj", "model.language_model.layers.8.mlp.experts.59.up_proj", "model.language_model.layers.8.mlp.experts.60.up_proj", "model.language_model.layers.8.mlp.experts.61.up_proj", "model.language_model.layers.8.mlp.experts.62.up_proj", "model.language_model.layers.8.mlp.experts.63.up_proj", "model.language_model.layers.8.mlp.experts.64.up_proj", "model.language_model.layers.8.mlp.experts.65.up_proj", "model.language_model.layers.8.mlp.experts.66.up_proj", "model.language_model.layers.8.mlp.experts.67.up_proj", "model.language_model.layers.8.mlp.experts.68.up_proj", "model.language_model.layers.8.mlp.experts.69.up_proj", "model.language_model.layers.8.mlp.experts.70.up_proj", "model.language_model.layers.8.mlp.experts.71.up_proj", "model.language_model.layers.8.mlp.experts.72.up_proj", "model.language_model.layers.8.mlp.experts.73.up_proj", "model.language_model.layers.8.mlp.experts.74.up_proj", "model.language_model.layers.8.mlp.experts.75.up_proj", "model.language_model.layers.8.mlp.experts.76.up_proj", "model.language_model.layers.8.mlp.experts.77.up_proj", "model.language_model.layers.8.mlp.experts.78.up_proj", "model.language_model.layers.8.mlp.experts.79.up_proj", "model.language_model.layers.8.mlp.experts.80.up_proj", "model.language_model.layers.8.mlp.experts.81.up_proj", "model.language_model.layers.8.mlp.experts.82.up_proj", "model.language_model.layers.8.mlp.experts.83.up_proj", "model.language_model.layers.8.mlp.experts.84.up_proj", "model.language_model.layers.8.mlp.experts.85.up_proj", "model.language_model.layers.8.mlp.experts.86.up_proj", "model.language_model.layers.8.mlp.experts.87.up_proj", "model.language_model.layers.8.mlp.experts.88.up_proj", "model.language_model.layers.8.mlp.experts.89.up_proj", "model.language_model.layers.8.mlp.experts.90.up_proj", "model.language_model.layers.8.mlp.experts.91.up_proj", "model.language_model.layers.8.mlp.experts.92.up_proj", "model.language_model.layers.8.mlp.experts.93.up_proj", "model.language_model.layers.8.mlp.experts.94.up_proj", "model.language_model.layers.8.mlp.experts.95.up_proj", "model.language_model.layers.8.mlp.experts.96.up_proj", "model.language_model.layers.8.mlp.experts.97.up_proj", "model.language_model.layers.8.mlp.experts.98.up_proj", "model.language_model.layers.8.mlp.experts.99.up_proj", "model.language_model.layers.8.mlp.experts.100.up_proj", "model.language_model.layers.8.mlp.experts.101.up_proj", "model.language_model.layers.8.mlp.experts.102.up_proj", "model.language_model.layers.8.mlp.experts.103.up_proj", "model.language_model.layers.8.mlp.experts.104.up_proj", "model.language_model.layers.8.mlp.experts.105.up_proj", "model.language_model.layers.8.mlp.experts.106.up_proj", "model.language_model.layers.8.mlp.experts.107.up_proj", "model.language_model.layers.8.mlp.experts.108.up_proj", "model.language_model.layers.8.mlp.experts.109.up_proj", "model.language_model.layers.8.mlp.experts.110.up_proj", "model.language_model.layers.8.mlp.experts.111.up_proj", "model.language_model.layers.8.mlp.experts.112.up_proj", "model.language_model.layers.8.mlp.experts.113.up_proj", "model.language_model.layers.8.mlp.experts.114.up_proj", "model.language_model.layers.8.mlp.experts.115.up_proj", "model.language_model.layers.8.mlp.experts.116.up_proj", "model.language_model.layers.8.mlp.experts.117.up_proj", "model.language_model.layers.8.mlp.experts.118.up_proj", "model.language_model.layers.8.mlp.experts.119.up_proj", "model.language_model.layers.8.mlp.experts.120.up_proj", "model.language_model.layers.8.mlp.experts.121.up_proj", "model.language_model.layers.8.mlp.experts.122.up_proj", "model.language_model.layers.8.mlp.experts.123.up_proj", "model.language_model.layers.8.mlp.experts.124.up_proj", "model.language_model.layers.8.mlp.experts.125.up_proj", "model.language_model.layers.8.mlp.experts.126.up_proj", "model.language_model.layers.8.mlp.experts.127.up_proj", "model.language_model.layers.8.mlp.experts.0.down_proj", "model.language_model.layers.8.mlp.experts.1.down_proj", "model.language_model.layers.8.mlp.experts.2.down_proj", "model.language_model.layers.8.mlp.experts.3.down_proj", "model.language_model.layers.8.mlp.experts.4.down_proj", "model.language_model.layers.8.mlp.experts.5.down_proj", "model.language_model.layers.8.mlp.experts.6.down_proj", "model.language_model.layers.8.mlp.experts.7.down_proj", "model.language_model.layers.8.mlp.experts.8.down_proj", "model.language_model.layers.8.mlp.experts.9.down_proj", "model.language_model.layers.8.mlp.experts.10.down_proj", "model.language_model.layers.8.mlp.experts.11.down_proj", "model.language_model.layers.8.mlp.experts.12.down_proj", "model.language_model.layers.8.mlp.experts.13.down_proj", "model.language_model.layers.8.mlp.experts.14.down_proj", "model.language_model.layers.8.mlp.experts.15.down_proj", "model.language_model.layers.8.mlp.experts.16.down_proj", "model.language_model.layers.8.mlp.experts.17.down_proj", "model.language_model.layers.8.mlp.experts.18.down_proj", "model.language_model.layers.8.mlp.experts.19.down_proj", "model.language_model.layers.8.mlp.experts.20.down_proj", "model.language_model.layers.8.mlp.experts.21.down_proj", "model.language_model.layers.8.mlp.experts.22.down_proj", "model.language_model.layers.8.mlp.experts.23.down_proj", "model.language_model.layers.8.mlp.experts.24.down_proj", "model.language_model.layers.8.mlp.experts.25.down_proj", "model.language_model.layers.8.mlp.experts.26.down_proj", "model.language_model.layers.8.mlp.experts.27.down_proj", "model.language_model.layers.8.mlp.experts.28.down_proj", "model.language_model.layers.8.mlp.experts.29.down_proj", "model.language_model.layers.8.mlp.experts.30.down_proj", "model.language_model.layers.8.mlp.experts.31.down_proj", "model.language_model.layers.8.mlp.experts.32.down_proj", "model.language_model.layers.8.mlp.experts.33.down_proj", "model.language_model.layers.8.mlp.experts.34.down_proj", "model.language_model.layers.8.mlp.experts.35.down_proj", "model.language_model.layers.8.mlp.experts.36.down_proj", "model.language_model.layers.8.mlp.experts.37.down_proj", "model.language_model.layers.8.mlp.experts.38.down_proj", "model.language_model.layers.8.mlp.experts.39.down_proj", "model.language_model.layers.8.mlp.experts.40.down_proj", "model.language_model.layers.8.mlp.experts.41.down_proj", "model.language_model.layers.8.mlp.experts.42.down_proj", "model.language_model.layers.8.mlp.experts.43.down_proj", "model.language_model.layers.8.mlp.experts.44.down_proj", "model.language_model.layers.8.mlp.experts.45.down_proj", "model.language_model.layers.8.mlp.experts.46.down_proj", "model.language_model.layers.8.mlp.experts.47.down_proj", "model.language_model.layers.8.mlp.experts.48.down_proj", "model.language_model.layers.8.mlp.experts.49.down_proj", "model.language_model.layers.8.mlp.experts.50.down_proj", "model.language_model.layers.8.mlp.experts.51.down_proj", "model.language_model.layers.8.mlp.experts.52.down_proj", "model.language_model.layers.8.mlp.experts.53.down_proj", "model.language_model.layers.8.mlp.experts.54.down_proj", "model.language_model.layers.8.mlp.experts.55.down_proj", "model.language_model.layers.8.mlp.experts.56.down_proj", "model.language_model.layers.8.mlp.experts.57.down_proj", "model.language_model.layers.8.mlp.experts.58.down_proj", "model.language_model.layers.8.mlp.experts.59.down_proj", "model.language_model.layers.8.mlp.experts.60.down_proj", "model.language_model.layers.8.mlp.experts.61.down_proj", "model.language_model.layers.8.mlp.experts.62.down_proj", "model.language_model.layers.8.mlp.experts.63.down_proj", "model.language_model.layers.8.mlp.experts.64.down_proj", "model.language_model.layers.8.mlp.experts.65.down_proj", "model.language_model.layers.8.mlp.experts.66.down_proj", "model.language_model.layers.8.mlp.experts.67.down_proj", "model.language_model.layers.8.mlp.experts.68.down_proj", "model.language_model.layers.8.mlp.experts.69.down_proj", "model.language_model.layers.8.mlp.experts.70.down_proj", "model.language_model.layers.8.mlp.experts.71.down_proj", "model.language_model.layers.8.mlp.experts.72.down_proj", "model.language_model.layers.8.mlp.experts.73.down_proj", "model.language_model.layers.8.mlp.experts.74.down_proj", "model.language_model.layers.8.mlp.experts.75.down_proj", "model.language_model.layers.8.mlp.experts.76.down_proj", "model.language_model.layers.8.mlp.experts.77.down_proj", "model.language_model.layers.8.mlp.experts.78.down_proj", "model.language_model.layers.8.mlp.experts.79.down_proj", "model.language_model.layers.8.mlp.experts.80.down_proj", "model.language_model.layers.8.mlp.experts.81.down_proj", "model.language_model.layers.8.mlp.experts.82.down_proj", "model.language_model.layers.8.mlp.experts.83.down_proj", "model.language_model.layers.8.mlp.experts.84.down_proj", "model.language_model.layers.8.mlp.experts.85.down_proj", "model.language_model.layers.8.mlp.experts.86.down_proj", "model.language_model.layers.8.mlp.experts.87.down_proj", "model.language_model.layers.8.mlp.experts.88.down_proj", "model.language_model.layers.8.mlp.experts.89.down_proj", "model.language_model.layers.8.mlp.experts.90.down_proj", "model.language_model.layers.8.mlp.experts.91.down_proj", "model.language_model.layers.8.mlp.experts.92.down_proj", "model.language_model.layers.8.mlp.experts.93.down_proj", "model.language_model.layers.8.mlp.experts.94.down_proj", "model.language_model.layers.8.mlp.experts.95.down_proj", "model.language_model.layers.8.mlp.experts.96.down_proj", "model.language_model.layers.8.mlp.experts.97.down_proj", "model.language_model.layers.8.mlp.experts.98.down_proj", "model.language_model.layers.8.mlp.experts.99.down_proj", "model.language_model.layers.8.mlp.experts.100.down_proj", "model.language_model.layers.8.mlp.experts.101.down_proj", "model.language_model.layers.8.mlp.experts.102.down_proj", "model.language_model.layers.8.mlp.experts.103.down_proj", "model.language_model.layers.8.mlp.experts.104.down_proj", "model.language_model.layers.8.mlp.experts.105.down_proj", "model.language_model.layers.8.mlp.experts.106.down_proj", "model.language_model.layers.8.mlp.experts.107.down_proj", "model.language_model.layers.8.mlp.experts.108.down_proj", "model.language_model.layers.8.mlp.experts.109.down_proj", "model.language_model.layers.8.mlp.experts.110.down_proj", "model.language_model.layers.8.mlp.experts.111.down_proj", "model.language_model.layers.8.mlp.experts.112.down_proj", "model.language_model.layers.8.mlp.experts.113.down_proj", "model.language_model.layers.8.mlp.experts.114.down_proj", "model.language_model.layers.8.mlp.experts.115.down_proj", "model.language_model.layers.8.mlp.experts.116.down_proj", "model.language_model.layers.8.mlp.experts.117.down_proj", "model.language_model.layers.8.mlp.experts.118.down_proj", "model.language_model.layers.8.mlp.experts.119.down_proj", "model.language_model.layers.8.mlp.experts.120.down_proj", "model.language_model.layers.8.mlp.experts.121.down_proj", "model.language_model.layers.8.mlp.experts.122.down_proj", "model.language_model.layers.8.mlp.experts.123.down_proj", "model.language_model.layers.8.mlp.experts.124.down_proj", "model.language_model.layers.8.mlp.experts.125.down_proj", "model.language_model.layers.8.mlp.experts.126.down_proj", "model.language_model.layers.8.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.0016236869618296568, "dbits": 603979776 } ] }, { "idx": 18, "layers": [ "model.language_model.layers.9.self_attn.q_proj", "model.language_model.layers.9.self_attn.k_proj", "model.language_model.layers.9.self_attn.v_proj", "model.language_model.layers.9.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0012734594754874734, "dbits": 18874368 } ] }, { "idx": 19, "layers": [ "model.language_model.layers.9.mlp.experts.0.gate_proj", "model.language_model.layers.9.mlp.experts.1.gate_proj", "model.language_model.layers.9.mlp.experts.2.gate_proj", "model.language_model.layers.9.mlp.experts.3.gate_proj", "model.language_model.layers.9.mlp.experts.4.gate_proj", "model.language_model.layers.9.mlp.experts.5.gate_proj", "model.language_model.layers.9.mlp.experts.6.gate_proj", "model.language_model.layers.9.mlp.experts.7.gate_proj", "model.language_model.layers.9.mlp.experts.8.gate_proj", "model.language_model.layers.9.mlp.experts.9.gate_proj", "model.language_model.layers.9.mlp.experts.10.gate_proj", "model.language_model.layers.9.mlp.experts.11.gate_proj", "model.language_model.layers.9.mlp.experts.12.gate_proj", "model.language_model.layers.9.mlp.experts.13.gate_proj", "model.language_model.layers.9.mlp.experts.14.gate_proj", "model.language_model.layers.9.mlp.experts.15.gate_proj", "model.language_model.layers.9.mlp.experts.16.gate_proj", "model.language_model.layers.9.mlp.experts.17.gate_proj", "model.language_model.layers.9.mlp.experts.18.gate_proj", "model.language_model.layers.9.mlp.experts.19.gate_proj", "model.language_model.layers.9.mlp.experts.20.gate_proj", "model.language_model.layers.9.mlp.experts.21.gate_proj", "model.language_model.layers.9.mlp.experts.22.gate_proj", "model.language_model.layers.9.mlp.experts.23.gate_proj", "model.language_model.layers.9.mlp.experts.24.gate_proj", "model.language_model.layers.9.mlp.experts.25.gate_proj", "model.language_model.layers.9.mlp.experts.26.gate_proj", "model.language_model.layers.9.mlp.experts.27.gate_proj", "model.language_model.layers.9.mlp.experts.28.gate_proj", "model.language_model.layers.9.mlp.experts.29.gate_proj", "model.language_model.layers.9.mlp.experts.30.gate_proj", "model.language_model.layers.9.mlp.experts.31.gate_proj", "model.language_model.layers.9.mlp.experts.32.gate_proj", "model.language_model.layers.9.mlp.experts.33.gate_proj", "model.language_model.layers.9.mlp.experts.34.gate_proj", "model.language_model.layers.9.mlp.experts.35.gate_proj", "model.language_model.layers.9.mlp.experts.36.gate_proj", "model.language_model.layers.9.mlp.experts.37.gate_proj", "model.language_model.layers.9.mlp.experts.38.gate_proj", "model.language_model.layers.9.mlp.experts.39.gate_proj", "model.language_model.layers.9.mlp.experts.40.gate_proj", "model.language_model.layers.9.mlp.experts.41.gate_proj", "model.language_model.layers.9.mlp.experts.42.gate_proj", "model.language_model.layers.9.mlp.experts.43.gate_proj", "model.language_model.layers.9.mlp.experts.44.gate_proj", "model.language_model.layers.9.mlp.experts.45.gate_proj", "model.language_model.layers.9.mlp.experts.46.gate_proj", "model.language_model.layers.9.mlp.experts.47.gate_proj", "model.language_model.layers.9.mlp.experts.48.gate_proj", "model.language_model.layers.9.mlp.experts.49.gate_proj", "model.language_model.layers.9.mlp.experts.50.gate_proj", "model.language_model.layers.9.mlp.experts.51.gate_proj", "model.language_model.layers.9.mlp.experts.52.gate_proj", "model.language_model.layers.9.mlp.experts.53.gate_proj", "model.language_model.layers.9.mlp.experts.54.gate_proj", "model.language_model.layers.9.mlp.experts.55.gate_proj", "model.language_model.layers.9.mlp.experts.56.gate_proj", "model.language_model.layers.9.mlp.experts.57.gate_proj", "model.language_model.layers.9.mlp.experts.58.gate_proj", "model.language_model.layers.9.mlp.experts.59.gate_proj", "model.language_model.layers.9.mlp.experts.60.gate_proj", "model.language_model.layers.9.mlp.experts.61.gate_proj", "model.language_model.layers.9.mlp.experts.62.gate_proj", "model.language_model.layers.9.mlp.experts.63.gate_proj", "model.language_model.layers.9.mlp.experts.64.gate_proj", "model.language_model.layers.9.mlp.experts.65.gate_proj", "model.language_model.layers.9.mlp.experts.66.gate_proj", "model.language_model.layers.9.mlp.experts.67.gate_proj", "model.language_model.layers.9.mlp.experts.68.gate_proj", "model.language_model.layers.9.mlp.experts.69.gate_proj", "model.language_model.layers.9.mlp.experts.70.gate_proj", "model.language_model.layers.9.mlp.experts.71.gate_proj", "model.language_model.layers.9.mlp.experts.72.gate_proj", "model.language_model.layers.9.mlp.experts.73.gate_proj", "model.language_model.layers.9.mlp.experts.74.gate_proj", "model.language_model.layers.9.mlp.experts.75.gate_proj", "model.language_model.layers.9.mlp.experts.76.gate_proj", "model.language_model.layers.9.mlp.experts.77.gate_proj", "model.language_model.layers.9.mlp.experts.78.gate_proj", "model.language_model.layers.9.mlp.experts.79.gate_proj", "model.language_model.layers.9.mlp.experts.80.gate_proj", "model.language_model.layers.9.mlp.experts.81.gate_proj", "model.language_model.layers.9.mlp.experts.82.gate_proj", "model.language_model.layers.9.mlp.experts.83.gate_proj", "model.language_model.layers.9.mlp.experts.84.gate_proj", "model.language_model.layers.9.mlp.experts.85.gate_proj", "model.language_model.layers.9.mlp.experts.86.gate_proj", "model.language_model.layers.9.mlp.experts.87.gate_proj", "model.language_model.layers.9.mlp.experts.88.gate_proj", "model.language_model.layers.9.mlp.experts.89.gate_proj", "model.language_model.layers.9.mlp.experts.90.gate_proj", "model.language_model.layers.9.mlp.experts.91.gate_proj", "model.language_model.layers.9.mlp.experts.92.gate_proj", "model.language_model.layers.9.mlp.experts.93.gate_proj", "model.language_model.layers.9.mlp.experts.94.gate_proj", "model.language_model.layers.9.mlp.experts.95.gate_proj", "model.language_model.layers.9.mlp.experts.96.gate_proj", "model.language_model.layers.9.mlp.experts.97.gate_proj", "model.language_model.layers.9.mlp.experts.98.gate_proj", "model.language_model.layers.9.mlp.experts.99.gate_proj", "model.language_model.layers.9.mlp.experts.100.gate_proj", "model.language_model.layers.9.mlp.experts.101.gate_proj", "model.language_model.layers.9.mlp.experts.102.gate_proj", "model.language_model.layers.9.mlp.experts.103.gate_proj", "model.language_model.layers.9.mlp.experts.104.gate_proj", "model.language_model.layers.9.mlp.experts.105.gate_proj", "model.language_model.layers.9.mlp.experts.106.gate_proj", "model.language_model.layers.9.mlp.experts.107.gate_proj", "model.language_model.layers.9.mlp.experts.108.gate_proj", "model.language_model.layers.9.mlp.experts.109.gate_proj", "model.language_model.layers.9.mlp.experts.110.gate_proj", "model.language_model.layers.9.mlp.experts.111.gate_proj", "model.language_model.layers.9.mlp.experts.112.gate_proj", "model.language_model.layers.9.mlp.experts.113.gate_proj", "model.language_model.layers.9.mlp.experts.114.gate_proj", "model.language_model.layers.9.mlp.experts.115.gate_proj", "model.language_model.layers.9.mlp.experts.116.gate_proj", "model.language_model.layers.9.mlp.experts.117.gate_proj", "model.language_model.layers.9.mlp.experts.118.gate_proj", "model.language_model.layers.9.mlp.experts.119.gate_proj", "model.language_model.layers.9.mlp.experts.120.gate_proj", "model.language_model.layers.9.mlp.experts.121.gate_proj", "model.language_model.layers.9.mlp.experts.122.gate_proj", "model.language_model.layers.9.mlp.experts.123.gate_proj", "model.language_model.layers.9.mlp.experts.124.gate_proj", "model.language_model.layers.9.mlp.experts.125.gate_proj", "model.language_model.layers.9.mlp.experts.126.gate_proj", "model.language_model.layers.9.mlp.experts.127.gate_proj", "model.language_model.layers.9.mlp.experts.0.up_proj", "model.language_model.layers.9.mlp.experts.1.up_proj", "model.language_model.layers.9.mlp.experts.2.up_proj", "model.language_model.layers.9.mlp.experts.3.up_proj", "model.language_model.layers.9.mlp.experts.4.up_proj", "model.language_model.layers.9.mlp.experts.5.up_proj", "model.language_model.layers.9.mlp.experts.6.up_proj", "model.language_model.layers.9.mlp.experts.7.up_proj", "model.language_model.layers.9.mlp.experts.8.up_proj", "model.language_model.layers.9.mlp.experts.9.up_proj", "model.language_model.layers.9.mlp.experts.10.up_proj", "model.language_model.layers.9.mlp.experts.11.up_proj", "model.language_model.layers.9.mlp.experts.12.up_proj", "model.language_model.layers.9.mlp.experts.13.up_proj", "model.language_model.layers.9.mlp.experts.14.up_proj", "model.language_model.layers.9.mlp.experts.15.up_proj", "model.language_model.layers.9.mlp.experts.16.up_proj", "model.language_model.layers.9.mlp.experts.17.up_proj", "model.language_model.layers.9.mlp.experts.18.up_proj", "model.language_model.layers.9.mlp.experts.19.up_proj", "model.language_model.layers.9.mlp.experts.20.up_proj", "model.language_model.layers.9.mlp.experts.21.up_proj", "model.language_model.layers.9.mlp.experts.22.up_proj", "model.language_model.layers.9.mlp.experts.23.up_proj", "model.language_model.layers.9.mlp.experts.24.up_proj", "model.language_model.layers.9.mlp.experts.25.up_proj", "model.language_model.layers.9.mlp.experts.26.up_proj", "model.language_model.layers.9.mlp.experts.27.up_proj", "model.language_model.layers.9.mlp.experts.28.up_proj", "model.language_model.layers.9.mlp.experts.29.up_proj", "model.language_model.layers.9.mlp.experts.30.up_proj", "model.language_model.layers.9.mlp.experts.31.up_proj", "model.language_model.layers.9.mlp.experts.32.up_proj", "model.language_model.layers.9.mlp.experts.33.up_proj", "model.language_model.layers.9.mlp.experts.34.up_proj", "model.language_model.layers.9.mlp.experts.35.up_proj", "model.language_model.layers.9.mlp.experts.36.up_proj", "model.language_model.layers.9.mlp.experts.37.up_proj", "model.language_model.layers.9.mlp.experts.38.up_proj", "model.language_model.layers.9.mlp.experts.39.up_proj", "model.language_model.layers.9.mlp.experts.40.up_proj", "model.language_model.layers.9.mlp.experts.41.up_proj", "model.language_model.layers.9.mlp.experts.42.up_proj", "model.language_model.layers.9.mlp.experts.43.up_proj", "model.language_model.layers.9.mlp.experts.44.up_proj", "model.language_model.layers.9.mlp.experts.45.up_proj", "model.language_model.layers.9.mlp.experts.46.up_proj", "model.language_model.layers.9.mlp.experts.47.up_proj", "model.language_model.layers.9.mlp.experts.48.up_proj", "model.language_model.layers.9.mlp.experts.49.up_proj", "model.language_model.layers.9.mlp.experts.50.up_proj", "model.language_model.layers.9.mlp.experts.51.up_proj", "model.language_model.layers.9.mlp.experts.52.up_proj", "model.language_model.layers.9.mlp.experts.53.up_proj", "model.language_model.layers.9.mlp.experts.54.up_proj", "model.language_model.layers.9.mlp.experts.55.up_proj", "model.language_model.layers.9.mlp.experts.56.up_proj", "model.language_model.layers.9.mlp.experts.57.up_proj", "model.language_model.layers.9.mlp.experts.58.up_proj", "model.language_model.layers.9.mlp.experts.59.up_proj", "model.language_model.layers.9.mlp.experts.60.up_proj", "model.language_model.layers.9.mlp.experts.61.up_proj", "model.language_model.layers.9.mlp.experts.62.up_proj", "model.language_model.layers.9.mlp.experts.63.up_proj", "model.language_model.layers.9.mlp.experts.64.up_proj", "model.language_model.layers.9.mlp.experts.65.up_proj", "model.language_model.layers.9.mlp.experts.66.up_proj", "model.language_model.layers.9.mlp.experts.67.up_proj", "model.language_model.layers.9.mlp.experts.68.up_proj", "model.language_model.layers.9.mlp.experts.69.up_proj", "model.language_model.layers.9.mlp.experts.70.up_proj", "model.language_model.layers.9.mlp.experts.71.up_proj", "model.language_model.layers.9.mlp.experts.72.up_proj", "model.language_model.layers.9.mlp.experts.73.up_proj", "model.language_model.layers.9.mlp.experts.74.up_proj", "model.language_model.layers.9.mlp.experts.75.up_proj", "model.language_model.layers.9.mlp.experts.76.up_proj", "model.language_model.layers.9.mlp.experts.77.up_proj", "model.language_model.layers.9.mlp.experts.78.up_proj", "model.language_model.layers.9.mlp.experts.79.up_proj", "model.language_model.layers.9.mlp.experts.80.up_proj", "model.language_model.layers.9.mlp.experts.81.up_proj", "model.language_model.layers.9.mlp.experts.82.up_proj", "model.language_model.layers.9.mlp.experts.83.up_proj", "model.language_model.layers.9.mlp.experts.84.up_proj", "model.language_model.layers.9.mlp.experts.85.up_proj", "model.language_model.layers.9.mlp.experts.86.up_proj", "model.language_model.layers.9.mlp.experts.87.up_proj", "model.language_model.layers.9.mlp.experts.88.up_proj", "model.language_model.layers.9.mlp.experts.89.up_proj", "model.language_model.layers.9.mlp.experts.90.up_proj", "model.language_model.layers.9.mlp.experts.91.up_proj", "model.language_model.layers.9.mlp.experts.92.up_proj", "model.language_model.layers.9.mlp.experts.93.up_proj", "model.language_model.layers.9.mlp.experts.94.up_proj", "model.language_model.layers.9.mlp.experts.95.up_proj", "model.language_model.layers.9.mlp.experts.96.up_proj", "model.language_model.layers.9.mlp.experts.97.up_proj", "model.language_model.layers.9.mlp.experts.98.up_proj", "model.language_model.layers.9.mlp.experts.99.up_proj", "model.language_model.layers.9.mlp.experts.100.up_proj", "model.language_model.layers.9.mlp.experts.101.up_proj", "model.language_model.layers.9.mlp.experts.102.up_proj", "model.language_model.layers.9.mlp.experts.103.up_proj", "model.language_model.layers.9.mlp.experts.104.up_proj", "model.language_model.layers.9.mlp.experts.105.up_proj", "model.language_model.layers.9.mlp.experts.106.up_proj", "model.language_model.layers.9.mlp.experts.107.up_proj", "model.language_model.layers.9.mlp.experts.108.up_proj", "model.language_model.layers.9.mlp.experts.109.up_proj", "model.language_model.layers.9.mlp.experts.110.up_proj", "model.language_model.layers.9.mlp.experts.111.up_proj", "model.language_model.layers.9.mlp.experts.112.up_proj", "model.language_model.layers.9.mlp.experts.113.up_proj", "model.language_model.layers.9.mlp.experts.114.up_proj", "model.language_model.layers.9.mlp.experts.115.up_proj", "model.language_model.layers.9.mlp.experts.116.up_proj", "model.language_model.layers.9.mlp.experts.117.up_proj", "model.language_model.layers.9.mlp.experts.118.up_proj", "model.language_model.layers.9.mlp.experts.119.up_proj", "model.language_model.layers.9.mlp.experts.120.up_proj", "model.language_model.layers.9.mlp.experts.121.up_proj", "model.language_model.layers.9.mlp.experts.122.up_proj", "model.language_model.layers.9.mlp.experts.123.up_proj", "model.language_model.layers.9.mlp.experts.124.up_proj", "model.language_model.layers.9.mlp.experts.125.up_proj", "model.language_model.layers.9.mlp.experts.126.up_proj", "model.language_model.layers.9.mlp.experts.127.up_proj", "model.language_model.layers.9.mlp.experts.0.down_proj", "model.language_model.layers.9.mlp.experts.1.down_proj", "model.language_model.layers.9.mlp.experts.2.down_proj", "model.language_model.layers.9.mlp.experts.3.down_proj", "model.language_model.layers.9.mlp.experts.4.down_proj", "model.language_model.layers.9.mlp.experts.5.down_proj", "model.language_model.layers.9.mlp.experts.6.down_proj", "model.language_model.layers.9.mlp.experts.7.down_proj", "model.language_model.layers.9.mlp.experts.8.down_proj", "model.language_model.layers.9.mlp.experts.9.down_proj", "model.language_model.layers.9.mlp.experts.10.down_proj", "model.language_model.layers.9.mlp.experts.11.down_proj", "model.language_model.layers.9.mlp.experts.12.down_proj", "model.language_model.layers.9.mlp.experts.13.down_proj", "model.language_model.layers.9.mlp.experts.14.down_proj", "model.language_model.layers.9.mlp.experts.15.down_proj", "model.language_model.layers.9.mlp.experts.16.down_proj", "model.language_model.layers.9.mlp.experts.17.down_proj", "model.language_model.layers.9.mlp.experts.18.down_proj", "model.language_model.layers.9.mlp.experts.19.down_proj", "model.language_model.layers.9.mlp.experts.20.down_proj", "model.language_model.layers.9.mlp.experts.21.down_proj", "model.language_model.layers.9.mlp.experts.22.down_proj", "model.language_model.layers.9.mlp.experts.23.down_proj", "model.language_model.layers.9.mlp.experts.24.down_proj", "model.language_model.layers.9.mlp.experts.25.down_proj", "model.language_model.layers.9.mlp.experts.26.down_proj", "model.language_model.layers.9.mlp.experts.27.down_proj", "model.language_model.layers.9.mlp.experts.28.down_proj", "model.language_model.layers.9.mlp.experts.29.down_proj", "model.language_model.layers.9.mlp.experts.30.down_proj", "model.language_model.layers.9.mlp.experts.31.down_proj", "model.language_model.layers.9.mlp.experts.32.down_proj", "model.language_model.layers.9.mlp.experts.33.down_proj", "model.language_model.layers.9.mlp.experts.34.down_proj", "model.language_model.layers.9.mlp.experts.35.down_proj", "model.language_model.layers.9.mlp.experts.36.down_proj", "model.language_model.layers.9.mlp.experts.37.down_proj", "model.language_model.layers.9.mlp.experts.38.down_proj", "model.language_model.layers.9.mlp.experts.39.down_proj", "model.language_model.layers.9.mlp.experts.40.down_proj", "model.language_model.layers.9.mlp.experts.41.down_proj", "model.language_model.layers.9.mlp.experts.42.down_proj", "model.language_model.layers.9.mlp.experts.43.down_proj", "model.language_model.layers.9.mlp.experts.44.down_proj", "model.language_model.layers.9.mlp.experts.45.down_proj", "model.language_model.layers.9.mlp.experts.46.down_proj", "model.language_model.layers.9.mlp.experts.47.down_proj", "model.language_model.layers.9.mlp.experts.48.down_proj", "model.language_model.layers.9.mlp.experts.49.down_proj", "model.language_model.layers.9.mlp.experts.50.down_proj", "model.language_model.layers.9.mlp.experts.51.down_proj", "model.language_model.layers.9.mlp.experts.52.down_proj", "model.language_model.layers.9.mlp.experts.53.down_proj", "model.language_model.layers.9.mlp.experts.54.down_proj", "model.language_model.layers.9.mlp.experts.55.down_proj", "model.language_model.layers.9.mlp.experts.56.down_proj", "model.language_model.layers.9.mlp.experts.57.down_proj", "model.language_model.layers.9.mlp.experts.58.down_proj", "model.language_model.layers.9.mlp.experts.59.down_proj", "model.language_model.layers.9.mlp.experts.60.down_proj", "model.language_model.layers.9.mlp.experts.61.down_proj", "model.language_model.layers.9.mlp.experts.62.down_proj", "model.language_model.layers.9.mlp.experts.63.down_proj", "model.language_model.layers.9.mlp.experts.64.down_proj", "model.language_model.layers.9.mlp.experts.65.down_proj", "model.language_model.layers.9.mlp.experts.66.down_proj", "model.language_model.layers.9.mlp.experts.67.down_proj", "model.language_model.layers.9.mlp.experts.68.down_proj", "model.language_model.layers.9.mlp.experts.69.down_proj", "model.language_model.layers.9.mlp.experts.70.down_proj", "model.language_model.layers.9.mlp.experts.71.down_proj", "model.language_model.layers.9.mlp.experts.72.down_proj", "model.language_model.layers.9.mlp.experts.73.down_proj", "model.language_model.layers.9.mlp.experts.74.down_proj", "model.language_model.layers.9.mlp.experts.75.down_proj", "model.language_model.layers.9.mlp.experts.76.down_proj", "model.language_model.layers.9.mlp.experts.77.down_proj", "model.language_model.layers.9.mlp.experts.78.down_proj", "model.language_model.layers.9.mlp.experts.79.down_proj", "model.language_model.layers.9.mlp.experts.80.down_proj", "model.language_model.layers.9.mlp.experts.81.down_proj", "model.language_model.layers.9.mlp.experts.82.down_proj", "model.language_model.layers.9.mlp.experts.83.down_proj", "model.language_model.layers.9.mlp.experts.84.down_proj", "model.language_model.layers.9.mlp.experts.85.down_proj", "model.language_model.layers.9.mlp.experts.86.down_proj", "model.language_model.layers.9.mlp.experts.87.down_proj", "model.language_model.layers.9.mlp.experts.88.down_proj", "model.language_model.layers.9.mlp.experts.89.down_proj", "model.language_model.layers.9.mlp.experts.90.down_proj", "model.language_model.layers.9.mlp.experts.91.down_proj", "model.language_model.layers.9.mlp.experts.92.down_proj", "model.language_model.layers.9.mlp.experts.93.down_proj", "model.language_model.layers.9.mlp.experts.94.down_proj", "model.language_model.layers.9.mlp.experts.95.down_proj", "model.language_model.layers.9.mlp.experts.96.down_proj", "model.language_model.layers.9.mlp.experts.97.down_proj", "model.language_model.layers.9.mlp.experts.98.down_proj", "model.language_model.layers.9.mlp.experts.99.down_proj", "model.language_model.layers.9.mlp.experts.100.down_proj", "model.language_model.layers.9.mlp.experts.101.down_proj", "model.language_model.layers.9.mlp.experts.102.down_proj", "model.language_model.layers.9.mlp.experts.103.down_proj", "model.language_model.layers.9.mlp.experts.104.down_proj", "model.language_model.layers.9.mlp.experts.105.down_proj", "model.language_model.layers.9.mlp.experts.106.down_proj", "model.language_model.layers.9.mlp.experts.107.down_proj", "model.language_model.layers.9.mlp.experts.108.down_proj", "model.language_model.layers.9.mlp.experts.109.down_proj", "model.language_model.layers.9.mlp.experts.110.down_proj", "model.language_model.layers.9.mlp.experts.111.down_proj", "model.language_model.layers.9.mlp.experts.112.down_proj", "model.language_model.layers.9.mlp.experts.113.down_proj", "model.language_model.layers.9.mlp.experts.114.down_proj", "model.language_model.layers.9.mlp.experts.115.down_proj", "model.language_model.layers.9.mlp.experts.116.down_proj", "model.language_model.layers.9.mlp.experts.117.down_proj", "model.language_model.layers.9.mlp.experts.118.down_proj", "model.language_model.layers.9.mlp.experts.119.down_proj", "model.language_model.layers.9.mlp.experts.120.down_proj", "model.language_model.layers.9.mlp.experts.121.down_proj", "model.language_model.layers.9.mlp.experts.122.down_proj", "model.language_model.layers.9.mlp.experts.123.down_proj", "model.language_model.layers.9.mlp.experts.124.down_proj", "model.language_model.layers.9.mlp.experts.125.down_proj", "model.language_model.layers.9.mlp.experts.126.down_proj", "model.language_model.layers.9.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.0027041195891797515, "dbits": 603979776 } ] }, { "idx": 20, "layers": [ "model.language_model.layers.10.self_attn.q_proj", "model.language_model.layers.10.self_attn.k_proj", "model.language_model.layers.10.self_attn.v_proj", "model.language_model.layers.10.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0011825214140117196, "dbits": 18874368 } ] }, { "idx": 21, "layers": [ "model.language_model.layers.10.mlp.experts.0.gate_proj", "model.language_model.layers.10.mlp.experts.1.gate_proj", "model.language_model.layers.10.mlp.experts.2.gate_proj", "model.language_model.layers.10.mlp.experts.3.gate_proj", "model.language_model.layers.10.mlp.experts.4.gate_proj", "model.language_model.layers.10.mlp.experts.5.gate_proj", "model.language_model.layers.10.mlp.experts.6.gate_proj", "model.language_model.layers.10.mlp.experts.7.gate_proj", "model.language_model.layers.10.mlp.experts.8.gate_proj", "model.language_model.layers.10.mlp.experts.9.gate_proj", "model.language_model.layers.10.mlp.experts.10.gate_proj", "model.language_model.layers.10.mlp.experts.11.gate_proj", "model.language_model.layers.10.mlp.experts.12.gate_proj", "model.language_model.layers.10.mlp.experts.13.gate_proj", "model.language_model.layers.10.mlp.experts.14.gate_proj", "model.language_model.layers.10.mlp.experts.15.gate_proj", "model.language_model.layers.10.mlp.experts.16.gate_proj", "model.language_model.layers.10.mlp.experts.17.gate_proj", "model.language_model.layers.10.mlp.experts.18.gate_proj", "model.language_model.layers.10.mlp.experts.19.gate_proj", "model.language_model.layers.10.mlp.experts.20.gate_proj", "model.language_model.layers.10.mlp.experts.21.gate_proj", "model.language_model.layers.10.mlp.experts.22.gate_proj", "model.language_model.layers.10.mlp.experts.23.gate_proj", "model.language_model.layers.10.mlp.experts.24.gate_proj", "model.language_model.layers.10.mlp.experts.25.gate_proj", "model.language_model.layers.10.mlp.experts.26.gate_proj", "model.language_model.layers.10.mlp.experts.27.gate_proj", "model.language_model.layers.10.mlp.experts.28.gate_proj", "model.language_model.layers.10.mlp.experts.29.gate_proj", "model.language_model.layers.10.mlp.experts.30.gate_proj", "model.language_model.layers.10.mlp.experts.31.gate_proj", "model.language_model.layers.10.mlp.experts.32.gate_proj", "model.language_model.layers.10.mlp.experts.33.gate_proj", "model.language_model.layers.10.mlp.experts.34.gate_proj", "model.language_model.layers.10.mlp.experts.35.gate_proj", "model.language_model.layers.10.mlp.experts.36.gate_proj", "model.language_model.layers.10.mlp.experts.37.gate_proj", "model.language_model.layers.10.mlp.experts.38.gate_proj", "model.language_model.layers.10.mlp.experts.39.gate_proj", "model.language_model.layers.10.mlp.experts.40.gate_proj", "model.language_model.layers.10.mlp.experts.41.gate_proj", "model.language_model.layers.10.mlp.experts.42.gate_proj", "model.language_model.layers.10.mlp.experts.43.gate_proj", "model.language_model.layers.10.mlp.experts.44.gate_proj", "model.language_model.layers.10.mlp.experts.45.gate_proj", "model.language_model.layers.10.mlp.experts.46.gate_proj", "model.language_model.layers.10.mlp.experts.47.gate_proj", "model.language_model.layers.10.mlp.experts.48.gate_proj", "model.language_model.layers.10.mlp.experts.49.gate_proj", "model.language_model.layers.10.mlp.experts.50.gate_proj", "model.language_model.layers.10.mlp.experts.51.gate_proj", "model.language_model.layers.10.mlp.experts.52.gate_proj", "model.language_model.layers.10.mlp.experts.53.gate_proj", "model.language_model.layers.10.mlp.experts.54.gate_proj", "model.language_model.layers.10.mlp.experts.55.gate_proj", "model.language_model.layers.10.mlp.experts.56.gate_proj", "model.language_model.layers.10.mlp.experts.57.gate_proj", "model.language_model.layers.10.mlp.experts.58.gate_proj", "model.language_model.layers.10.mlp.experts.59.gate_proj", "model.language_model.layers.10.mlp.experts.60.gate_proj", "model.language_model.layers.10.mlp.experts.61.gate_proj", "model.language_model.layers.10.mlp.experts.62.gate_proj", "model.language_model.layers.10.mlp.experts.63.gate_proj", "model.language_model.layers.10.mlp.experts.64.gate_proj", "model.language_model.layers.10.mlp.experts.65.gate_proj", "model.language_model.layers.10.mlp.experts.66.gate_proj", "model.language_model.layers.10.mlp.experts.67.gate_proj", "model.language_model.layers.10.mlp.experts.68.gate_proj", "model.language_model.layers.10.mlp.experts.69.gate_proj", "model.language_model.layers.10.mlp.experts.70.gate_proj", "model.language_model.layers.10.mlp.experts.71.gate_proj", "model.language_model.layers.10.mlp.experts.72.gate_proj", "model.language_model.layers.10.mlp.experts.73.gate_proj", "model.language_model.layers.10.mlp.experts.74.gate_proj", "model.language_model.layers.10.mlp.experts.75.gate_proj", "model.language_model.layers.10.mlp.experts.76.gate_proj", "model.language_model.layers.10.mlp.experts.77.gate_proj", "model.language_model.layers.10.mlp.experts.78.gate_proj", "model.language_model.layers.10.mlp.experts.79.gate_proj", "model.language_model.layers.10.mlp.experts.80.gate_proj", "model.language_model.layers.10.mlp.experts.81.gate_proj", "model.language_model.layers.10.mlp.experts.82.gate_proj", "model.language_model.layers.10.mlp.experts.83.gate_proj", "model.language_model.layers.10.mlp.experts.84.gate_proj", "model.language_model.layers.10.mlp.experts.85.gate_proj", "model.language_model.layers.10.mlp.experts.86.gate_proj", "model.language_model.layers.10.mlp.experts.87.gate_proj", "model.language_model.layers.10.mlp.experts.88.gate_proj", "model.language_model.layers.10.mlp.experts.89.gate_proj", "model.language_model.layers.10.mlp.experts.90.gate_proj", "model.language_model.layers.10.mlp.experts.91.gate_proj", "model.language_model.layers.10.mlp.experts.92.gate_proj", "model.language_model.layers.10.mlp.experts.93.gate_proj", "model.language_model.layers.10.mlp.experts.94.gate_proj", "model.language_model.layers.10.mlp.experts.95.gate_proj", "model.language_model.layers.10.mlp.experts.96.gate_proj", "model.language_model.layers.10.mlp.experts.97.gate_proj", "model.language_model.layers.10.mlp.experts.98.gate_proj", "model.language_model.layers.10.mlp.experts.99.gate_proj", "model.language_model.layers.10.mlp.experts.100.gate_proj", "model.language_model.layers.10.mlp.experts.101.gate_proj", "model.language_model.layers.10.mlp.experts.102.gate_proj", "model.language_model.layers.10.mlp.experts.103.gate_proj", "model.language_model.layers.10.mlp.experts.104.gate_proj", "model.language_model.layers.10.mlp.experts.105.gate_proj", "model.language_model.layers.10.mlp.experts.106.gate_proj", "model.language_model.layers.10.mlp.experts.107.gate_proj", "model.language_model.layers.10.mlp.experts.108.gate_proj", "model.language_model.layers.10.mlp.experts.109.gate_proj", "model.language_model.layers.10.mlp.experts.110.gate_proj", "model.language_model.layers.10.mlp.experts.111.gate_proj", "model.language_model.layers.10.mlp.experts.112.gate_proj", "model.language_model.layers.10.mlp.experts.113.gate_proj", "model.language_model.layers.10.mlp.experts.114.gate_proj", "model.language_model.layers.10.mlp.experts.115.gate_proj", "model.language_model.layers.10.mlp.experts.116.gate_proj", "model.language_model.layers.10.mlp.experts.117.gate_proj", "model.language_model.layers.10.mlp.experts.118.gate_proj", "model.language_model.layers.10.mlp.experts.119.gate_proj", "model.language_model.layers.10.mlp.experts.120.gate_proj", "model.language_model.layers.10.mlp.experts.121.gate_proj", "model.language_model.layers.10.mlp.experts.122.gate_proj", "model.language_model.layers.10.mlp.experts.123.gate_proj", "model.language_model.layers.10.mlp.experts.124.gate_proj", "model.language_model.layers.10.mlp.experts.125.gate_proj", "model.language_model.layers.10.mlp.experts.126.gate_proj", "model.language_model.layers.10.mlp.experts.127.gate_proj", "model.language_model.layers.10.mlp.experts.0.up_proj", "model.language_model.layers.10.mlp.experts.1.up_proj", "model.language_model.layers.10.mlp.experts.2.up_proj", "model.language_model.layers.10.mlp.experts.3.up_proj", "model.language_model.layers.10.mlp.experts.4.up_proj", "model.language_model.layers.10.mlp.experts.5.up_proj", "model.language_model.layers.10.mlp.experts.6.up_proj", "model.language_model.layers.10.mlp.experts.7.up_proj", "model.language_model.layers.10.mlp.experts.8.up_proj", "model.language_model.layers.10.mlp.experts.9.up_proj", "model.language_model.layers.10.mlp.experts.10.up_proj", "model.language_model.layers.10.mlp.experts.11.up_proj", "model.language_model.layers.10.mlp.experts.12.up_proj", "model.language_model.layers.10.mlp.experts.13.up_proj", "model.language_model.layers.10.mlp.experts.14.up_proj", "model.language_model.layers.10.mlp.experts.15.up_proj", "model.language_model.layers.10.mlp.experts.16.up_proj", "model.language_model.layers.10.mlp.experts.17.up_proj", "model.language_model.layers.10.mlp.experts.18.up_proj", "model.language_model.layers.10.mlp.experts.19.up_proj", "model.language_model.layers.10.mlp.experts.20.up_proj", "model.language_model.layers.10.mlp.experts.21.up_proj", "model.language_model.layers.10.mlp.experts.22.up_proj", "model.language_model.layers.10.mlp.experts.23.up_proj", "model.language_model.layers.10.mlp.experts.24.up_proj", "model.language_model.layers.10.mlp.experts.25.up_proj", "model.language_model.layers.10.mlp.experts.26.up_proj", "model.language_model.layers.10.mlp.experts.27.up_proj", "model.language_model.layers.10.mlp.experts.28.up_proj", "model.language_model.layers.10.mlp.experts.29.up_proj", "model.language_model.layers.10.mlp.experts.30.up_proj", "model.language_model.layers.10.mlp.experts.31.up_proj", "model.language_model.layers.10.mlp.experts.32.up_proj", "model.language_model.layers.10.mlp.experts.33.up_proj", "model.language_model.layers.10.mlp.experts.34.up_proj", "model.language_model.layers.10.mlp.experts.35.up_proj", "model.language_model.layers.10.mlp.experts.36.up_proj", "model.language_model.layers.10.mlp.experts.37.up_proj", "model.language_model.layers.10.mlp.experts.38.up_proj", "model.language_model.layers.10.mlp.experts.39.up_proj", "model.language_model.layers.10.mlp.experts.40.up_proj", "model.language_model.layers.10.mlp.experts.41.up_proj", "model.language_model.layers.10.mlp.experts.42.up_proj", "model.language_model.layers.10.mlp.experts.43.up_proj", "model.language_model.layers.10.mlp.experts.44.up_proj", "model.language_model.layers.10.mlp.experts.45.up_proj", "model.language_model.layers.10.mlp.experts.46.up_proj", "model.language_model.layers.10.mlp.experts.47.up_proj", "model.language_model.layers.10.mlp.experts.48.up_proj", "model.language_model.layers.10.mlp.experts.49.up_proj", "model.language_model.layers.10.mlp.experts.50.up_proj", "model.language_model.layers.10.mlp.experts.51.up_proj", "model.language_model.layers.10.mlp.experts.52.up_proj", "model.language_model.layers.10.mlp.experts.53.up_proj", "model.language_model.layers.10.mlp.experts.54.up_proj", "model.language_model.layers.10.mlp.experts.55.up_proj", "model.language_model.layers.10.mlp.experts.56.up_proj", "model.language_model.layers.10.mlp.experts.57.up_proj", "model.language_model.layers.10.mlp.experts.58.up_proj", "model.language_model.layers.10.mlp.experts.59.up_proj", "model.language_model.layers.10.mlp.experts.60.up_proj", "model.language_model.layers.10.mlp.experts.61.up_proj", "model.language_model.layers.10.mlp.experts.62.up_proj", "model.language_model.layers.10.mlp.experts.63.up_proj", "model.language_model.layers.10.mlp.experts.64.up_proj", "model.language_model.layers.10.mlp.experts.65.up_proj", "model.language_model.layers.10.mlp.experts.66.up_proj", "model.language_model.layers.10.mlp.experts.67.up_proj", "model.language_model.layers.10.mlp.experts.68.up_proj", "model.language_model.layers.10.mlp.experts.69.up_proj", "model.language_model.layers.10.mlp.experts.70.up_proj", "model.language_model.layers.10.mlp.experts.71.up_proj", "model.language_model.layers.10.mlp.experts.72.up_proj", "model.language_model.layers.10.mlp.experts.73.up_proj", "model.language_model.layers.10.mlp.experts.74.up_proj", "model.language_model.layers.10.mlp.experts.75.up_proj", "model.language_model.layers.10.mlp.experts.76.up_proj", "model.language_model.layers.10.mlp.experts.77.up_proj", "model.language_model.layers.10.mlp.experts.78.up_proj", "model.language_model.layers.10.mlp.experts.79.up_proj", "model.language_model.layers.10.mlp.experts.80.up_proj", "model.language_model.layers.10.mlp.experts.81.up_proj", "model.language_model.layers.10.mlp.experts.82.up_proj", "model.language_model.layers.10.mlp.experts.83.up_proj", "model.language_model.layers.10.mlp.experts.84.up_proj", "model.language_model.layers.10.mlp.experts.85.up_proj", "model.language_model.layers.10.mlp.experts.86.up_proj", "model.language_model.layers.10.mlp.experts.87.up_proj", "model.language_model.layers.10.mlp.experts.88.up_proj", "model.language_model.layers.10.mlp.experts.89.up_proj", "model.language_model.layers.10.mlp.experts.90.up_proj", "model.language_model.layers.10.mlp.experts.91.up_proj", "model.language_model.layers.10.mlp.experts.92.up_proj", "model.language_model.layers.10.mlp.experts.93.up_proj", "model.language_model.layers.10.mlp.experts.94.up_proj", "model.language_model.layers.10.mlp.experts.95.up_proj", "model.language_model.layers.10.mlp.experts.96.up_proj", "model.language_model.layers.10.mlp.experts.97.up_proj", "model.language_model.layers.10.mlp.experts.98.up_proj", "model.language_model.layers.10.mlp.experts.99.up_proj", "model.language_model.layers.10.mlp.experts.100.up_proj", "model.language_model.layers.10.mlp.experts.101.up_proj", "model.language_model.layers.10.mlp.experts.102.up_proj", "model.language_model.layers.10.mlp.experts.103.up_proj", "model.language_model.layers.10.mlp.experts.104.up_proj", "model.language_model.layers.10.mlp.experts.105.up_proj", "model.language_model.layers.10.mlp.experts.106.up_proj", "model.language_model.layers.10.mlp.experts.107.up_proj", "model.language_model.layers.10.mlp.experts.108.up_proj", "model.language_model.layers.10.mlp.experts.109.up_proj", "model.language_model.layers.10.mlp.experts.110.up_proj", "model.language_model.layers.10.mlp.experts.111.up_proj", "model.language_model.layers.10.mlp.experts.112.up_proj", "model.language_model.layers.10.mlp.experts.113.up_proj", "model.language_model.layers.10.mlp.experts.114.up_proj", "model.language_model.layers.10.mlp.experts.115.up_proj", "model.language_model.layers.10.mlp.experts.116.up_proj", "model.language_model.layers.10.mlp.experts.117.up_proj", "model.language_model.layers.10.mlp.experts.118.up_proj", "model.language_model.layers.10.mlp.experts.119.up_proj", "model.language_model.layers.10.mlp.experts.120.up_proj", "model.language_model.layers.10.mlp.experts.121.up_proj", "model.language_model.layers.10.mlp.experts.122.up_proj", "model.language_model.layers.10.mlp.experts.123.up_proj", "model.language_model.layers.10.mlp.experts.124.up_proj", "model.language_model.layers.10.mlp.experts.125.up_proj", "model.language_model.layers.10.mlp.experts.126.up_proj", "model.language_model.layers.10.mlp.experts.127.up_proj", "model.language_model.layers.10.mlp.experts.0.down_proj", "model.language_model.layers.10.mlp.experts.1.down_proj", "model.language_model.layers.10.mlp.experts.2.down_proj", "model.language_model.layers.10.mlp.experts.3.down_proj", "model.language_model.layers.10.mlp.experts.4.down_proj", "model.language_model.layers.10.mlp.experts.5.down_proj", "model.language_model.layers.10.mlp.experts.6.down_proj", "model.language_model.layers.10.mlp.experts.7.down_proj", "model.language_model.layers.10.mlp.experts.8.down_proj", "model.language_model.layers.10.mlp.experts.9.down_proj", "model.language_model.layers.10.mlp.experts.10.down_proj", "model.language_model.layers.10.mlp.experts.11.down_proj", "model.language_model.layers.10.mlp.experts.12.down_proj", "model.language_model.layers.10.mlp.experts.13.down_proj", "model.language_model.layers.10.mlp.experts.14.down_proj", "model.language_model.layers.10.mlp.experts.15.down_proj", "model.language_model.layers.10.mlp.experts.16.down_proj", "model.language_model.layers.10.mlp.experts.17.down_proj", "model.language_model.layers.10.mlp.experts.18.down_proj", "model.language_model.layers.10.mlp.experts.19.down_proj", "model.language_model.layers.10.mlp.experts.20.down_proj", "model.language_model.layers.10.mlp.experts.21.down_proj", "model.language_model.layers.10.mlp.experts.22.down_proj", "model.language_model.layers.10.mlp.experts.23.down_proj", "model.language_model.layers.10.mlp.experts.24.down_proj", "model.language_model.layers.10.mlp.experts.25.down_proj", "model.language_model.layers.10.mlp.experts.26.down_proj", "model.language_model.layers.10.mlp.experts.27.down_proj", "model.language_model.layers.10.mlp.experts.28.down_proj", "model.language_model.layers.10.mlp.experts.29.down_proj", "model.language_model.layers.10.mlp.experts.30.down_proj", "model.language_model.layers.10.mlp.experts.31.down_proj", "model.language_model.layers.10.mlp.experts.32.down_proj", "model.language_model.layers.10.mlp.experts.33.down_proj", "model.language_model.layers.10.mlp.experts.34.down_proj", "model.language_model.layers.10.mlp.experts.35.down_proj", "model.language_model.layers.10.mlp.experts.36.down_proj", "model.language_model.layers.10.mlp.experts.37.down_proj", "model.language_model.layers.10.mlp.experts.38.down_proj", "model.language_model.layers.10.mlp.experts.39.down_proj", "model.language_model.layers.10.mlp.experts.40.down_proj", "model.language_model.layers.10.mlp.experts.41.down_proj", "model.language_model.layers.10.mlp.experts.42.down_proj", "model.language_model.layers.10.mlp.experts.43.down_proj", "model.language_model.layers.10.mlp.experts.44.down_proj", "model.language_model.layers.10.mlp.experts.45.down_proj", "model.language_model.layers.10.mlp.experts.46.down_proj", "model.language_model.layers.10.mlp.experts.47.down_proj", "model.language_model.layers.10.mlp.experts.48.down_proj", "model.language_model.layers.10.mlp.experts.49.down_proj", "model.language_model.layers.10.mlp.experts.50.down_proj", "model.language_model.layers.10.mlp.experts.51.down_proj", "model.language_model.layers.10.mlp.experts.52.down_proj", "model.language_model.layers.10.mlp.experts.53.down_proj", "model.language_model.layers.10.mlp.experts.54.down_proj", "model.language_model.layers.10.mlp.experts.55.down_proj", "model.language_model.layers.10.mlp.experts.56.down_proj", "model.language_model.layers.10.mlp.experts.57.down_proj", "model.language_model.layers.10.mlp.experts.58.down_proj", "model.language_model.layers.10.mlp.experts.59.down_proj", "model.language_model.layers.10.mlp.experts.60.down_proj", "model.language_model.layers.10.mlp.experts.61.down_proj", "model.language_model.layers.10.mlp.experts.62.down_proj", "model.language_model.layers.10.mlp.experts.63.down_proj", "model.language_model.layers.10.mlp.experts.64.down_proj", "model.language_model.layers.10.mlp.experts.65.down_proj", "model.language_model.layers.10.mlp.experts.66.down_proj", "model.language_model.layers.10.mlp.experts.67.down_proj", "model.language_model.layers.10.mlp.experts.68.down_proj", "model.language_model.layers.10.mlp.experts.69.down_proj", "model.language_model.layers.10.mlp.experts.70.down_proj", "model.language_model.layers.10.mlp.experts.71.down_proj", "model.language_model.layers.10.mlp.experts.72.down_proj", "model.language_model.layers.10.mlp.experts.73.down_proj", "model.language_model.layers.10.mlp.experts.74.down_proj", "model.language_model.layers.10.mlp.experts.75.down_proj", "model.language_model.layers.10.mlp.experts.76.down_proj", "model.language_model.layers.10.mlp.experts.77.down_proj", "model.language_model.layers.10.mlp.experts.78.down_proj", "model.language_model.layers.10.mlp.experts.79.down_proj", "model.language_model.layers.10.mlp.experts.80.down_proj", "model.language_model.layers.10.mlp.experts.81.down_proj", "model.language_model.layers.10.mlp.experts.82.down_proj", "model.language_model.layers.10.mlp.experts.83.down_proj", "model.language_model.layers.10.mlp.experts.84.down_proj", "model.language_model.layers.10.mlp.experts.85.down_proj", "model.language_model.layers.10.mlp.experts.86.down_proj", "model.language_model.layers.10.mlp.experts.87.down_proj", "model.language_model.layers.10.mlp.experts.88.down_proj", "model.language_model.layers.10.mlp.experts.89.down_proj", "model.language_model.layers.10.mlp.experts.90.down_proj", "model.language_model.layers.10.mlp.experts.91.down_proj", "model.language_model.layers.10.mlp.experts.92.down_proj", "model.language_model.layers.10.mlp.experts.93.down_proj", "model.language_model.layers.10.mlp.experts.94.down_proj", "model.language_model.layers.10.mlp.experts.95.down_proj", "model.language_model.layers.10.mlp.experts.96.down_proj", "model.language_model.layers.10.mlp.experts.97.down_proj", "model.language_model.layers.10.mlp.experts.98.down_proj", "model.language_model.layers.10.mlp.experts.99.down_proj", "model.language_model.layers.10.mlp.experts.100.down_proj", "model.language_model.layers.10.mlp.experts.101.down_proj", "model.language_model.layers.10.mlp.experts.102.down_proj", "model.language_model.layers.10.mlp.experts.103.down_proj", "model.language_model.layers.10.mlp.experts.104.down_proj", "model.language_model.layers.10.mlp.experts.105.down_proj", "model.language_model.layers.10.mlp.experts.106.down_proj", "model.language_model.layers.10.mlp.experts.107.down_proj", "model.language_model.layers.10.mlp.experts.108.down_proj", "model.language_model.layers.10.mlp.experts.109.down_proj", "model.language_model.layers.10.mlp.experts.110.down_proj", "model.language_model.layers.10.mlp.experts.111.down_proj", "model.language_model.layers.10.mlp.experts.112.down_proj", "model.language_model.layers.10.mlp.experts.113.down_proj", "model.language_model.layers.10.mlp.experts.114.down_proj", "model.language_model.layers.10.mlp.experts.115.down_proj", "model.language_model.layers.10.mlp.experts.116.down_proj", "model.language_model.layers.10.mlp.experts.117.down_proj", "model.language_model.layers.10.mlp.experts.118.down_proj", "model.language_model.layers.10.mlp.experts.119.down_proj", "model.language_model.layers.10.mlp.experts.120.down_proj", "model.language_model.layers.10.mlp.experts.121.down_proj", "model.language_model.layers.10.mlp.experts.122.down_proj", "model.language_model.layers.10.mlp.experts.123.down_proj", "model.language_model.layers.10.mlp.experts.124.down_proj", "model.language_model.layers.10.mlp.experts.125.down_proj", "model.language_model.layers.10.mlp.experts.126.down_proj", "model.language_model.layers.10.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.0015961226075887625, "dbits": 603979776 } ] }, { "idx": 22, "layers": [ "model.language_model.layers.11.self_attn.q_proj", "model.language_model.layers.11.self_attn.k_proj", "model.language_model.layers.11.self_attn.v_proj", "model.language_model.layers.11.self_attn.o_proj" ], "candidates": [ { "dkld": 0.0007645189762115479, "dbits": 18874368 } ] }, { "idx": 23, "layers": [ "model.language_model.layers.11.mlp.experts.0.gate_proj", "model.language_model.layers.11.mlp.experts.1.gate_proj", "model.language_model.layers.11.mlp.experts.2.gate_proj", "model.language_model.layers.11.mlp.experts.3.gate_proj", "model.language_model.layers.11.mlp.experts.4.gate_proj", "model.language_model.layers.11.mlp.experts.5.gate_proj", "model.language_model.layers.11.mlp.experts.6.gate_proj", "model.language_model.layers.11.mlp.experts.7.gate_proj", "model.language_model.layers.11.mlp.experts.8.gate_proj", "model.language_model.layers.11.mlp.experts.9.gate_proj", "model.language_model.layers.11.mlp.experts.10.gate_proj", "model.language_model.layers.11.mlp.experts.11.gate_proj", "model.language_model.layers.11.mlp.experts.12.gate_proj", "model.language_model.layers.11.mlp.experts.13.gate_proj", "model.language_model.layers.11.mlp.experts.14.gate_proj", "model.language_model.layers.11.mlp.experts.15.gate_proj", "model.language_model.layers.11.mlp.experts.16.gate_proj", "model.language_model.layers.11.mlp.experts.17.gate_proj", "model.language_model.layers.11.mlp.experts.18.gate_proj", "model.language_model.layers.11.mlp.experts.19.gate_proj", "model.language_model.layers.11.mlp.experts.20.gate_proj", "model.language_model.layers.11.mlp.experts.21.gate_proj", "model.language_model.layers.11.mlp.experts.22.gate_proj", "model.language_model.layers.11.mlp.experts.23.gate_proj", "model.language_model.layers.11.mlp.experts.24.gate_proj", "model.language_model.layers.11.mlp.experts.25.gate_proj", "model.language_model.layers.11.mlp.experts.26.gate_proj", "model.language_model.layers.11.mlp.experts.27.gate_proj", "model.language_model.layers.11.mlp.experts.28.gate_proj", "model.language_model.layers.11.mlp.experts.29.gate_proj", "model.language_model.layers.11.mlp.experts.30.gate_proj", "model.language_model.layers.11.mlp.experts.31.gate_proj", "model.language_model.layers.11.mlp.experts.32.gate_proj", "model.language_model.layers.11.mlp.experts.33.gate_proj", "model.language_model.layers.11.mlp.experts.34.gate_proj", "model.language_model.layers.11.mlp.experts.35.gate_proj", "model.language_model.layers.11.mlp.experts.36.gate_proj", "model.language_model.layers.11.mlp.experts.37.gate_proj", "model.language_model.layers.11.mlp.experts.38.gate_proj", "model.language_model.layers.11.mlp.experts.39.gate_proj", "model.language_model.layers.11.mlp.experts.40.gate_proj", "model.language_model.layers.11.mlp.experts.41.gate_proj", "model.language_model.layers.11.mlp.experts.42.gate_proj", "model.language_model.layers.11.mlp.experts.43.gate_proj", "model.language_model.layers.11.mlp.experts.44.gate_proj", "model.language_model.layers.11.mlp.experts.45.gate_proj", "model.language_model.layers.11.mlp.experts.46.gate_proj", "model.language_model.layers.11.mlp.experts.47.gate_proj", "model.language_model.layers.11.mlp.experts.48.gate_proj", "model.language_model.layers.11.mlp.experts.49.gate_proj", "model.language_model.layers.11.mlp.experts.50.gate_proj", "model.language_model.layers.11.mlp.experts.51.gate_proj", "model.language_model.layers.11.mlp.experts.52.gate_proj", "model.language_model.layers.11.mlp.experts.53.gate_proj", "model.language_model.layers.11.mlp.experts.54.gate_proj", "model.language_model.layers.11.mlp.experts.55.gate_proj", "model.language_model.layers.11.mlp.experts.56.gate_proj", "model.language_model.layers.11.mlp.experts.57.gate_proj", "model.language_model.layers.11.mlp.experts.58.gate_proj", "model.language_model.layers.11.mlp.experts.59.gate_proj", "model.language_model.layers.11.mlp.experts.60.gate_proj", "model.language_model.layers.11.mlp.experts.61.gate_proj", "model.language_model.layers.11.mlp.experts.62.gate_proj", "model.language_model.layers.11.mlp.experts.63.gate_proj", "model.language_model.layers.11.mlp.experts.64.gate_proj", "model.language_model.layers.11.mlp.experts.65.gate_proj", "model.language_model.layers.11.mlp.experts.66.gate_proj", "model.language_model.layers.11.mlp.experts.67.gate_proj", "model.language_model.layers.11.mlp.experts.68.gate_proj", "model.language_model.layers.11.mlp.experts.69.gate_proj", "model.language_model.layers.11.mlp.experts.70.gate_proj", "model.language_model.layers.11.mlp.experts.71.gate_proj", "model.language_model.layers.11.mlp.experts.72.gate_proj", "model.language_model.layers.11.mlp.experts.73.gate_proj", "model.language_model.layers.11.mlp.experts.74.gate_proj", "model.language_model.layers.11.mlp.experts.75.gate_proj", "model.language_model.layers.11.mlp.experts.76.gate_proj", "model.language_model.layers.11.mlp.experts.77.gate_proj", "model.language_model.layers.11.mlp.experts.78.gate_proj", "model.language_model.layers.11.mlp.experts.79.gate_proj", "model.language_model.layers.11.mlp.experts.80.gate_proj", "model.language_model.layers.11.mlp.experts.81.gate_proj", "model.language_model.layers.11.mlp.experts.82.gate_proj", "model.language_model.layers.11.mlp.experts.83.gate_proj", "model.language_model.layers.11.mlp.experts.84.gate_proj", "model.language_model.layers.11.mlp.experts.85.gate_proj", "model.language_model.layers.11.mlp.experts.86.gate_proj", "model.language_model.layers.11.mlp.experts.87.gate_proj", "model.language_model.layers.11.mlp.experts.88.gate_proj", "model.language_model.layers.11.mlp.experts.89.gate_proj", "model.language_model.layers.11.mlp.experts.90.gate_proj", "model.language_model.layers.11.mlp.experts.91.gate_proj", "model.language_model.layers.11.mlp.experts.92.gate_proj", "model.language_model.layers.11.mlp.experts.93.gate_proj", "model.language_model.layers.11.mlp.experts.94.gate_proj", "model.language_model.layers.11.mlp.experts.95.gate_proj", "model.language_model.layers.11.mlp.experts.96.gate_proj", "model.language_model.layers.11.mlp.experts.97.gate_proj", "model.language_model.layers.11.mlp.experts.98.gate_proj", "model.language_model.layers.11.mlp.experts.99.gate_proj", "model.language_model.layers.11.mlp.experts.100.gate_proj", "model.language_model.layers.11.mlp.experts.101.gate_proj", "model.language_model.layers.11.mlp.experts.102.gate_proj", "model.language_model.layers.11.mlp.experts.103.gate_proj", "model.language_model.layers.11.mlp.experts.104.gate_proj", "model.language_model.layers.11.mlp.experts.105.gate_proj", "model.language_model.layers.11.mlp.experts.106.gate_proj", "model.language_model.layers.11.mlp.experts.107.gate_proj", "model.language_model.layers.11.mlp.experts.108.gate_proj", "model.language_model.layers.11.mlp.experts.109.gate_proj", "model.language_model.layers.11.mlp.experts.110.gate_proj", "model.language_model.layers.11.mlp.experts.111.gate_proj", "model.language_model.layers.11.mlp.experts.112.gate_proj", "model.language_model.layers.11.mlp.experts.113.gate_proj", "model.language_model.layers.11.mlp.experts.114.gate_proj", "model.language_model.layers.11.mlp.experts.115.gate_proj", "model.language_model.layers.11.mlp.experts.116.gate_proj", "model.language_model.layers.11.mlp.experts.117.gate_proj", "model.language_model.layers.11.mlp.experts.118.gate_proj", "model.language_model.layers.11.mlp.experts.119.gate_proj", "model.language_model.layers.11.mlp.experts.120.gate_proj", "model.language_model.layers.11.mlp.experts.121.gate_proj", "model.language_model.layers.11.mlp.experts.122.gate_proj", "model.language_model.layers.11.mlp.experts.123.gate_proj", "model.language_model.layers.11.mlp.experts.124.gate_proj", "model.language_model.layers.11.mlp.experts.125.gate_proj", "model.language_model.layers.11.mlp.experts.126.gate_proj", "model.language_model.layers.11.mlp.experts.127.gate_proj", "model.language_model.layers.11.mlp.experts.0.up_proj", "model.language_model.layers.11.mlp.experts.1.up_proj", "model.language_model.layers.11.mlp.experts.2.up_proj", "model.language_model.layers.11.mlp.experts.3.up_proj", "model.language_model.layers.11.mlp.experts.4.up_proj", "model.language_model.layers.11.mlp.experts.5.up_proj", "model.language_model.layers.11.mlp.experts.6.up_proj", "model.language_model.layers.11.mlp.experts.7.up_proj", "model.language_model.layers.11.mlp.experts.8.up_proj", "model.language_model.layers.11.mlp.experts.9.up_proj", "model.language_model.layers.11.mlp.experts.10.up_proj", "model.language_model.layers.11.mlp.experts.11.up_proj", "model.language_model.layers.11.mlp.experts.12.up_proj", "model.language_model.layers.11.mlp.experts.13.up_proj", "model.language_model.layers.11.mlp.experts.14.up_proj", "model.language_model.layers.11.mlp.experts.15.up_proj", "model.language_model.layers.11.mlp.experts.16.up_proj", "model.language_model.layers.11.mlp.experts.17.up_proj", "model.language_model.layers.11.mlp.experts.18.up_proj", "model.language_model.layers.11.mlp.experts.19.up_proj", "model.language_model.layers.11.mlp.experts.20.up_proj", "model.language_model.layers.11.mlp.experts.21.up_proj", "model.language_model.layers.11.mlp.experts.22.up_proj", "model.language_model.layers.11.mlp.experts.23.up_proj", "model.language_model.layers.11.mlp.experts.24.up_proj", "model.language_model.layers.11.mlp.experts.25.up_proj", "model.language_model.layers.11.mlp.experts.26.up_proj", "model.language_model.layers.11.mlp.experts.27.up_proj", "model.language_model.layers.11.mlp.experts.28.up_proj", "model.language_model.layers.11.mlp.experts.29.up_proj", "model.language_model.layers.11.mlp.experts.30.up_proj", "model.language_model.layers.11.mlp.experts.31.up_proj", "model.language_model.layers.11.mlp.experts.32.up_proj", "model.language_model.layers.11.mlp.experts.33.up_proj", "model.language_model.layers.11.mlp.experts.34.up_proj", "model.language_model.layers.11.mlp.experts.35.up_proj", "model.language_model.layers.11.mlp.experts.36.up_proj", "model.language_model.layers.11.mlp.experts.37.up_proj", "model.language_model.layers.11.mlp.experts.38.up_proj", "model.language_model.layers.11.mlp.experts.39.up_proj", "model.language_model.layers.11.mlp.experts.40.up_proj", "model.language_model.layers.11.mlp.experts.41.up_proj", "model.language_model.layers.11.mlp.experts.42.up_proj", "model.language_model.layers.11.mlp.experts.43.up_proj", "model.language_model.layers.11.mlp.experts.44.up_proj", "model.language_model.layers.11.mlp.experts.45.up_proj", "model.language_model.layers.11.mlp.experts.46.up_proj", "model.language_model.layers.11.mlp.experts.47.up_proj", "model.language_model.layers.11.mlp.experts.48.up_proj", "model.language_model.layers.11.mlp.experts.49.up_proj", "model.language_model.layers.11.mlp.experts.50.up_proj", "model.language_model.layers.11.mlp.experts.51.up_proj", "model.language_model.layers.11.mlp.experts.52.up_proj", "model.language_model.layers.11.mlp.experts.53.up_proj", "model.language_model.layers.11.mlp.experts.54.up_proj", "model.language_model.layers.11.mlp.experts.55.up_proj", "model.language_model.layers.11.mlp.experts.56.up_proj", "model.language_model.layers.11.mlp.experts.57.up_proj", "model.language_model.layers.11.mlp.experts.58.up_proj", "model.language_model.layers.11.mlp.experts.59.up_proj", "model.language_model.layers.11.mlp.experts.60.up_proj", "model.language_model.layers.11.mlp.experts.61.up_proj", "model.language_model.layers.11.mlp.experts.62.up_proj", "model.language_model.layers.11.mlp.experts.63.up_proj", "model.language_model.layers.11.mlp.experts.64.up_proj", "model.language_model.layers.11.mlp.experts.65.up_proj", "model.language_model.layers.11.mlp.experts.66.up_proj", "model.language_model.layers.11.mlp.experts.67.up_proj", "model.language_model.layers.11.mlp.experts.68.up_proj", "model.language_model.layers.11.mlp.experts.69.up_proj", "model.language_model.layers.11.mlp.experts.70.up_proj", "model.language_model.layers.11.mlp.experts.71.up_proj", "model.language_model.layers.11.mlp.experts.72.up_proj", "model.language_model.layers.11.mlp.experts.73.up_proj", "model.language_model.layers.11.mlp.experts.74.up_proj", "model.language_model.layers.11.mlp.experts.75.up_proj", "model.language_model.layers.11.mlp.experts.76.up_proj", "model.language_model.layers.11.mlp.experts.77.up_proj", "model.language_model.layers.11.mlp.experts.78.up_proj", "model.language_model.layers.11.mlp.experts.79.up_proj", "model.language_model.layers.11.mlp.experts.80.up_proj", "model.language_model.layers.11.mlp.experts.81.up_proj", "model.language_model.layers.11.mlp.experts.82.up_proj", "model.language_model.layers.11.mlp.experts.83.up_proj", "model.language_model.layers.11.mlp.experts.84.up_proj", "model.language_model.layers.11.mlp.experts.85.up_proj", "model.language_model.layers.11.mlp.experts.86.up_proj", "model.language_model.layers.11.mlp.experts.87.up_proj", "model.language_model.layers.11.mlp.experts.88.up_proj", "model.language_model.layers.11.mlp.experts.89.up_proj", "model.language_model.layers.11.mlp.experts.90.up_proj", "model.language_model.layers.11.mlp.experts.91.up_proj", "model.language_model.layers.11.mlp.experts.92.up_proj", "model.language_model.layers.11.mlp.experts.93.up_proj", "model.language_model.layers.11.mlp.experts.94.up_proj", "model.language_model.layers.11.mlp.experts.95.up_proj", "model.language_model.layers.11.mlp.experts.96.up_proj", "model.language_model.layers.11.mlp.experts.97.up_proj", "model.language_model.layers.11.mlp.experts.98.up_proj", "model.language_model.layers.11.mlp.experts.99.up_proj", "model.language_model.layers.11.mlp.experts.100.up_proj", "model.language_model.layers.11.mlp.experts.101.up_proj", "model.language_model.layers.11.mlp.experts.102.up_proj", "model.language_model.layers.11.mlp.experts.103.up_proj", "model.language_model.layers.11.mlp.experts.104.up_proj", "model.language_model.layers.11.mlp.experts.105.up_proj", "model.language_model.layers.11.mlp.experts.106.up_proj", "model.language_model.layers.11.mlp.experts.107.up_proj", "model.language_model.layers.11.mlp.experts.108.up_proj", "model.language_model.layers.11.mlp.experts.109.up_proj", "model.language_model.layers.11.mlp.experts.110.up_proj", "model.language_model.layers.11.mlp.experts.111.up_proj", "model.language_model.layers.11.mlp.experts.112.up_proj", "model.language_model.layers.11.mlp.experts.113.up_proj", "model.language_model.layers.11.mlp.experts.114.up_proj", "model.language_model.layers.11.mlp.experts.115.up_proj", "model.language_model.layers.11.mlp.experts.116.up_proj", "model.language_model.layers.11.mlp.experts.117.up_proj", "model.language_model.layers.11.mlp.experts.118.up_proj", "model.language_model.layers.11.mlp.experts.119.up_proj", "model.language_model.layers.11.mlp.experts.120.up_proj", "model.language_model.layers.11.mlp.experts.121.up_proj", "model.language_model.layers.11.mlp.experts.122.up_proj", "model.language_model.layers.11.mlp.experts.123.up_proj", "model.language_model.layers.11.mlp.experts.124.up_proj", "model.language_model.layers.11.mlp.experts.125.up_proj", "model.language_model.layers.11.mlp.experts.126.up_proj", "model.language_model.layers.11.mlp.experts.127.up_proj", "model.language_model.layers.11.mlp.experts.0.down_proj", "model.language_model.layers.11.mlp.experts.1.down_proj", "model.language_model.layers.11.mlp.experts.2.down_proj", "model.language_model.layers.11.mlp.experts.3.down_proj", "model.language_model.layers.11.mlp.experts.4.down_proj", "model.language_model.layers.11.mlp.experts.5.down_proj", "model.language_model.layers.11.mlp.experts.6.down_proj", "model.language_model.layers.11.mlp.experts.7.down_proj", "model.language_model.layers.11.mlp.experts.8.down_proj", "model.language_model.layers.11.mlp.experts.9.down_proj", "model.language_model.layers.11.mlp.experts.10.down_proj", "model.language_model.layers.11.mlp.experts.11.down_proj", "model.language_model.layers.11.mlp.experts.12.down_proj", "model.language_model.layers.11.mlp.experts.13.down_proj", "model.language_model.layers.11.mlp.experts.14.down_proj", "model.language_model.layers.11.mlp.experts.15.down_proj", "model.language_model.layers.11.mlp.experts.16.down_proj", "model.language_model.layers.11.mlp.experts.17.down_proj", "model.language_model.layers.11.mlp.experts.18.down_proj", "model.language_model.layers.11.mlp.experts.19.down_proj", "model.language_model.layers.11.mlp.experts.20.down_proj", "model.language_model.layers.11.mlp.experts.21.down_proj", "model.language_model.layers.11.mlp.experts.22.down_proj", "model.language_model.layers.11.mlp.experts.23.down_proj", "model.language_model.layers.11.mlp.experts.24.down_proj", "model.language_model.layers.11.mlp.experts.25.down_proj", "model.language_model.layers.11.mlp.experts.26.down_proj", "model.language_model.layers.11.mlp.experts.27.down_proj", "model.language_model.layers.11.mlp.experts.28.down_proj", "model.language_model.layers.11.mlp.experts.29.down_proj", "model.language_model.layers.11.mlp.experts.30.down_proj", "model.language_model.layers.11.mlp.experts.31.down_proj", "model.language_model.layers.11.mlp.experts.32.down_proj", "model.language_model.layers.11.mlp.experts.33.down_proj", "model.language_model.layers.11.mlp.experts.34.down_proj", "model.language_model.layers.11.mlp.experts.35.down_proj", "model.language_model.layers.11.mlp.experts.36.down_proj", "model.language_model.layers.11.mlp.experts.37.down_proj", "model.language_model.layers.11.mlp.experts.38.down_proj", "model.language_model.layers.11.mlp.experts.39.down_proj", "model.language_model.layers.11.mlp.experts.40.down_proj", "model.language_model.layers.11.mlp.experts.41.down_proj", "model.language_model.layers.11.mlp.experts.42.down_proj", "model.language_model.layers.11.mlp.experts.43.down_proj", "model.language_model.layers.11.mlp.experts.44.down_proj", "model.language_model.layers.11.mlp.experts.45.down_proj", "model.language_model.layers.11.mlp.experts.46.down_proj", "model.language_model.layers.11.mlp.experts.47.down_proj", "model.language_model.layers.11.mlp.experts.48.down_proj", "model.language_model.layers.11.mlp.experts.49.down_proj", "model.language_model.layers.11.mlp.experts.50.down_proj", "model.language_model.layers.11.mlp.experts.51.down_proj", "model.language_model.layers.11.mlp.experts.52.down_proj", "model.language_model.layers.11.mlp.experts.53.down_proj", "model.language_model.layers.11.mlp.experts.54.down_proj", "model.language_model.layers.11.mlp.experts.55.down_proj", "model.language_model.layers.11.mlp.experts.56.down_proj", "model.language_model.layers.11.mlp.experts.57.down_proj", "model.language_model.layers.11.mlp.experts.58.down_proj", "model.language_model.layers.11.mlp.experts.59.down_proj", "model.language_model.layers.11.mlp.experts.60.down_proj", "model.language_model.layers.11.mlp.experts.61.down_proj", "model.language_model.layers.11.mlp.experts.62.down_proj", "model.language_model.layers.11.mlp.experts.63.down_proj", "model.language_model.layers.11.mlp.experts.64.down_proj", "model.language_model.layers.11.mlp.experts.65.down_proj", "model.language_model.layers.11.mlp.experts.66.down_proj", "model.language_model.layers.11.mlp.experts.67.down_proj", "model.language_model.layers.11.mlp.experts.68.down_proj", "model.language_model.layers.11.mlp.experts.69.down_proj", "model.language_model.layers.11.mlp.experts.70.down_proj", "model.language_model.layers.11.mlp.experts.71.down_proj", "model.language_model.layers.11.mlp.experts.72.down_proj", "model.language_model.layers.11.mlp.experts.73.down_proj", "model.language_model.layers.11.mlp.experts.74.down_proj", "model.language_model.layers.11.mlp.experts.75.down_proj", "model.language_model.layers.11.mlp.experts.76.down_proj", "model.language_model.layers.11.mlp.experts.77.down_proj", "model.language_model.layers.11.mlp.experts.78.down_proj", "model.language_model.layers.11.mlp.experts.79.down_proj", "model.language_model.layers.11.mlp.experts.80.down_proj", "model.language_model.layers.11.mlp.experts.81.down_proj", "model.language_model.layers.11.mlp.experts.82.down_proj", "model.language_model.layers.11.mlp.experts.83.down_proj", "model.language_model.layers.11.mlp.experts.84.down_proj", "model.language_model.layers.11.mlp.experts.85.down_proj", "model.language_model.layers.11.mlp.experts.86.down_proj", "model.language_model.layers.11.mlp.experts.87.down_proj", "model.language_model.layers.11.mlp.experts.88.down_proj", "model.language_model.layers.11.mlp.experts.89.down_proj", "model.language_model.layers.11.mlp.experts.90.down_proj", "model.language_model.layers.11.mlp.experts.91.down_proj", "model.language_model.layers.11.mlp.experts.92.down_proj", "model.language_model.layers.11.mlp.experts.93.down_proj", "model.language_model.layers.11.mlp.experts.94.down_proj", "model.language_model.layers.11.mlp.experts.95.down_proj", "model.language_model.layers.11.mlp.experts.96.down_proj", "model.language_model.layers.11.mlp.experts.97.down_proj", "model.language_model.layers.11.mlp.experts.98.down_proj", "model.language_model.layers.11.mlp.experts.99.down_proj", "model.language_model.layers.11.mlp.experts.100.down_proj", "model.language_model.layers.11.mlp.experts.101.down_proj", "model.language_model.layers.11.mlp.experts.102.down_proj", "model.language_model.layers.11.mlp.experts.103.down_proj", "model.language_model.layers.11.mlp.experts.104.down_proj", "model.language_model.layers.11.mlp.experts.105.down_proj", "model.language_model.layers.11.mlp.experts.106.down_proj", "model.language_model.layers.11.mlp.experts.107.down_proj", "model.language_model.layers.11.mlp.experts.108.down_proj", "model.language_model.layers.11.mlp.experts.109.down_proj", "model.language_model.layers.11.mlp.experts.110.down_proj", "model.language_model.layers.11.mlp.experts.111.down_proj", "model.language_model.layers.11.mlp.experts.112.down_proj", "model.language_model.layers.11.mlp.experts.113.down_proj", "model.language_model.layers.11.mlp.experts.114.down_proj", "model.language_model.layers.11.mlp.experts.115.down_proj", "model.language_model.layers.11.mlp.experts.116.down_proj", "model.language_model.layers.11.mlp.experts.117.down_proj", "model.language_model.layers.11.mlp.experts.118.down_proj", "model.language_model.layers.11.mlp.experts.119.down_proj", "model.language_model.layers.11.mlp.experts.120.down_proj", "model.language_model.layers.11.mlp.experts.121.down_proj", "model.language_model.layers.11.mlp.experts.122.down_proj", "model.language_model.layers.11.mlp.experts.123.down_proj", "model.language_model.layers.11.mlp.experts.124.down_proj", "model.language_model.layers.11.mlp.experts.125.down_proj", "model.language_model.layers.11.mlp.experts.126.down_proj", "model.language_model.layers.11.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.0020969027653336553, "dbits": 603979776 } ] }, { "idx": 24, "layers": [ "model.language_model.layers.12.self_attn.q_proj", "model.language_model.layers.12.self_attn.k_proj", "model.language_model.layers.12.self_attn.v_proj", "model.language_model.layers.12.self_attn.o_proj" ], "candidates": [ { "dkld": 0.00018256278708576879, "dbits": 18874368 } ] }, { "idx": 25, "layers": [ "model.language_model.layers.12.mlp.experts.0.gate_proj", "model.language_model.layers.12.mlp.experts.1.gate_proj", "model.language_model.layers.12.mlp.experts.2.gate_proj", "model.language_model.layers.12.mlp.experts.3.gate_proj", "model.language_model.layers.12.mlp.experts.4.gate_proj", "model.language_model.layers.12.mlp.experts.5.gate_proj", "model.language_model.layers.12.mlp.experts.6.gate_proj", "model.language_model.layers.12.mlp.experts.7.gate_proj", "model.language_model.layers.12.mlp.experts.8.gate_proj", "model.language_model.layers.12.mlp.experts.9.gate_proj", "model.language_model.layers.12.mlp.experts.10.gate_proj", "model.language_model.layers.12.mlp.experts.11.gate_proj", "model.language_model.layers.12.mlp.experts.12.gate_proj", "model.language_model.layers.12.mlp.experts.13.gate_proj", "model.language_model.layers.12.mlp.experts.14.gate_proj", "model.language_model.layers.12.mlp.experts.15.gate_proj", "model.language_model.layers.12.mlp.experts.16.gate_proj", "model.language_model.layers.12.mlp.experts.17.gate_proj", "model.language_model.layers.12.mlp.experts.18.gate_proj", "model.language_model.layers.12.mlp.experts.19.gate_proj", "model.language_model.layers.12.mlp.experts.20.gate_proj", "model.language_model.layers.12.mlp.experts.21.gate_proj", "model.language_model.layers.12.mlp.experts.22.gate_proj", "model.language_model.layers.12.mlp.experts.23.gate_proj", "model.language_model.layers.12.mlp.experts.24.gate_proj", "model.language_model.layers.12.mlp.experts.25.gate_proj", "model.language_model.layers.12.mlp.experts.26.gate_proj", "model.language_model.layers.12.mlp.experts.27.gate_proj", "model.language_model.layers.12.mlp.experts.28.gate_proj", "model.language_model.layers.12.mlp.experts.29.gate_proj", "model.language_model.layers.12.mlp.experts.30.gate_proj", "model.language_model.layers.12.mlp.experts.31.gate_proj", "model.language_model.layers.12.mlp.experts.32.gate_proj", "model.language_model.layers.12.mlp.experts.33.gate_proj", "model.language_model.layers.12.mlp.experts.34.gate_proj", "model.language_model.layers.12.mlp.experts.35.gate_proj", "model.language_model.layers.12.mlp.experts.36.gate_proj", "model.language_model.layers.12.mlp.experts.37.gate_proj", "model.language_model.layers.12.mlp.experts.38.gate_proj", "model.language_model.layers.12.mlp.experts.39.gate_proj", "model.language_model.layers.12.mlp.experts.40.gate_proj", "model.language_model.layers.12.mlp.experts.41.gate_proj", "model.language_model.layers.12.mlp.experts.42.gate_proj", "model.language_model.layers.12.mlp.experts.43.gate_proj", "model.language_model.layers.12.mlp.experts.44.gate_proj", "model.language_model.layers.12.mlp.experts.45.gate_proj", "model.language_model.layers.12.mlp.experts.46.gate_proj", "model.language_model.layers.12.mlp.experts.47.gate_proj", "model.language_model.layers.12.mlp.experts.48.gate_proj", "model.language_model.layers.12.mlp.experts.49.gate_proj", "model.language_model.layers.12.mlp.experts.50.gate_proj", "model.language_model.layers.12.mlp.experts.51.gate_proj", "model.language_model.layers.12.mlp.experts.52.gate_proj", "model.language_model.layers.12.mlp.experts.53.gate_proj", "model.language_model.layers.12.mlp.experts.54.gate_proj", "model.language_model.layers.12.mlp.experts.55.gate_proj", "model.language_model.layers.12.mlp.experts.56.gate_proj", "model.language_model.layers.12.mlp.experts.57.gate_proj", "model.language_model.layers.12.mlp.experts.58.gate_proj", "model.language_model.layers.12.mlp.experts.59.gate_proj", "model.language_model.layers.12.mlp.experts.60.gate_proj", "model.language_model.layers.12.mlp.experts.61.gate_proj", "model.language_model.layers.12.mlp.experts.62.gate_proj", "model.language_model.layers.12.mlp.experts.63.gate_proj", "model.language_model.layers.12.mlp.experts.64.gate_proj", "model.language_model.layers.12.mlp.experts.65.gate_proj", "model.language_model.layers.12.mlp.experts.66.gate_proj", "model.language_model.layers.12.mlp.experts.67.gate_proj", "model.language_model.layers.12.mlp.experts.68.gate_proj", "model.language_model.layers.12.mlp.experts.69.gate_proj", "model.language_model.layers.12.mlp.experts.70.gate_proj", "model.language_model.layers.12.mlp.experts.71.gate_proj", "model.language_model.layers.12.mlp.experts.72.gate_proj", "model.language_model.layers.12.mlp.experts.73.gate_proj", "model.language_model.layers.12.mlp.experts.74.gate_proj", "model.language_model.layers.12.mlp.experts.75.gate_proj", "model.language_model.layers.12.mlp.experts.76.gate_proj", "model.language_model.layers.12.mlp.experts.77.gate_proj", "model.language_model.layers.12.mlp.experts.78.gate_proj", "model.language_model.layers.12.mlp.experts.79.gate_proj", "model.language_model.layers.12.mlp.experts.80.gate_proj", "model.language_model.layers.12.mlp.experts.81.gate_proj", "model.language_model.layers.12.mlp.experts.82.gate_proj", "model.language_model.layers.12.mlp.experts.83.gate_proj", "model.language_model.layers.12.mlp.experts.84.gate_proj", "model.language_model.layers.12.mlp.experts.85.gate_proj", "model.language_model.layers.12.mlp.experts.86.gate_proj", "model.language_model.layers.12.mlp.experts.87.gate_proj", "model.language_model.layers.12.mlp.experts.88.gate_proj", "model.language_model.layers.12.mlp.experts.89.gate_proj", "model.language_model.layers.12.mlp.experts.90.gate_proj", "model.language_model.layers.12.mlp.experts.91.gate_proj", "model.language_model.layers.12.mlp.experts.92.gate_proj", "model.language_model.layers.12.mlp.experts.93.gate_proj", "model.language_model.layers.12.mlp.experts.94.gate_proj", "model.language_model.layers.12.mlp.experts.95.gate_proj", "model.language_model.layers.12.mlp.experts.96.gate_proj", "model.language_model.layers.12.mlp.experts.97.gate_proj", "model.language_model.layers.12.mlp.experts.98.gate_proj", "model.language_model.layers.12.mlp.experts.99.gate_proj", "model.language_model.layers.12.mlp.experts.100.gate_proj", "model.language_model.layers.12.mlp.experts.101.gate_proj", "model.language_model.layers.12.mlp.experts.102.gate_proj", "model.language_model.layers.12.mlp.experts.103.gate_proj", "model.language_model.layers.12.mlp.experts.104.gate_proj", "model.language_model.layers.12.mlp.experts.105.gate_proj", "model.language_model.layers.12.mlp.experts.106.gate_proj", "model.language_model.layers.12.mlp.experts.107.gate_proj", "model.language_model.layers.12.mlp.experts.108.gate_proj", "model.language_model.layers.12.mlp.experts.109.gate_proj", "model.language_model.layers.12.mlp.experts.110.gate_proj", "model.language_model.layers.12.mlp.experts.111.gate_proj", "model.language_model.layers.12.mlp.experts.112.gate_proj", "model.language_model.layers.12.mlp.experts.113.gate_proj", "model.language_model.layers.12.mlp.experts.114.gate_proj", "model.language_model.layers.12.mlp.experts.115.gate_proj", "model.language_model.layers.12.mlp.experts.116.gate_proj", "model.language_model.layers.12.mlp.experts.117.gate_proj", "model.language_model.layers.12.mlp.experts.118.gate_proj", "model.language_model.layers.12.mlp.experts.119.gate_proj", "model.language_model.layers.12.mlp.experts.120.gate_proj", "model.language_model.layers.12.mlp.experts.121.gate_proj", "model.language_model.layers.12.mlp.experts.122.gate_proj", "model.language_model.layers.12.mlp.experts.123.gate_proj", "model.language_model.layers.12.mlp.experts.124.gate_proj", "model.language_model.layers.12.mlp.experts.125.gate_proj", "model.language_model.layers.12.mlp.experts.126.gate_proj", "model.language_model.layers.12.mlp.experts.127.gate_proj", "model.language_model.layers.12.mlp.experts.0.up_proj", "model.language_model.layers.12.mlp.experts.1.up_proj", "model.language_model.layers.12.mlp.experts.2.up_proj", "model.language_model.layers.12.mlp.experts.3.up_proj", "model.language_model.layers.12.mlp.experts.4.up_proj", "model.language_model.layers.12.mlp.experts.5.up_proj", "model.language_model.layers.12.mlp.experts.6.up_proj", "model.language_model.layers.12.mlp.experts.7.up_proj", "model.language_model.layers.12.mlp.experts.8.up_proj", "model.language_model.layers.12.mlp.experts.9.up_proj", "model.language_model.layers.12.mlp.experts.10.up_proj", "model.language_model.layers.12.mlp.experts.11.up_proj", "model.language_model.layers.12.mlp.experts.12.up_proj", "model.language_model.layers.12.mlp.experts.13.up_proj", "model.language_model.layers.12.mlp.experts.14.up_proj", "model.language_model.layers.12.mlp.experts.15.up_proj", "model.language_model.layers.12.mlp.experts.16.up_proj", "model.language_model.layers.12.mlp.experts.17.up_proj", "model.language_model.layers.12.mlp.experts.18.up_proj", "model.language_model.layers.12.mlp.experts.19.up_proj", "model.language_model.layers.12.mlp.experts.20.up_proj", "model.language_model.layers.12.mlp.experts.21.up_proj", "model.language_model.layers.12.mlp.experts.22.up_proj", "model.language_model.layers.12.mlp.experts.23.up_proj", "model.language_model.layers.12.mlp.experts.24.up_proj", "model.language_model.layers.12.mlp.experts.25.up_proj", "model.language_model.layers.12.mlp.experts.26.up_proj", "model.language_model.layers.12.mlp.experts.27.up_proj", "model.language_model.layers.12.mlp.experts.28.up_proj", "model.language_model.layers.12.mlp.experts.29.up_proj", "model.language_model.layers.12.mlp.experts.30.up_proj", "model.language_model.layers.12.mlp.experts.31.up_proj", "model.language_model.layers.12.mlp.experts.32.up_proj", "model.language_model.layers.12.mlp.experts.33.up_proj", "model.language_model.layers.12.mlp.experts.34.up_proj", "model.language_model.layers.12.mlp.experts.35.up_proj", "model.language_model.layers.12.mlp.experts.36.up_proj", "model.language_model.layers.12.mlp.experts.37.up_proj", "model.language_model.layers.12.mlp.experts.38.up_proj", "model.language_model.layers.12.mlp.experts.39.up_proj", "model.language_model.layers.12.mlp.experts.40.up_proj", "model.language_model.layers.12.mlp.experts.41.up_proj", "model.language_model.layers.12.mlp.experts.42.up_proj", "model.language_model.layers.12.mlp.experts.43.up_proj", "model.language_model.layers.12.mlp.experts.44.up_proj", "model.language_model.layers.12.mlp.experts.45.up_proj", "model.language_model.layers.12.mlp.experts.46.up_proj", "model.language_model.layers.12.mlp.experts.47.up_proj", "model.language_model.layers.12.mlp.experts.48.up_proj", "model.language_model.layers.12.mlp.experts.49.up_proj", "model.language_model.layers.12.mlp.experts.50.up_proj", "model.language_model.layers.12.mlp.experts.51.up_proj", "model.language_model.layers.12.mlp.experts.52.up_proj", "model.language_model.layers.12.mlp.experts.53.up_proj", "model.language_model.layers.12.mlp.experts.54.up_proj", "model.language_model.layers.12.mlp.experts.55.up_proj", "model.language_model.layers.12.mlp.experts.56.up_proj", "model.language_model.layers.12.mlp.experts.57.up_proj", "model.language_model.layers.12.mlp.experts.58.up_proj", "model.language_model.layers.12.mlp.experts.59.up_proj", "model.language_model.layers.12.mlp.experts.60.up_proj", "model.language_model.layers.12.mlp.experts.61.up_proj", "model.language_model.layers.12.mlp.experts.62.up_proj", "model.language_model.layers.12.mlp.experts.63.up_proj", "model.language_model.layers.12.mlp.experts.64.up_proj", "model.language_model.layers.12.mlp.experts.65.up_proj", "model.language_model.layers.12.mlp.experts.66.up_proj", "model.language_model.layers.12.mlp.experts.67.up_proj", "model.language_model.layers.12.mlp.experts.68.up_proj", "model.language_model.layers.12.mlp.experts.69.up_proj", "model.language_model.layers.12.mlp.experts.70.up_proj", "model.language_model.layers.12.mlp.experts.71.up_proj", "model.language_model.layers.12.mlp.experts.72.up_proj", "model.language_model.layers.12.mlp.experts.73.up_proj", "model.language_model.layers.12.mlp.experts.74.up_proj", "model.language_model.layers.12.mlp.experts.75.up_proj", "model.language_model.layers.12.mlp.experts.76.up_proj", "model.language_model.layers.12.mlp.experts.77.up_proj", "model.language_model.layers.12.mlp.experts.78.up_proj", "model.language_model.layers.12.mlp.experts.79.up_proj", "model.language_model.layers.12.mlp.experts.80.up_proj", "model.language_model.layers.12.mlp.experts.81.up_proj", "model.language_model.layers.12.mlp.experts.82.up_proj", "model.language_model.layers.12.mlp.experts.83.up_proj", "model.language_model.layers.12.mlp.experts.84.up_proj", "model.language_model.layers.12.mlp.experts.85.up_proj", "model.language_model.layers.12.mlp.experts.86.up_proj", "model.language_model.layers.12.mlp.experts.87.up_proj", "model.language_model.layers.12.mlp.experts.88.up_proj", "model.language_model.layers.12.mlp.experts.89.up_proj", "model.language_model.layers.12.mlp.experts.90.up_proj", "model.language_model.layers.12.mlp.experts.91.up_proj", "model.language_model.layers.12.mlp.experts.92.up_proj", "model.language_model.layers.12.mlp.experts.93.up_proj", "model.language_model.layers.12.mlp.experts.94.up_proj", "model.language_model.layers.12.mlp.experts.95.up_proj", "model.language_model.layers.12.mlp.experts.96.up_proj", "model.language_model.layers.12.mlp.experts.97.up_proj", "model.language_model.layers.12.mlp.experts.98.up_proj", "model.language_model.layers.12.mlp.experts.99.up_proj", "model.language_model.layers.12.mlp.experts.100.up_proj", "model.language_model.layers.12.mlp.experts.101.up_proj", "model.language_model.layers.12.mlp.experts.102.up_proj", "model.language_model.layers.12.mlp.experts.103.up_proj", "model.language_model.layers.12.mlp.experts.104.up_proj", "model.language_model.layers.12.mlp.experts.105.up_proj", "model.language_model.layers.12.mlp.experts.106.up_proj", "model.language_model.layers.12.mlp.experts.107.up_proj", "model.language_model.layers.12.mlp.experts.108.up_proj", "model.language_model.layers.12.mlp.experts.109.up_proj", "model.language_model.layers.12.mlp.experts.110.up_proj", "model.language_model.layers.12.mlp.experts.111.up_proj", "model.language_model.layers.12.mlp.experts.112.up_proj", "model.language_model.layers.12.mlp.experts.113.up_proj", "model.language_model.layers.12.mlp.experts.114.up_proj", "model.language_model.layers.12.mlp.experts.115.up_proj", "model.language_model.layers.12.mlp.experts.116.up_proj", "model.language_model.layers.12.mlp.experts.117.up_proj", "model.language_model.layers.12.mlp.experts.118.up_proj", "model.language_model.layers.12.mlp.experts.119.up_proj", "model.language_model.layers.12.mlp.experts.120.up_proj", "model.language_model.layers.12.mlp.experts.121.up_proj", "model.language_model.layers.12.mlp.experts.122.up_proj", "model.language_model.layers.12.mlp.experts.123.up_proj", "model.language_model.layers.12.mlp.experts.124.up_proj", "model.language_model.layers.12.mlp.experts.125.up_proj", "model.language_model.layers.12.mlp.experts.126.up_proj", "model.language_model.layers.12.mlp.experts.127.up_proj", "model.language_model.layers.12.mlp.experts.0.down_proj", "model.language_model.layers.12.mlp.experts.1.down_proj", "model.language_model.layers.12.mlp.experts.2.down_proj", "model.language_model.layers.12.mlp.experts.3.down_proj", "model.language_model.layers.12.mlp.experts.4.down_proj", "model.language_model.layers.12.mlp.experts.5.down_proj", "model.language_model.layers.12.mlp.experts.6.down_proj", "model.language_model.layers.12.mlp.experts.7.down_proj", "model.language_model.layers.12.mlp.experts.8.down_proj", "model.language_model.layers.12.mlp.experts.9.down_proj", "model.language_model.layers.12.mlp.experts.10.down_proj", "model.language_model.layers.12.mlp.experts.11.down_proj", "model.language_model.layers.12.mlp.experts.12.down_proj", "model.language_model.layers.12.mlp.experts.13.down_proj", "model.language_model.layers.12.mlp.experts.14.down_proj", "model.language_model.layers.12.mlp.experts.15.down_proj", "model.language_model.layers.12.mlp.experts.16.down_proj", "model.language_model.layers.12.mlp.experts.17.down_proj", "model.language_model.layers.12.mlp.experts.18.down_proj", "model.language_model.layers.12.mlp.experts.19.down_proj", "model.language_model.layers.12.mlp.experts.20.down_proj", "model.language_model.layers.12.mlp.experts.21.down_proj", "model.language_model.layers.12.mlp.experts.22.down_proj", "model.language_model.layers.12.mlp.experts.23.down_proj", "model.language_model.layers.12.mlp.experts.24.down_proj", "model.language_model.layers.12.mlp.experts.25.down_proj", "model.language_model.layers.12.mlp.experts.26.down_proj", "model.language_model.layers.12.mlp.experts.27.down_proj", "model.language_model.layers.12.mlp.experts.28.down_proj", "model.language_model.layers.12.mlp.experts.29.down_proj", "model.language_model.layers.12.mlp.experts.30.down_proj", "model.language_model.layers.12.mlp.experts.31.down_proj", "model.language_model.layers.12.mlp.experts.32.down_proj", "model.language_model.layers.12.mlp.experts.33.down_proj", "model.language_model.layers.12.mlp.experts.34.down_proj", "model.language_model.layers.12.mlp.experts.35.down_proj", "model.language_model.layers.12.mlp.experts.36.down_proj", "model.language_model.layers.12.mlp.experts.37.down_proj", "model.language_model.layers.12.mlp.experts.38.down_proj", "model.language_model.layers.12.mlp.experts.39.down_proj", "model.language_model.layers.12.mlp.experts.40.down_proj", "model.language_model.layers.12.mlp.experts.41.down_proj", "model.language_model.layers.12.mlp.experts.42.down_proj", "model.language_model.layers.12.mlp.experts.43.down_proj", "model.language_model.layers.12.mlp.experts.44.down_proj", "model.language_model.layers.12.mlp.experts.45.down_proj", "model.language_model.layers.12.mlp.experts.46.down_proj", "model.language_model.layers.12.mlp.experts.47.down_proj", "model.language_model.layers.12.mlp.experts.48.down_proj", "model.language_model.layers.12.mlp.experts.49.down_proj", "model.language_model.layers.12.mlp.experts.50.down_proj", "model.language_model.layers.12.mlp.experts.51.down_proj", "model.language_model.layers.12.mlp.experts.52.down_proj", "model.language_model.layers.12.mlp.experts.53.down_proj", "model.language_model.layers.12.mlp.experts.54.down_proj", "model.language_model.layers.12.mlp.experts.55.down_proj", "model.language_model.layers.12.mlp.experts.56.down_proj", "model.language_model.layers.12.mlp.experts.57.down_proj", "model.language_model.layers.12.mlp.experts.58.down_proj", "model.language_model.layers.12.mlp.experts.59.down_proj", "model.language_model.layers.12.mlp.experts.60.down_proj", "model.language_model.layers.12.mlp.experts.61.down_proj", "model.language_model.layers.12.mlp.experts.62.down_proj", "model.language_model.layers.12.mlp.experts.63.down_proj", "model.language_model.layers.12.mlp.experts.64.down_proj", "model.language_model.layers.12.mlp.experts.65.down_proj", "model.language_model.layers.12.mlp.experts.66.down_proj", "model.language_model.layers.12.mlp.experts.67.down_proj", "model.language_model.layers.12.mlp.experts.68.down_proj", "model.language_model.layers.12.mlp.experts.69.down_proj", "model.language_model.layers.12.mlp.experts.70.down_proj", "model.language_model.layers.12.mlp.experts.71.down_proj", "model.language_model.layers.12.mlp.experts.72.down_proj", "model.language_model.layers.12.mlp.experts.73.down_proj", "model.language_model.layers.12.mlp.experts.74.down_proj", "model.language_model.layers.12.mlp.experts.75.down_proj", "model.language_model.layers.12.mlp.experts.76.down_proj", "model.language_model.layers.12.mlp.experts.77.down_proj", "model.language_model.layers.12.mlp.experts.78.down_proj", "model.language_model.layers.12.mlp.experts.79.down_proj", "model.language_model.layers.12.mlp.experts.80.down_proj", "model.language_model.layers.12.mlp.experts.81.down_proj", "model.language_model.layers.12.mlp.experts.82.down_proj", "model.language_model.layers.12.mlp.experts.83.down_proj", "model.language_model.layers.12.mlp.experts.84.down_proj", "model.language_model.layers.12.mlp.experts.85.down_proj", "model.language_model.layers.12.mlp.experts.86.down_proj", "model.language_model.layers.12.mlp.experts.87.down_proj", "model.language_model.layers.12.mlp.experts.88.down_proj", "model.language_model.layers.12.mlp.experts.89.down_proj", "model.language_model.layers.12.mlp.experts.90.down_proj", "model.language_model.layers.12.mlp.experts.91.down_proj", "model.language_model.layers.12.mlp.experts.92.down_proj", "model.language_model.layers.12.mlp.experts.93.down_proj", "model.language_model.layers.12.mlp.experts.94.down_proj", "model.language_model.layers.12.mlp.experts.95.down_proj", "model.language_model.layers.12.mlp.experts.96.down_proj", "model.language_model.layers.12.mlp.experts.97.down_proj", "model.language_model.layers.12.mlp.experts.98.down_proj", "model.language_model.layers.12.mlp.experts.99.down_proj", "model.language_model.layers.12.mlp.experts.100.down_proj", "model.language_model.layers.12.mlp.experts.101.down_proj", "model.language_model.layers.12.mlp.experts.102.down_proj", "model.language_model.layers.12.mlp.experts.103.down_proj", "model.language_model.layers.12.mlp.experts.104.down_proj", "model.language_model.layers.12.mlp.experts.105.down_proj", "model.language_model.layers.12.mlp.experts.106.down_proj", "model.language_model.layers.12.mlp.experts.107.down_proj", "model.language_model.layers.12.mlp.experts.108.down_proj", "model.language_model.layers.12.mlp.experts.109.down_proj", "model.language_model.layers.12.mlp.experts.110.down_proj", "model.language_model.layers.12.mlp.experts.111.down_proj", "model.language_model.layers.12.mlp.experts.112.down_proj", "model.language_model.layers.12.mlp.experts.113.down_proj", "model.language_model.layers.12.mlp.experts.114.down_proj", "model.language_model.layers.12.mlp.experts.115.down_proj", "model.language_model.layers.12.mlp.experts.116.down_proj", "model.language_model.layers.12.mlp.experts.117.down_proj", "model.language_model.layers.12.mlp.experts.118.down_proj", "model.language_model.layers.12.mlp.experts.119.down_proj", "model.language_model.layers.12.mlp.experts.120.down_proj", "model.language_model.layers.12.mlp.experts.121.down_proj", "model.language_model.layers.12.mlp.experts.122.down_proj", "model.language_model.layers.12.mlp.experts.123.down_proj", "model.language_model.layers.12.mlp.experts.124.down_proj", "model.language_model.layers.12.mlp.experts.125.down_proj", "model.language_model.layers.12.mlp.experts.126.down_proj", "model.language_model.layers.12.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": 0.026037120260298252, "dbits": 603979776 } ] }, { "idx": 26, "layers": [ "model.language_model.layers.13.self_attn.q_proj", "model.language_model.layers.13.self_attn.k_proj", "model.language_model.layers.13.self_attn.v_proj", "model.language_model.layers.13.self_attn.o_proj" ], "candidates": [ { "dkld": 0.0008125194348394815, "dbits": 18874368 } ] }, { "idx": 27, "layers": [ "model.language_model.layers.13.mlp.experts.0.gate_proj", "model.language_model.layers.13.mlp.experts.1.gate_proj", "model.language_model.layers.13.mlp.experts.2.gate_proj", "model.language_model.layers.13.mlp.experts.3.gate_proj", "model.language_model.layers.13.mlp.experts.4.gate_proj", "model.language_model.layers.13.mlp.experts.5.gate_proj", "model.language_model.layers.13.mlp.experts.6.gate_proj", "model.language_model.layers.13.mlp.experts.7.gate_proj", "model.language_model.layers.13.mlp.experts.8.gate_proj", "model.language_model.layers.13.mlp.experts.9.gate_proj", "model.language_model.layers.13.mlp.experts.10.gate_proj", "model.language_model.layers.13.mlp.experts.11.gate_proj", "model.language_model.layers.13.mlp.experts.12.gate_proj", "model.language_model.layers.13.mlp.experts.13.gate_proj", "model.language_model.layers.13.mlp.experts.14.gate_proj", "model.language_model.layers.13.mlp.experts.15.gate_proj", "model.language_model.layers.13.mlp.experts.16.gate_proj", "model.language_model.layers.13.mlp.experts.17.gate_proj", "model.language_model.layers.13.mlp.experts.18.gate_proj", "model.language_model.layers.13.mlp.experts.19.gate_proj", "model.language_model.layers.13.mlp.experts.20.gate_proj", "model.language_model.layers.13.mlp.experts.21.gate_proj", "model.language_model.layers.13.mlp.experts.22.gate_proj", "model.language_model.layers.13.mlp.experts.23.gate_proj", "model.language_model.layers.13.mlp.experts.24.gate_proj", "model.language_model.layers.13.mlp.experts.25.gate_proj", "model.language_model.layers.13.mlp.experts.26.gate_proj", "model.language_model.layers.13.mlp.experts.27.gate_proj", "model.language_model.layers.13.mlp.experts.28.gate_proj", "model.language_model.layers.13.mlp.experts.29.gate_proj", "model.language_model.layers.13.mlp.experts.30.gate_proj", "model.language_model.layers.13.mlp.experts.31.gate_proj", "model.language_model.layers.13.mlp.experts.32.gate_proj", "model.language_model.layers.13.mlp.experts.33.gate_proj", "model.language_model.layers.13.mlp.experts.34.gate_proj", "model.language_model.layers.13.mlp.experts.35.gate_proj", "model.language_model.layers.13.mlp.experts.36.gate_proj", "model.language_model.layers.13.mlp.experts.37.gate_proj", "model.language_model.layers.13.mlp.experts.38.gate_proj", "model.language_model.layers.13.mlp.experts.39.gate_proj", "model.language_model.layers.13.mlp.experts.40.gate_proj", "model.language_model.layers.13.mlp.experts.41.gate_proj", "model.language_model.layers.13.mlp.experts.42.gate_proj", "model.language_model.layers.13.mlp.experts.43.gate_proj", "model.language_model.layers.13.mlp.experts.44.gate_proj", "model.language_model.layers.13.mlp.experts.45.gate_proj", "model.language_model.layers.13.mlp.experts.46.gate_proj", "model.language_model.layers.13.mlp.experts.47.gate_proj", "model.language_model.layers.13.mlp.experts.48.gate_proj", "model.language_model.layers.13.mlp.experts.49.gate_proj", "model.language_model.layers.13.mlp.experts.50.gate_proj", "model.language_model.layers.13.mlp.experts.51.gate_proj", "model.language_model.layers.13.mlp.experts.52.gate_proj", "model.language_model.layers.13.mlp.experts.53.gate_proj", "model.language_model.layers.13.mlp.experts.54.gate_proj", "model.language_model.layers.13.mlp.experts.55.gate_proj", "model.language_model.layers.13.mlp.experts.56.gate_proj", "model.language_model.layers.13.mlp.experts.57.gate_proj", "model.language_model.layers.13.mlp.experts.58.gate_proj", "model.language_model.layers.13.mlp.experts.59.gate_proj", "model.language_model.layers.13.mlp.experts.60.gate_proj", "model.language_model.layers.13.mlp.experts.61.gate_proj", "model.language_model.layers.13.mlp.experts.62.gate_proj", "model.language_model.layers.13.mlp.experts.63.gate_proj", "model.language_model.layers.13.mlp.experts.64.gate_proj", "model.language_model.layers.13.mlp.experts.65.gate_proj", "model.language_model.layers.13.mlp.experts.66.gate_proj", "model.language_model.layers.13.mlp.experts.67.gate_proj", "model.language_model.layers.13.mlp.experts.68.gate_proj", "model.language_model.layers.13.mlp.experts.69.gate_proj", "model.language_model.layers.13.mlp.experts.70.gate_proj", "model.language_model.layers.13.mlp.experts.71.gate_proj", "model.language_model.layers.13.mlp.experts.72.gate_proj", "model.language_model.layers.13.mlp.experts.73.gate_proj", "model.language_model.layers.13.mlp.experts.74.gate_proj", "model.language_model.layers.13.mlp.experts.75.gate_proj", "model.language_model.layers.13.mlp.experts.76.gate_proj", "model.language_model.layers.13.mlp.experts.77.gate_proj", "model.language_model.layers.13.mlp.experts.78.gate_proj", "model.language_model.layers.13.mlp.experts.79.gate_proj", "model.language_model.layers.13.mlp.experts.80.gate_proj", "model.language_model.layers.13.mlp.experts.81.gate_proj", "model.language_model.layers.13.mlp.experts.82.gate_proj", "model.language_model.layers.13.mlp.experts.83.gate_proj", "model.language_model.layers.13.mlp.experts.84.gate_proj", "model.language_model.layers.13.mlp.experts.85.gate_proj", "model.language_model.layers.13.mlp.experts.86.gate_proj", "model.language_model.layers.13.mlp.experts.87.gate_proj", "model.language_model.layers.13.mlp.experts.88.gate_proj", "model.language_model.layers.13.mlp.experts.89.gate_proj", "model.language_model.layers.13.mlp.experts.90.gate_proj", "model.language_model.layers.13.mlp.experts.91.gate_proj", "model.language_model.layers.13.mlp.experts.92.gate_proj", "model.language_model.layers.13.mlp.experts.93.gate_proj", "model.language_model.layers.13.mlp.experts.94.gate_proj", "model.language_model.layers.13.mlp.experts.95.gate_proj", "model.language_model.layers.13.mlp.experts.96.gate_proj", "model.language_model.layers.13.mlp.experts.97.gate_proj", "model.language_model.layers.13.mlp.experts.98.gate_proj", "model.language_model.layers.13.mlp.experts.99.gate_proj", "model.language_model.layers.13.mlp.experts.100.gate_proj", "model.language_model.layers.13.mlp.experts.101.gate_proj", "model.language_model.layers.13.mlp.experts.102.gate_proj", "model.language_model.layers.13.mlp.experts.103.gate_proj", "model.language_model.layers.13.mlp.experts.104.gate_proj", "model.language_model.layers.13.mlp.experts.105.gate_proj", "model.language_model.layers.13.mlp.experts.106.gate_proj", "model.language_model.layers.13.mlp.experts.107.gate_proj", "model.language_model.layers.13.mlp.experts.108.gate_proj", "model.language_model.layers.13.mlp.experts.109.gate_proj", "model.language_model.layers.13.mlp.experts.110.gate_proj", "model.language_model.layers.13.mlp.experts.111.gate_proj", "model.language_model.layers.13.mlp.experts.112.gate_proj", "model.language_model.layers.13.mlp.experts.113.gate_proj", "model.language_model.layers.13.mlp.experts.114.gate_proj", "model.language_model.layers.13.mlp.experts.115.gate_proj", "model.language_model.layers.13.mlp.experts.116.gate_proj", "model.language_model.layers.13.mlp.experts.117.gate_proj", "model.language_model.layers.13.mlp.experts.118.gate_proj", "model.language_model.layers.13.mlp.experts.119.gate_proj", "model.language_model.layers.13.mlp.experts.120.gate_proj", "model.language_model.layers.13.mlp.experts.121.gate_proj", "model.language_model.layers.13.mlp.experts.122.gate_proj", "model.language_model.layers.13.mlp.experts.123.gate_proj", "model.language_model.layers.13.mlp.experts.124.gate_proj", "model.language_model.layers.13.mlp.experts.125.gate_proj", "model.language_model.layers.13.mlp.experts.126.gate_proj", "model.language_model.layers.13.mlp.experts.127.gate_proj", "model.language_model.layers.13.mlp.experts.0.up_proj", "model.language_model.layers.13.mlp.experts.1.up_proj", "model.language_model.layers.13.mlp.experts.2.up_proj", "model.language_model.layers.13.mlp.experts.3.up_proj", "model.language_model.layers.13.mlp.experts.4.up_proj", "model.language_model.layers.13.mlp.experts.5.up_proj", "model.language_model.layers.13.mlp.experts.6.up_proj", "model.language_model.layers.13.mlp.experts.7.up_proj", "model.language_model.layers.13.mlp.experts.8.up_proj", "model.language_model.layers.13.mlp.experts.9.up_proj", "model.language_model.layers.13.mlp.experts.10.up_proj", "model.language_model.layers.13.mlp.experts.11.up_proj", "model.language_model.layers.13.mlp.experts.12.up_proj", "model.language_model.layers.13.mlp.experts.13.up_proj", "model.language_model.layers.13.mlp.experts.14.up_proj", "model.language_model.layers.13.mlp.experts.15.up_proj", "model.language_model.layers.13.mlp.experts.16.up_proj", "model.language_model.layers.13.mlp.experts.17.up_proj", "model.language_model.layers.13.mlp.experts.18.up_proj", "model.language_model.layers.13.mlp.experts.19.up_proj", "model.language_model.layers.13.mlp.experts.20.up_proj", "model.language_model.layers.13.mlp.experts.21.up_proj", "model.language_model.layers.13.mlp.experts.22.up_proj", "model.language_model.layers.13.mlp.experts.23.up_proj", "model.language_model.layers.13.mlp.experts.24.up_proj", "model.language_model.layers.13.mlp.experts.25.up_proj", "model.language_model.layers.13.mlp.experts.26.up_proj", "model.language_model.layers.13.mlp.experts.27.up_proj", "model.language_model.layers.13.mlp.experts.28.up_proj", "model.language_model.layers.13.mlp.experts.29.up_proj", "model.language_model.layers.13.mlp.experts.30.up_proj", "model.language_model.layers.13.mlp.experts.31.up_proj", "model.language_model.layers.13.mlp.experts.32.up_proj", "model.language_model.layers.13.mlp.experts.33.up_proj", "model.language_model.layers.13.mlp.experts.34.up_proj", "model.language_model.layers.13.mlp.experts.35.up_proj", "model.language_model.layers.13.mlp.experts.36.up_proj", "model.language_model.layers.13.mlp.experts.37.up_proj", "model.language_model.layers.13.mlp.experts.38.up_proj", "model.language_model.layers.13.mlp.experts.39.up_proj", "model.language_model.layers.13.mlp.experts.40.up_proj", "model.language_model.layers.13.mlp.experts.41.up_proj", "model.language_model.layers.13.mlp.experts.42.up_proj", "model.language_model.layers.13.mlp.experts.43.up_proj", "model.language_model.layers.13.mlp.experts.44.up_proj", "model.language_model.layers.13.mlp.experts.45.up_proj", "model.language_model.layers.13.mlp.experts.46.up_proj", "model.language_model.layers.13.mlp.experts.47.up_proj", "model.language_model.layers.13.mlp.experts.48.up_proj", "model.language_model.layers.13.mlp.experts.49.up_proj", "model.language_model.layers.13.mlp.experts.50.up_proj", "model.language_model.layers.13.mlp.experts.51.up_proj", "model.language_model.layers.13.mlp.experts.52.up_proj", "model.language_model.layers.13.mlp.experts.53.up_proj", "model.language_model.layers.13.mlp.experts.54.up_proj", "model.language_model.layers.13.mlp.experts.55.up_proj", "model.language_model.layers.13.mlp.experts.56.up_proj", "model.language_model.layers.13.mlp.experts.57.up_proj", "model.language_model.layers.13.mlp.experts.58.up_proj", "model.language_model.layers.13.mlp.experts.59.up_proj", "model.language_model.layers.13.mlp.experts.60.up_proj", "model.language_model.layers.13.mlp.experts.61.up_proj", "model.language_model.layers.13.mlp.experts.62.up_proj", "model.language_model.layers.13.mlp.experts.63.up_proj", "model.language_model.layers.13.mlp.experts.64.up_proj", "model.language_model.layers.13.mlp.experts.65.up_proj", "model.language_model.layers.13.mlp.experts.66.up_proj", "model.language_model.layers.13.mlp.experts.67.up_proj", "model.language_model.layers.13.mlp.experts.68.up_proj", "model.language_model.layers.13.mlp.experts.69.up_proj", "model.language_model.layers.13.mlp.experts.70.up_proj", "model.language_model.layers.13.mlp.experts.71.up_proj", "model.language_model.layers.13.mlp.experts.72.up_proj", "model.language_model.layers.13.mlp.experts.73.up_proj", "model.language_model.layers.13.mlp.experts.74.up_proj", "model.language_model.layers.13.mlp.experts.75.up_proj", "model.language_model.layers.13.mlp.experts.76.up_proj", "model.language_model.layers.13.mlp.experts.77.up_proj", "model.language_model.layers.13.mlp.experts.78.up_proj", "model.language_model.layers.13.mlp.experts.79.up_proj", "model.language_model.layers.13.mlp.experts.80.up_proj", "model.language_model.layers.13.mlp.experts.81.up_proj", "model.language_model.layers.13.mlp.experts.82.up_proj", "model.language_model.layers.13.mlp.experts.83.up_proj", "model.language_model.layers.13.mlp.experts.84.up_proj", "model.language_model.layers.13.mlp.experts.85.up_proj", "model.language_model.layers.13.mlp.experts.86.up_proj", "model.language_model.layers.13.mlp.experts.87.up_proj", "model.language_model.layers.13.mlp.experts.88.up_proj", "model.language_model.layers.13.mlp.experts.89.up_proj", "model.language_model.layers.13.mlp.experts.90.up_proj", "model.language_model.layers.13.mlp.experts.91.up_proj", "model.language_model.layers.13.mlp.experts.92.up_proj", "model.language_model.layers.13.mlp.experts.93.up_proj", "model.language_model.layers.13.mlp.experts.94.up_proj", "model.language_model.layers.13.mlp.experts.95.up_proj", "model.language_model.layers.13.mlp.experts.96.up_proj", "model.language_model.layers.13.mlp.experts.97.up_proj", "model.language_model.layers.13.mlp.experts.98.up_proj", "model.language_model.layers.13.mlp.experts.99.up_proj", "model.language_model.layers.13.mlp.experts.100.up_proj", "model.language_model.layers.13.mlp.experts.101.up_proj", "model.language_model.layers.13.mlp.experts.102.up_proj", "model.language_model.layers.13.mlp.experts.103.up_proj", "model.language_model.layers.13.mlp.experts.104.up_proj", "model.language_model.layers.13.mlp.experts.105.up_proj", "model.language_model.layers.13.mlp.experts.106.up_proj", "model.language_model.layers.13.mlp.experts.107.up_proj", "model.language_model.layers.13.mlp.experts.108.up_proj", "model.language_model.layers.13.mlp.experts.109.up_proj", "model.language_model.layers.13.mlp.experts.110.up_proj", "model.language_model.layers.13.mlp.experts.111.up_proj", "model.language_model.layers.13.mlp.experts.112.up_proj", "model.language_model.layers.13.mlp.experts.113.up_proj", "model.language_model.layers.13.mlp.experts.114.up_proj", "model.language_model.layers.13.mlp.experts.115.up_proj", "model.language_model.layers.13.mlp.experts.116.up_proj", "model.language_model.layers.13.mlp.experts.117.up_proj", "model.language_model.layers.13.mlp.experts.118.up_proj", "model.language_model.layers.13.mlp.experts.119.up_proj", "model.language_model.layers.13.mlp.experts.120.up_proj", "model.language_model.layers.13.mlp.experts.121.up_proj", "model.language_model.layers.13.mlp.experts.122.up_proj", "model.language_model.layers.13.mlp.experts.123.up_proj", "model.language_model.layers.13.mlp.experts.124.up_proj", "model.language_model.layers.13.mlp.experts.125.up_proj", "model.language_model.layers.13.mlp.experts.126.up_proj", "model.language_model.layers.13.mlp.experts.127.up_proj", "model.language_model.layers.13.mlp.experts.0.down_proj", "model.language_model.layers.13.mlp.experts.1.down_proj", "model.language_model.layers.13.mlp.experts.2.down_proj", "model.language_model.layers.13.mlp.experts.3.down_proj", "model.language_model.layers.13.mlp.experts.4.down_proj", "model.language_model.layers.13.mlp.experts.5.down_proj", "model.language_model.layers.13.mlp.experts.6.down_proj", "model.language_model.layers.13.mlp.experts.7.down_proj", "model.language_model.layers.13.mlp.experts.8.down_proj", "model.language_model.layers.13.mlp.experts.9.down_proj", "model.language_model.layers.13.mlp.experts.10.down_proj", "model.language_model.layers.13.mlp.experts.11.down_proj", "model.language_model.layers.13.mlp.experts.12.down_proj", "model.language_model.layers.13.mlp.experts.13.down_proj", "model.language_model.layers.13.mlp.experts.14.down_proj", "model.language_model.layers.13.mlp.experts.15.down_proj", "model.language_model.layers.13.mlp.experts.16.down_proj", "model.language_model.layers.13.mlp.experts.17.down_proj", "model.language_model.layers.13.mlp.experts.18.down_proj", "model.language_model.layers.13.mlp.experts.19.down_proj", "model.language_model.layers.13.mlp.experts.20.down_proj", "model.language_model.layers.13.mlp.experts.21.down_proj", "model.language_model.layers.13.mlp.experts.22.down_proj", "model.language_model.layers.13.mlp.experts.23.down_proj", "model.language_model.layers.13.mlp.experts.24.down_proj", "model.language_model.layers.13.mlp.experts.25.down_proj", "model.language_model.layers.13.mlp.experts.26.down_proj", "model.language_model.layers.13.mlp.experts.27.down_proj", "model.language_model.layers.13.mlp.experts.28.down_proj", "model.language_model.layers.13.mlp.experts.29.down_proj", "model.language_model.layers.13.mlp.experts.30.down_proj", "model.language_model.layers.13.mlp.experts.31.down_proj", "model.language_model.layers.13.mlp.experts.32.down_proj", "model.language_model.layers.13.mlp.experts.33.down_proj", "model.language_model.layers.13.mlp.experts.34.down_proj", "model.language_model.layers.13.mlp.experts.35.down_proj", "model.language_model.layers.13.mlp.experts.36.down_proj", "model.language_model.layers.13.mlp.experts.37.down_proj", "model.language_model.layers.13.mlp.experts.38.down_proj", "model.language_model.layers.13.mlp.experts.39.down_proj", "model.language_model.layers.13.mlp.experts.40.down_proj", "model.language_model.layers.13.mlp.experts.41.down_proj", "model.language_model.layers.13.mlp.experts.42.down_proj", "model.language_model.layers.13.mlp.experts.43.down_proj", "model.language_model.layers.13.mlp.experts.44.down_proj", "model.language_model.layers.13.mlp.experts.45.down_proj", "model.language_model.layers.13.mlp.experts.46.down_proj", "model.language_model.layers.13.mlp.experts.47.down_proj", "model.language_model.layers.13.mlp.experts.48.down_proj", "model.language_model.layers.13.mlp.experts.49.down_proj", "model.language_model.layers.13.mlp.experts.50.down_proj", "model.language_model.layers.13.mlp.experts.51.down_proj", "model.language_model.layers.13.mlp.experts.52.down_proj", "model.language_model.layers.13.mlp.experts.53.down_proj", "model.language_model.layers.13.mlp.experts.54.down_proj", "model.language_model.layers.13.mlp.experts.55.down_proj", "model.language_model.layers.13.mlp.experts.56.down_proj", "model.language_model.layers.13.mlp.experts.57.down_proj", "model.language_model.layers.13.mlp.experts.58.down_proj", "model.language_model.layers.13.mlp.experts.59.down_proj", "model.language_model.layers.13.mlp.experts.60.down_proj", "model.language_model.layers.13.mlp.experts.61.down_proj", "model.language_model.layers.13.mlp.experts.62.down_proj", "model.language_model.layers.13.mlp.experts.63.down_proj", "model.language_model.layers.13.mlp.experts.64.down_proj", "model.language_model.layers.13.mlp.experts.65.down_proj", "model.language_model.layers.13.mlp.experts.66.down_proj", "model.language_model.layers.13.mlp.experts.67.down_proj", "model.language_model.layers.13.mlp.experts.68.down_proj", "model.language_model.layers.13.mlp.experts.69.down_proj", "model.language_model.layers.13.mlp.experts.70.down_proj", "model.language_model.layers.13.mlp.experts.71.down_proj", "model.language_model.layers.13.mlp.experts.72.down_proj", "model.language_model.layers.13.mlp.experts.73.down_proj", "model.language_model.layers.13.mlp.experts.74.down_proj", "model.language_model.layers.13.mlp.experts.75.down_proj", "model.language_model.layers.13.mlp.experts.76.down_proj", "model.language_model.layers.13.mlp.experts.77.down_proj", "model.language_model.layers.13.mlp.experts.78.down_proj", "model.language_model.layers.13.mlp.experts.79.down_proj", "model.language_model.layers.13.mlp.experts.80.down_proj", "model.language_model.layers.13.mlp.experts.81.down_proj", "model.language_model.layers.13.mlp.experts.82.down_proj", "model.language_model.layers.13.mlp.experts.83.down_proj", "model.language_model.layers.13.mlp.experts.84.down_proj", "model.language_model.layers.13.mlp.experts.85.down_proj", "model.language_model.layers.13.mlp.experts.86.down_proj", "model.language_model.layers.13.mlp.experts.87.down_proj", "model.language_model.layers.13.mlp.experts.88.down_proj", "model.language_model.layers.13.mlp.experts.89.down_proj", "model.language_model.layers.13.mlp.experts.90.down_proj", "model.language_model.layers.13.mlp.experts.91.down_proj", "model.language_model.layers.13.mlp.experts.92.down_proj", "model.language_model.layers.13.mlp.experts.93.down_proj", "model.language_model.layers.13.mlp.experts.94.down_proj", "model.language_model.layers.13.mlp.experts.95.down_proj", "model.language_model.layers.13.mlp.experts.96.down_proj", "model.language_model.layers.13.mlp.experts.97.down_proj", "model.language_model.layers.13.mlp.experts.98.down_proj", "model.language_model.layers.13.mlp.experts.99.down_proj", "model.language_model.layers.13.mlp.experts.100.down_proj", "model.language_model.layers.13.mlp.experts.101.down_proj", "model.language_model.layers.13.mlp.experts.102.down_proj", "model.language_model.layers.13.mlp.experts.103.down_proj", "model.language_model.layers.13.mlp.experts.104.down_proj", "model.language_model.layers.13.mlp.experts.105.down_proj", "model.language_model.layers.13.mlp.experts.106.down_proj", "model.language_model.layers.13.mlp.experts.107.down_proj", "model.language_model.layers.13.mlp.experts.108.down_proj", "model.language_model.layers.13.mlp.experts.109.down_proj", "model.language_model.layers.13.mlp.experts.110.down_proj", "model.language_model.layers.13.mlp.experts.111.down_proj", "model.language_model.layers.13.mlp.experts.112.down_proj", "model.language_model.layers.13.mlp.experts.113.down_proj", "model.language_model.layers.13.mlp.experts.114.down_proj", "model.language_model.layers.13.mlp.experts.115.down_proj", "model.language_model.layers.13.mlp.experts.116.down_proj", "model.language_model.layers.13.mlp.experts.117.down_proj", "model.language_model.layers.13.mlp.experts.118.down_proj", "model.language_model.layers.13.mlp.experts.119.down_proj", "model.language_model.layers.13.mlp.experts.120.down_proj", "model.language_model.layers.13.mlp.experts.121.down_proj", "model.language_model.layers.13.mlp.experts.122.down_proj", "model.language_model.layers.13.mlp.experts.123.down_proj", "model.language_model.layers.13.mlp.experts.124.down_proj", "model.language_model.layers.13.mlp.experts.125.down_proj", "model.language_model.layers.13.mlp.experts.126.down_proj", "model.language_model.layers.13.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.0014988730661571054, "dbits": 603979776 } ] }, { "idx": 28, "layers": [ "model.language_model.layers.14.self_attn.q_proj", "model.language_model.layers.14.self_attn.k_proj", "model.language_model.layers.14.self_attn.v_proj", "model.language_model.layers.14.self_attn.o_proj" ], "candidates": [ { "dkld": 0.001802235655486581, "dbits": 18874368 } ] }, { "idx": 29, "layers": [ "model.language_model.layers.14.mlp.experts.0.gate_proj", "model.language_model.layers.14.mlp.experts.1.gate_proj", "model.language_model.layers.14.mlp.experts.2.gate_proj", "model.language_model.layers.14.mlp.experts.3.gate_proj", "model.language_model.layers.14.mlp.experts.4.gate_proj", "model.language_model.layers.14.mlp.experts.5.gate_proj", "model.language_model.layers.14.mlp.experts.6.gate_proj", "model.language_model.layers.14.mlp.experts.7.gate_proj", "model.language_model.layers.14.mlp.experts.8.gate_proj", "model.language_model.layers.14.mlp.experts.9.gate_proj", "model.language_model.layers.14.mlp.experts.10.gate_proj", "model.language_model.layers.14.mlp.experts.11.gate_proj", "model.language_model.layers.14.mlp.experts.12.gate_proj", "model.language_model.layers.14.mlp.experts.13.gate_proj", "model.language_model.layers.14.mlp.experts.14.gate_proj", "model.language_model.layers.14.mlp.experts.15.gate_proj", "model.language_model.layers.14.mlp.experts.16.gate_proj", "model.language_model.layers.14.mlp.experts.17.gate_proj", "model.language_model.layers.14.mlp.experts.18.gate_proj", "model.language_model.layers.14.mlp.experts.19.gate_proj", "model.language_model.layers.14.mlp.experts.20.gate_proj", "model.language_model.layers.14.mlp.experts.21.gate_proj", "model.language_model.layers.14.mlp.experts.22.gate_proj", "model.language_model.layers.14.mlp.experts.23.gate_proj", "model.language_model.layers.14.mlp.experts.24.gate_proj", "model.language_model.layers.14.mlp.experts.25.gate_proj", "model.language_model.layers.14.mlp.experts.26.gate_proj", "model.language_model.layers.14.mlp.experts.27.gate_proj", "model.language_model.layers.14.mlp.experts.28.gate_proj", "model.language_model.layers.14.mlp.experts.29.gate_proj", "model.language_model.layers.14.mlp.experts.30.gate_proj", "model.language_model.layers.14.mlp.experts.31.gate_proj", "model.language_model.layers.14.mlp.experts.32.gate_proj", "model.language_model.layers.14.mlp.experts.33.gate_proj", "model.language_model.layers.14.mlp.experts.34.gate_proj", "model.language_model.layers.14.mlp.experts.35.gate_proj", "model.language_model.layers.14.mlp.experts.36.gate_proj", "model.language_model.layers.14.mlp.experts.37.gate_proj", "model.language_model.layers.14.mlp.experts.38.gate_proj", "model.language_model.layers.14.mlp.experts.39.gate_proj", "model.language_model.layers.14.mlp.experts.40.gate_proj", "model.language_model.layers.14.mlp.experts.41.gate_proj", "model.language_model.layers.14.mlp.experts.42.gate_proj", "model.language_model.layers.14.mlp.experts.43.gate_proj", "model.language_model.layers.14.mlp.experts.44.gate_proj", "model.language_model.layers.14.mlp.experts.45.gate_proj", "model.language_model.layers.14.mlp.experts.46.gate_proj", "model.language_model.layers.14.mlp.experts.47.gate_proj", "model.language_model.layers.14.mlp.experts.48.gate_proj", "model.language_model.layers.14.mlp.experts.49.gate_proj", "model.language_model.layers.14.mlp.experts.50.gate_proj", "model.language_model.layers.14.mlp.experts.51.gate_proj", "model.language_model.layers.14.mlp.experts.52.gate_proj", "model.language_model.layers.14.mlp.experts.53.gate_proj", "model.language_model.layers.14.mlp.experts.54.gate_proj", "model.language_model.layers.14.mlp.experts.55.gate_proj", "model.language_model.layers.14.mlp.experts.56.gate_proj", "model.language_model.layers.14.mlp.experts.57.gate_proj", "model.language_model.layers.14.mlp.experts.58.gate_proj", "model.language_model.layers.14.mlp.experts.59.gate_proj", "model.language_model.layers.14.mlp.experts.60.gate_proj", "model.language_model.layers.14.mlp.experts.61.gate_proj", "model.language_model.layers.14.mlp.experts.62.gate_proj", "model.language_model.layers.14.mlp.experts.63.gate_proj", "model.language_model.layers.14.mlp.experts.64.gate_proj", "model.language_model.layers.14.mlp.experts.65.gate_proj", "model.language_model.layers.14.mlp.experts.66.gate_proj", "model.language_model.layers.14.mlp.experts.67.gate_proj", "model.language_model.layers.14.mlp.experts.68.gate_proj", "model.language_model.layers.14.mlp.experts.69.gate_proj", "model.language_model.layers.14.mlp.experts.70.gate_proj", "model.language_model.layers.14.mlp.experts.71.gate_proj", "model.language_model.layers.14.mlp.experts.72.gate_proj", "model.language_model.layers.14.mlp.experts.73.gate_proj", "model.language_model.layers.14.mlp.experts.74.gate_proj", "model.language_model.layers.14.mlp.experts.75.gate_proj", "model.language_model.layers.14.mlp.experts.76.gate_proj", "model.language_model.layers.14.mlp.experts.77.gate_proj", "model.language_model.layers.14.mlp.experts.78.gate_proj", "model.language_model.layers.14.mlp.experts.79.gate_proj", "model.language_model.layers.14.mlp.experts.80.gate_proj", "model.language_model.layers.14.mlp.experts.81.gate_proj", "model.language_model.layers.14.mlp.experts.82.gate_proj", "model.language_model.layers.14.mlp.experts.83.gate_proj", "model.language_model.layers.14.mlp.experts.84.gate_proj", "model.language_model.layers.14.mlp.experts.85.gate_proj", "model.language_model.layers.14.mlp.experts.86.gate_proj", "model.language_model.layers.14.mlp.experts.87.gate_proj", "model.language_model.layers.14.mlp.experts.88.gate_proj", "model.language_model.layers.14.mlp.experts.89.gate_proj", "model.language_model.layers.14.mlp.experts.90.gate_proj", "model.language_model.layers.14.mlp.experts.91.gate_proj", "model.language_model.layers.14.mlp.experts.92.gate_proj", "model.language_model.layers.14.mlp.experts.93.gate_proj", "model.language_model.layers.14.mlp.experts.94.gate_proj", "model.language_model.layers.14.mlp.experts.95.gate_proj", "model.language_model.layers.14.mlp.experts.96.gate_proj", "model.language_model.layers.14.mlp.experts.97.gate_proj", "model.language_model.layers.14.mlp.experts.98.gate_proj", "model.language_model.layers.14.mlp.experts.99.gate_proj", "model.language_model.layers.14.mlp.experts.100.gate_proj", "model.language_model.layers.14.mlp.experts.101.gate_proj", "model.language_model.layers.14.mlp.experts.102.gate_proj", "model.language_model.layers.14.mlp.experts.103.gate_proj", "model.language_model.layers.14.mlp.experts.104.gate_proj", "model.language_model.layers.14.mlp.experts.105.gate_proj", "model.language_model.layers.14.mlp.experts.106.gate_proj", "model.language_model.layers.14.mlp.experts.107.gate_proj", "model.language_model.layers.14.mlp.experts.108.gate_proj", "model.language_model.layers.14.mlp.experts.109.gate_proj", "model.language_model.layers.14.mlp.experts.110.gate_proj", "model.language_model.layers.14.mlp.experts.111.gate_proj", "model.language_model.layers.14.mlp.experts.112.gate_proj", "model.language_model.layers.14.mlp.experts.113.gate_proj", "model.language_model.layers.14.mlp.experts.114.gate_proj", "model.language_model.layers.14.mlp.experts.115.gate_proj", "model.language_model.layers.14.mlp.experts.116.gate_proj", "model.language_model.layers.14.mlp.experts.117.gate_proj", "model.language_model.layers.14.mlp.experts.118.gate_proj", "model.language_model.layers.14.mlp.experts.119.gate_proj", "model.language_model.layers.14.mlp.experts.120.gate_proj", "model.language_model.layers.14.mlp.experts.121.gate_proj", "model.language_model.layers.14.mlp.experts.122.gate_proj", "model.language_model.layers.14.mlp.experts.123.gate_proj", "model.language_model.layers.14.mlp.experts.124.gate_proj", "model.language_model.layers.14.mlp.experts.125.gate_proj", "model.language_model.layers.14.mlp.experts.126.gate_proj", "model.language_model.layers.14.mlp.experts.127.gate_proj", "model.language_model.layers.14.mlp.experts.0.up_proj", "model.language_model.layers.14.mlp.experts.1.up_proj", "model.language_model.layers.14.mlp.experts.2.up_proj", "model.language_model.layers.14.mlp.experts.3.up_proj", "model.language_model.layers.14.mlp.experts.4.up_proj", "model.language_model.layers.14.mlp.experts.5.up_proj", "model.language_model.layers.14.mlp.experts.6.up_proj", "model.language_model.layers.14.mlp.experts.7.up_proj", "model.language_model.layers.14.mlp.experts.8.up_proj", "model.language_model.layers.14.mlp.experts.9.up_proj", "model.language_model.layers.14.mlp.experts.10.up_proj", "model.language_model.layers.14.mlp.experts.11.up_proj", "model.language_model.layers.14.mlp.experts.12.up_proj", "model.language_model.layers.14.mlp.experts.13.up_proj", "model.language_model.layers.14.mlp.experts.14.up_proj", "model.language_model.layers.14.mlp.experts.15.up_proj", "model.language_model.layers.14.mlp.experts.16.up_proj", "model.language_model.layers.14.mlp.experts.17.up_proj", "model.language_model.layers.14.mlp.experts.18.up_proj", "model.language_model.layers.14.mlp.experts.19.up_proj", "model.language_model.layers.14.mlp.experts.20.up_proj", "model.language_model.layers.14.mlp.experts.21.up_proj", "model.language_model.layers.14.mlp.experts.22.up_proj", "model.language_model.layers.14.mlp.experts.23.up_proj", "model.language_model.layers.14.mlp.experts.24.up_proj", "model.language_model.layers.14.mlp.experts.25.up_proj", "model.language_model.layers.14.mlp.experts.26.up_proj", "model.language_model.layers.14.mlp.experts.27.up_proj", "model.language_model.layers.14.mlp.experts.28.up_proj", "model.language_model.layers.14.mlp.experts.29.up_proj", "model.language_model.layers.14.mlp.experts.30.up_proj", "model.language_model.layers.14.mlp.experts.31.up_proj", "model.language_model.layers.14.mlp.experts.32.up_proj", "model.language_model.layers.14.mlp.experts.33.up_proj", "model.language_model.layers.14.mlp.experts.34.up_proj", "model.language_model.layers.14.mlp.experts.35.up_proj", "model.language_model.layers.14.mlp.experts.36.up_proj", "model.language_model.layers.14.mlp.experts.37.up_proj", "model.language_model.layers.14.mlp.experts.38.up_proj", "model.language_model.layers.14.mlp.experts.39.up_proj", "model.language_model.layers.14.mlp.experts.40.up_proj", "model.language_model.layers.14.mlp.experts.41.up_proj", "model.language_model.layers.14.mlp.experts.42.up_proj", "model.language_model.layers.14.mlp.experts.43.up_proj", "model.language_model.layers.14.mlp.experts.44.up_proj", "model.language_model.layers.14.mlp.experts.45.up_proj", "model.language_model.layers.14.mlp.experts.46.up_proj", "model.language_model.layers.14.mlp.experts.47.up_proj", "model.language_model.layers.14.mlp.experts.48.up_proj", "model.language_model.layers.14.mlp.experts.49.up_proj", "model.language_model.layers.14.mlp.experts.50.up_proj", "model.language_model.layers.14.mlp.experts.51.up_proj", "model.language_model.layers.14.mlp.experts.52.up_proj", "model.language_model.layers.14.mlp.experts.53.up_proj", "model.language_model.layers.14.mlp.experts.54.up_proj", "model.language_model.layers.14.mlp.experts.55.up_proj", "model.language_model.layers.14.mlp.experts.56.up_proj", "model.language_model.layers.14.mlp.experts.57.up_proj", "model.language_model.layers.14.mlp.experts.58.up_proj", "model.language_model.layers.14.mlp.experts.59.up_proj", "model.language_model.layers.14.mlp.experts.60.up_proj", "model.language_model.layers.14.mlp.experts.61.up_proj", "model.language_model.layers.14.mlp.experts.62.up_proj", "model.language_model.layers.14.mlp.experts.63.up_proj", "model.language_model.layers.14.mlp.experts.64.up_proj", "model.language_model.layers.14.mlp.experts.65.up_proj", "model.language_model.layers.14.mlp.experts.66.up_proj", "model.language_model.layers.14.mlp.experts.67.up_proj", "model.language_model.layers.14.mlp.experts.68.up_proj", "model.language_model.layers.14.mlp.experts.69.up_proj", "model.language_model.layers.14.mlp.experts.70.up_proj", "model.language_model.layers.14.mlp.experts.71.up_proj", "model.language_model.layers.14.mlp.experts.72.up_proj", "model.language_model.layers.14.mlp.experts.73.up_proj", "model.language_model.layers.14.mlp.experts.74.up_proj", "model.language_model.layers.14.mlp.experts.75.up_proj", "model.language_model.layers.14.mlp.experts.76.up_proj", "model.language_model.layers.14.mlp.experts.77.up_proj", "model.language_model.layers.14.mlp.experts.78.up_proj", "model.language_model.layers.14.mlp.experts.79.up_proj", "model.language_model.layers.14.mlp.experts.80.up_proj", "model.language_model.layers.14.mlp.experts.81.up_proj", "model.language_model.layers.14.mlp.experts.82.up_proj", "model.language_model.layers.14.mlp.experts.83.up_proj", "model.language_model.layers.14.mlp.experts.84.up_proj", "model.language_model.layers.14.mlp.experts.85.up_proj", "model.language_model.layers.14.mlp.experts.86.up_proj", "model.language_model.layers.14.mlp.experts.87.up_proj", "model.language_model.layers.14.mlp.experts.88.up_proj", "model.language_model.layers.14.mlp.experts.89.up_proj", "model.language_model.layers.14.mlp.experts.90.up_proj", "model.language_model.layers.14.mlp.experts.91.up_proj", "model.language_model.layers.14.mlp.experts.92.up_proj", "model.language_model.layers.14.mlp.experts.93.up_proj", "model.language_model.layers.14.mlp.experts.94.up_proj", "model.language_model.layers.14.mlp.experts.95.up_proj", "model.language_model.layers.14.mlp.experts.96.up_proj", "model.language_model.layers.14.mlp.experts.97.up_proj", "model.language_model.layers.14.mlp.experts.98.up_proj", "model.language_model.layers.14.mlp.experts.99.up_proj", "model.language_model.layers.14.mlp.experts.100.up_proj", "model.language_model.layers.14.mlp.experts.101.up_proj", "model.language_model.layers.14.mlp.experts.102.up_proj", "model.language_model.layers.14.mlp.experts.103.up_proj", "model.language_model.layers.14.mlp.experts.104.up_proj", "model.language_model.layers.14.mlp.experts.105.up_proj", "model.language_model.layers.14.mlp.experts.106.up_proj", "model.language_model.layers.14.mlp.experts.107.up_proj", "model.language_model.layers.14.mlp.experts.108.up_proj", "model.language_model.layers.14.mlp.experts.109.up_proj", "model.language_model.layers.14.mlp.experts.110.up_proj", "model.language_model.layers.14.mlp.experts.111.up_proj", "model.language_model.layers.14.mlp.experts.112.up_proj", "model.language_model.layers.14.mlp.experts.113.up_proj", "model.language_model.layers.14.mlp.experts.114.up_proj", "model.language_model.layers.14.mlp.experts.115.up_proj", "model.language_model.layers.14.mlp.experts.116.up_proj", "model.language_model.layers.14.mlp.experts.117.up_proj", "model.language_model.layers.14.mlp.experts.118.up_proj", "model.language_model.layers.14.mlp.experts.119.up_proj", "model.language_model.layers.14.mlp.experts.120.up_proj", "model.language_model.layers.14.mlp.experts.121.up_proj", "model.language_model.layers.14.mlp.experts.122.up_proj", "model.language_model.layers.14.mlp.experts.123.up_proj", "model.language_model.layers.14.mlp.experts.124.up_proj", "model.language_model.layers.14.mlp.experts.125.up_proj", "model.language_model.layers.14.mlp.experts.126.up_proj", "model.language_model.layers.14.mlp.experts.127.up_proj", "model.language_model.layers.14.mlp.experts.0.down_proj", "model.language_model.layers.14.mlp.experts.1.down_proj", "model.language_model.layers.14.mlp.experts.2.down_proj", "model.language_model.layers.14.mlp.experts.3.down_proj", "model.language_model.layers.14.mlp.experts.4.down_proj", "model.language_model.layers.14.mlp.experts.5.down_proj", "model.language_model.layers.14.mlp.experts.6.down_proj", "model.language_model.layers.14.mlp.experts.7.down_proj", "model.language_model.layers.14.mlp.experts.8.down_proj", "model.language_model.layers.14.mlp.experts.9.down_proj", "model.language_model.layers.14.mlp.experts.10.down_proj", "model.language_model.layers.14.mlp.experts.11.down_proj", "model.language_model.layers.14.mlp.experts.12.down_proj", "model.language_model.layers.14.mlp.experts.13.down_proj", "model.language_model.layers.14.mlp.experts.14.down_proj", "model.language_model.layers.14.mlp.experts.15.down_proj", "model.language_model.layers.14.mlp.experts.16.down_proj", "model.language_model.layers.14.mlp.experts.17.down_proj", "model.language_model.layers.14.mlp.experts.18.down_proj", "model.language_model.layers.14.mlp.experts.19.down_proj", "model.language_model.layers.14.mlp.experts.20.down_proj", "model.language_model.layers.14.mlp.experts.21.down_proj", "model.language_model.layers.14.mlp.experts.22.down_proj", "model.language_model.layers.14.mlp.experts.23.down_proj", "model.language_model.layers.14.mlp.experts.24.down_proj", "model.language_model.layers.14.mlp.experts.25.down_proj", "model.language_model.layers.14.mlp.experts.26.down_proj", "model.language_model.layers.14.mlp.experts.27.down_proj", "model.language_model.layers.14.mlp.experts.28.down_proj", "model.language_model.layers.14.mlp.experts.29.down_proj", "model.language_model.layers.14.mlp.experts.30.down_proj", "model.language_model.layers.14.mlp.experts.31.down_proj", "model.language_model.layers.14.mlp.experts.32.down_proj", "model.language_model.layers.14.mlp.experts.33.down_proj", "model.language_model.layers.14.mlp.experts.34.down_proj", "model.language_model.layers.14.mlp.experts.35.down_proj", "model.language_model.layers.14.mlp.experts.36.down_proj", "model.language_model.layers.14.mlp.experts.37.down_proj", "model.language_model.layers.14.mlp.experts.38.down_proj", "model.language_model.layers.14.mlp.experts.39.down_proj", "model.language_model.layers.14.mlp.experts.40.down_proj", "model.language_model.layers.14.mlp.experts.41.down_proj", "model.language_model.layers.14.mlp.experts.42.down_proj", "model.language_model.layers.14.mlp.experts.43.down_proj", "model.language_model.layers.14.mlp.experts.44.down_proj", "model.language_model.layers.14.mlp.experts.45.down_proj", "model.language_model.layers.14.mlp.experts.46.down_proj", "model.language_model.layers.14.mlp.experts.47.down_proj", "model.language_model.layers.14.mlp.experts.48.down_proj", "model.language_model.layers.14.mlp.experts.49.down_proj", "model.language_model.layers.14.mlp.experts.50.down_proj", "model.language_model.layers.14.mlp.experts.51.down_proj", "model.language_model.layers.14.mlp.experts.52.down_proj", "model.language_model.layers.14.mlp.experts.53.down_proj", "model.language_model.layers.14.mlp.experts.54.down_proj", "model.language_model.layers.14.mlp.experts.55.down_proj", "model.language_model.layers.14.mlp.experts.56.down_proj", "model.language_model.layers.14.mlp.experts.57.down_proj", "model.language_model.layers.14.mlp.experts.58.down_proj", "model.language_model.layers.14.mlp.experts.59.down_proj", "model.language_model.layers.14.mlp.experts.60.down_proj", "model.language_model.layers.14.mlp.experts.61.down_proj", "model.language_model.layers.14.mlp.experts.62.down_proj", "model.language_model.layers.14.mlp.experts.63.down_proj", "model.language_model.layers.14.mlp.experts.64.down_proj", "model.language_model.layers.14.mlp.experts.65.down_proj", "model.language_model.layers.14.mlp.experts.66.down_proj", "model.language_model.layers.14.mlp.experts.67.down_proj", "model.language_model.layers.14.mlp.experts.68.down_proj", "model.language_model.layers.14.mlp.experts.69.down_proj", "model.language_model.layers.14.mlp.experts.70.down_proj", "model.language_model.layers.14.mlp.experts.71.down_proj", "model.language_model.layers.14.mlp.experts.72.down_proj", "model.language_model.layers.14.mlp.experts.73.down_proj", "model.language_model.layers.14.mlp.experts.74.down_proj", "model.language_model.layers.14.mlp.experts.75.down_proj", "model.language_model.layers.14.mlp.experts.76.down_proj", "model.language_model.layers.14.mlp.experts.77.down_proj", "model.language_model.layers.14.mlp.experts.78.down_proj", "model.language_model.layers.14.mlp.experts.79.down_proj", "model.language_model.layers.14.mlp.experts.80.down_proj", "model.language_model.layers.14.mlp.experts.81.down_proj", "model.language_model.layers.14.mlp.experts.82.down_proj", "model.language_model.layers.14.mlp.experts.83.down_proj", "model.language_model.layers.14.mlp.experts.84.down_proj", "model.language_model.layers.14.mlp.experts.85.down_proj", "model.language_model.layers.14.mlp.experts.86.down_proj", "model.language_model.layers.14.mlp.experts.87.down_proj", "model.language_model.layers.14.mlp.experts.88.down_proj", "model.language_model.layers.14.mlp.experts.89.down_proj", "model.language_model.layers.14.mlp.experts.90.down_proj", "model.language_model.layers.14.mlp.experts.91.down_proj", "model.language_model.layers.14.mlp.experts.92.down_proj", "model.language_model.layers.14.mlp.experts.93.down_proj", "model.language_model.layers.14.mlp.experts.94.down_proj", "model.language_model.layers.14.mlp.experts.95.down_proj", "model.language_model.layers.14.mlp.experts.96.down_proj", "model.language_model.layers.14.mlp.experts.97.down_proj", "model.language_model.layers.14.mlp.experts.98.down_proj", "model.language_model.layers.14.mlp.experts.99.down_proj", "model.language_model.layers.14.mlp.experts.100.down_proj", "model.language_model.layers.14.mlp.experts.101.down_proj", "model.language_model.layers.14.mlp.experts.102.down_proj", "model.language_model.layers.14.mlp.experts.103.down_proj", "model.language_model.layers.14.mlp.experts.104.down_proj", "model.language_model.layers.14.mlp.experts.105.down_proj", "model.language_model.layers.14.mlp.experts.106.down_proj", "model.language_model.layers.14.mlp.experts.107.down_proj", "model.language_model.layers.14.mlp.experts.108.down_proj", "model.language_model.layers.14.mlp.experts.109.down_proj", "model.language_model.layers.14.mlp.experts.110.down_proj", "model.language_model.layers.14.mlp.experts.111.down_proj", "model.language_model.layers.14.mlp.experts.112.down_proj", "model.language_model.layers.14.mlp.experts.113.down_proj", "model.language_model.layers.14.mlp.experts.114.down_proj", "model.language_model.layers.14.mlp.experts.115.down_proj", "model.language_model.layers.14.mlp.experts.116.down_proj", "model.language_model.layers.14.mlp.experts.117.down_proj", "model.language_model.layers.14.mlp.experts.118.down_proj", "model.language_model.layers.14.mlp.experts.119.down_proj", "model.language_model.layers.14.mlp.experts.120.down_proj", "model.language_model.layers.14.mlp.experts.121.down_proj", "model.language_model.layers.14.mlp.experts.122.down_proj", "model.language_model.layers.14.mlp.experts.123.down_proj", "model.language_model.layers.14.mlp.experts.124.down_proj", "model.language_model.layers.14.mlp.experts.125.down_proj", "model.language_model.layers.14.mlp.experts.126.down_proj", "model.language_model.layers.14.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.0013290055096149417, "dbits": 603979776 } ] }, { "idx": 30, "layers": [ "model.language_model.layers.15.self_attn.q_proj", "model.language_model.layers.15.self_attn.k_proj", "model.language_model.layers.15.self_attn.v_proj", "model.language_model.layers.15.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0003152049146592645, "dbits": 18874368 } ] }, { "idx": 31, "layers": [ "model.language_model.layers.15.mlp.experts.0.gate_proj", "model.language_model.layers.15.mlp.experts.1.gate_proj", "model.language_model.layers.15.mlp.experts.2.gate_proj", "model.language_model.layers.15.mlp.experts.3.gate_proj", "model.language_model.layers.15.mlp.experts.4.gate_proj", "model.language_model.layers.15.mlp.experts.5.gate_proj", "model.language_model.layers.15.mlp.experts.6.gate_proj", "model.language_model.layers.15.mlp.experts.7.gate_proj", "model.language_model.layers.15.mlp.experts.8.gate_proj", "model.language_model.layers.15.mlp.experts.9.gate_proj", "model.language_model.layers.15.mlp.experts.10.gate_proj", "model.language_model.layers.15.mlp.experts.11.gate_proj", "model.language_model.layers.15.mlp.experts.12.gate_proj", "model.language_model.layers.15.mlp.experts.13.gate_proj", "model.language_model.layers.15.mlp.experts.14.gate_proj", "model.language_model.layers.15.mlp.experts.15.gate_proj", "model.language_model.layers.15.mlp.experts.16.gate_proj", "model.language_model.layers.15.mlp.experts.17.gate_proj", "model.language_model.layers.15.mlp.experts.18.gate_proj", "model.language_model.layers.15.mlp.experts.19.gate_proj", "model.language_model.layers.15.mlp.experts.20.gate_proj", "model.language_model.layers.15.mlp.experts.21.gate_proj", "model.language_model.layers.15.mlp.experts.22.gate_proj", "model.language_model.layers.15.mlp.experts.23.gate_proj", "model.language_model.layers.15.mlp.experts.24.gate_proj", "model.language_model.layers.15.mlp.experts.25.gate_proj", "model.language_model.layers.15.mlp.experts.26.gate_proj", "model.language_model.layers.15.mlp.experts.27.gate_proj", "model.language_model.layers.15.mlp.experts.28.gate_proj", "model.language_model.layers.15.mlp.experts.29.gate_proj", "model.language_model.layers.15.mlp.experts.30.gate_proj", "model.language_model.layers.15.mlp.experts.31.gate_proj", "model.language_model.layers.15.mlp.experts.32.gate_proj", "model.language_model.layers.15.mlp.experts.33.gate_proj", "model.language_model.layers.15.mlp.experts.34.gate_proj", "model.language_model.layers.15.mlp.experts.35.gate_proj", "model.language_model.layers.15.mlp.experts.36.gate_proj", "model.language_model.layers.15.mlp.experts.37.gate_proj", "model.language_model.layers.15.mlp.experts.38.gate_proj", "model.language_model.layers.15.mlp.experts.39.gate_proj", "model.language_model.layers.15.mlp.experts.40.gate_proj", "model.language_model.layers.15.mlp.experts.41.gate_proj", "model.language_model.layers.15.mlp.experts.42.gate_proj", "model.language_model.layers.15.mlp.experts.43.gate_proj", "model.language_model.layers.15.mlp.experts.44.gate_proj", "model.language_model.layers.15.mlp.experts.45.gate_proj", "model.language_model.layers.15.mlp.experts.46.gate_proj", "model.language_model.layers.15.mlp.experts.47.gate_proj", "model.language_model.layers.15.mlp.experts.48.gate_proj", "model.language_model.layers.15.mlp.experts.49.gate_proj", "model.language_model.layers.15.mlp.experts.50.gate_proj", "model.language_model.layers.15.mlp.experts.51.gate_proj", "model.language_model.layers.15.mlp.experts.52.gate_proj", "model.language_model.layers.15.mlp.experts.53.gate_proj", "model.language_model.layers.15.mlp.experts.54.gate_proj", "model.language_model.layers.15.mlp.experts.55.gate_proj", "model.language_model.layers.15.mlp.experts.56.gate_proj", "model.language_model.layers.15.mlp.experts.57.gate_proj", "model.language_model.layers.15.mlp.experts.58.gate_proj", "model.language_model.layers.15.mlp.experts.59.gate_proj", "model.language_model.layers.15.mlp.experts.60.gate_proj", "model.language_model.layers.15.mlp.experts.61.gate_proj", "model.language_model.layers.15.mlp.experts.62.gate_proj", "model.language_model.layers.15.mlp.experts.63.gate_proj", "model.language_model.layers.15.mlp.experts.64.gate_proj", "model.language_model.layers.15.mlp.experts.65.gate_proj", "model.language_model.layers.15.mlp.experts.66.gate_proj", "model.language_model.layers.15.mlp.experts.67.gate_proj", "model.language_model.layers.15.mlp.experts.68.gate_proj", "model.language_model.layers.15.mlp.experts.69.gate_proj", "model.language_model.layers.15.mlp.experts.70.gate_proj", "model.language_model.layers.15.mlp.experts.71.gate_proj", "model.language_model.layers.15.mlp.experts.72.gate_proj", "model.language_model.layers.15.mlp.experts.73.gate_proj", "model.language_model.layers.15.mlp.experts.74.gate_proj", "model.language_model.layers.15.mlp.experts.75.gate_proj", "model.language_model.layers.15.mlp.experts.76.gate_proj", "model.language_model.layers.15.mlp.experts.77.gate_proj", "model.language_model.layers.15.mlp.experts.78.gate_proj", "model.language_model.layers.15.mlp.experts.79.gate_proj", "model.language_model.layers.15.mlp.experts.80.gate_proj", "model.language_model.layers.15.mlp.experts.81.gate_proj", "model.language_model.layers.15.mlp.experts.82.gate_proj", "model.language_model.layers.15.mlp.experts.83.gate_proj", "model.language_model.layers.15.mlp.experts.84.gate_proj", "model.language_model.layers.15.mlp.experts.85.gate_proj", "model.language_model.layers.15.mlp.experts.86.gate_proj", "model.language_model.layers.15.mlp.experts.87.gate_proj", "model.language_model.layers.15.mlp.experts.88.gate_proj", "model.language_model.layers.15.mlp.experts.89.gate_proj", "model.language_model.layers.15.mlp.experts.90.gate_proj", "model.language_model.layers.15.mlp.experts.91.gate_proj", "model.language_model.layers.15.mlp.experts.92.gate_proj", "model.language_model.layers.15.mlp.experts.93.gate_proj", "model.language_model.layers.15.mlp.experts.94.gate_proj", "model.language_model.layers.15.mlp.experts.95.gate_proj", "model.language_model.layers.15.mlp.experts.96.gate_proj", "model.language_model.layers.15.mlp.experts.97.gate_proj", "model.language_model.layers.15.mlp.experts.98.gate_proj", "model.language_model.layers.15.mlp.experts.99.gate_proj", "model.language_model.layers.15.mlp.experts.100.gate_proj", "model.language_model.layers.15.mlp.experts.101.gate_proj", "model.language_model.layers.15.mlp.experts.102.gate_proj", "model.language_model.layers.15.mlp.experts.103.gate_proj", "model.language_model.layers.15.mlp.experts.104.gate_proj", "model.language_model.layers.15.mlp.experts.105.gate_proj", "model.language_model.layers.15.mlp.experts.106.gate_proj", "model.language_model.layers.15.mlp.experts.107.gate_proj", "model.language_model.layers.15.mlp.experts.108.gate_proj", "model.language_model.layers.15.mlp.experts.109.gate_proj", "model.language_model.layers.15.mlp.experts.110.gate_proj", "model.language_model.layers.15.mlp.experts.111.gate_proj", "model.language_model.layers.15.mlp.experts.112.gate_proj", "model.language_model.layers.15.mlp.experts.113.gate_proj", "model.language_model.layers.15.mlp.experts.114.gate_proj", "model.language_model.layers.15.mlp.experts.115.gate_proj", "model.language_model.layers.15.mlp.experts.116.gate_proj", "model.language_model.layers.15.mlp.experts.117.gate_proj", "model.language_model.layers.15.mlp.experts.118.gate_proj", "model.language_model.layers.15.mlp.experts.119.gate_proj", "model.language_model.layers.15.mlp.experts.120.gate_proj", "model.language_model.layers.15.mlp.experts.121.gate_proj", "model.language_model.layers.15.mlp.experts.122.gate_proj", "model.language_model.layers.15.mlp.experts.123.gate_proj", "model.language_model.layers.15.mlp.experts.124.gate_proj", "model.language_model.layers.15.mlp.experts.125.gate_proj", "model.language_model.layers.15.mlp.experts.126.gate_proj", "model.language_model.layers.15.mlp.experts.127.gate_proj", "model.language_model.layers.15.mlp.experts.0.up_proj", "model.language_model.layers.15.mlp.experts.1.up_proj", "model.language_model.layers.15.mlp.experts.2.up_proj", "model.language_model.layers.15.mlp.experts.3.up_proj", "model.language_model.layers.15.mlp.experts.4.up_proj", "model.language_model.layers.15.mlp.experts.5.up_proj", "model.language_model.layers.15.mlp.experts.6.up_proj", "model.language_model.layers.15.mlp.experts.7.up_proj", "model.language_model.layers.15.mlp.experts.8.up_proj", "model.language_model.layers.15.mlp.experts.9.up_proj", "model.language_model.layers.15.mlp.experts.10.up_proj", "model.language_model.layers.15.mlp.experts.11.up_proj", "model.language_model.layers.15.mlp.experts.12.up_proj", "model.language_model.layers.15.mlp.experts.13.up_proj", "model.language_model.layers.15.mlp.experts.14.up_proj", "model.language_model.layers.15.mlp.experts.15.up_proj", "model.language_model.layers.15.mlp.experts.16.up_proj", "model.language_model.layers.15.mlp.experts.17.up_proj", "model.language_model.layers.15.mlp.experts.18.up_proj", "model.language_model.layers.15.mlp.experts.19.up_proj", "model.language_model.layers.15.mlp.experts.20.up_proj", "model.language_model.layers.15.mlp.experts.21.up_proj", "model.language_model.layers.15.mlp.experts.22.up_proj", "model.language_model.layers.15.mlp.experts.23.up_proj", "model.language_model.layers.15.mlp.experts.24.up_proj", "model.language_model.layers.15.mlp.experts.25.up_proj", "model.language_model.layers.15.mlp.experts.26.up_proj", "model.language_model.layers.15.mlp.experts.27.up_proj", "model.language_model.layers.15.mlp.experts.28.up_proj", "model.language_model.layers.15.mlp.experts.29.up_proj", "model.language_model.layers.15.mlp.experts.30.up_proj", "model.language_model.layers.15.mlp.experts.31.up_proj", "model.language_model.layers.15.mlp.experts.32.up_proj", "model.language_model.layers.15.mlp.experts.33.up_proj", "model.language_model.layers.15.mlp.experts.34.up_proj", "model.language_model.layers.15.mlp.experts.35.up_proj", "model.language_model.layers.15.mlp.experts.36.up_proj", "model.language_model.layers.15.mlp.experts.37.up_proj", "model.language_model.layers.15.mlp.experts.38.up_proj", "model.language_model.layers.15.mlp.experts.39.up_proj", "model.language_model.layers.15.mlp.experts.40.up_proj", "model.language_model.layers.15.mlp.experts.41.up_proj", "model.language_model.layers.15.mlp.experts.42.up_proj", "model.language_model.layers.15.mlp.experts.43.up_proj", "model.language_model.layers.15.mlp.experts.44.up_proj", "model.language_model.layers.15.mlp.experts.45.up_proj", "model.language_model.layers.15.mlp.experts.46.up_proj", "model.language_model.layers.15.mlp.experts.47.up_proj", "model.language_model.layers.15.mlp.experts.48.up_proj", "model.language_model.layers.15.mlp.experts.49.up_proj", "model.language_model.layers.15.mlp.experts.50.up_proj", "model.language_model.layers.15.mlp.experts.51.up_proj", "model.language_model.layers.15.mlp.experts.52.up_proj", "model.language_model.layers.15.mlp.experts.53.up_proj", "model.language_model.layers.15.mlp.experts.54.up_proj", "model.language_model.layers.15.mlp.experts.55.up_proj", "model.language_model.layers.15.mlp.experts.56.up_proj", "model.language_model.layers.15.mlp.experts.57.up_proj", "model.language_model.layers.15.mlp.experts.58.up_proj", "model.language_model.layers.15.mlp.experts.59.up_proj", "model.language_model.layers.15.mlp.experts.60.up_proj", "model.language_model.layers.15.mlp.experts.61.up_proj", "model.language_model.layers.15.mlp.experts.62.up_proj", "model.language_model.layers.15.mlp.experts.63.up_proj", "model.language_model.layers.15.mlp.experts.64.up_proj", "model.language_model.layers.15.mlp.experts.65.up_proj", "model.language_model.layers.15.mlp.experts.66.up_proj", "model.language_model.layers.15.mlp.experts.67.up_proj", "model.language_model.layers.15.mlp.experts.68.up_proj", "model.language_model.layers.15.mlp.experts.69.up_proj", "model.language_model.layers.15.mlp.experts.70.up_proj", "model.language_model.layers.15.mlp.experts.71.up_proj", "model.language_model.layers.15.mlp.experts.72.up_proj", "model.language_model.layers.15.mlp.experts.73.up_proj", "model.language_model.layers.15.mlp.experts.74.up_proj", "model.language_model.layers.15.mlp.experts.75.up_proj", "model.language_model.layers.15.mlp.experts.76.up_proj", "model.language_model.layers.15.mlp.experts.77.up_proj", "model.language_model.layers.15.mlp.experts.78.up_proj", "model.language_model.layers.15.mlp.experts.79.up_proj", "model.language_model.layers.15.mlp.experts.80.up_proj", "model.language_model.layers.15.mlp.experts.81.up_proj", "model.language_model.layers.15.mlp.experts.82.up_proj", "model.language_model.layers.15.mlp.experts.83.up_proj", "model.language_model.layers.15.mlp.experts.84.up_proj", "model.language_model.layers.15.mlp.experts.85.up_proj", "model.language_model.layers.15.mlp.experts.86.up_proj", "model.language_model.layers.15.mlp.experts.87.up_proj", "model.language_model.layers.15.mlp.experts.88.up_proj", "model.language_model.layers.15.mlp.experts.89.up_proj", "model.language_model.layers.15.mlp.experts.90.up_proj", "model.language_model.layers.15.mlp.experts.91.up_proj", "model.language_model.layers.15.mlp.experts.92.up_proj", "model.language_model.layers.15.mlp.experts.93.up_proj", "model.language_model.layers.15.mlp.experts.94.up_proj", "model.language_model.layers.15.mlp.experts.95.up_proj", "model.language_model.layers.15.mlp.experts.96.up_proj", "model.language_model.layers.15.mlp.experts.97.up_proj", "model.language_model.layers.15.mlp.experts.98.up_proj", "model.language_model.layers.15.mlp.experts.99.up_proj", "model.language_model.layers.15.mlp.experts.100.up_proj", "model.language_model.layers.15.mlp.experts.101.up_proj", "model.language_model.layers.15.mlp.experts.102.up_proj", "model.language_model.layers.15.mlp.experts.103.up_proj", "model.language_model.layers.15.mlp.experts.104.up_proj", "model.language_model.layers.15.mlp.experts.105.up_proj", "model.language_model.layers.15.mlp.experts.106.up_proj", "model.language_model.layers.15.mlp.experts.107.up_proj", "model.language_model.layers.15.mlp.experts.108.up_proj", "model.language_model.layers.15.mlp.experts.109.up_proj", "model.language_model.layers.15.mlp.experts.110.up_proj", "model.language_model.layers.15.mlp.experts.111.up_proj", "model.language_model.layers.15.mlp.experts.112.up_proj", "model.language_model.layers.15.mlp.experts.113.up_proj", "model.language_model.layers.15.mlp.experts.114.up_proj", "model.language_model.layers.15.mlp.experts.115.up_proj", "model.language_model.layers.15.mlp.experts.116.up_proj", "model.language_model.layers.15.mlp.experts.117.up_proj", "model.language_model.layers.15.mlp.experts.118.up_proj", "model.language_model.layers.15.mlp.experts.119.up_proj", "model.language_model.layers.15.mlp.experts.120.up_proj", "model.language_model.layers.15.mlp.experts.121.up_proj", "model.language_model.layers.15.mlp.experts.122.up_proj", "model.language_model.layers.15.mlp.experts.123.up_proj", "model.language_model.layers.15.mlp.experts.124.up_proj", "model.language_model.layers.15.mlp.experts.125.up_proj", "model.language_model.layers.15.mlp.experts.126.up_proj", "model.language_model.layers.15.mlp.experts.127.up_proj", "model.language_model.layers.15.mlp.experts.0.down_proj", "model.language_model.layers.15.mlp.experts.1.down_proj", "model.language_model.layers.15.mlp.experts.2.down_proj", "model.language_model.layers.15.mlp.experts.3.down_proj", "model.language_model.layers.15.mlp.experts.4.down_proj", "model.language_model.layers.15.mlp.experts.5.down_proj", "model.language_model.layers.15.mlp.experts.6.down_proj", "model.language_model.layers.15.mlp.experts.7.down_proj", "model.language_model.layers.15.mlp.experts.8.down_proj", "model.language_model.layers.15.mlp.experts.9.down_proj", "model.language_model.layers.15.mlp.experts.10.down_proj", "model.language_model.layers.15.mlp.experts.11.down_proj", "model.language_model.layers.15.mlp.experts.12.down_proj", "model.language_model.layers.15.mlp.experts.13.down_proj", "model.language_model.layers.15.mlp.experts.14.down_proj", "model.language_model.layers.15.mlp.experts.15.down_proj", "model.language_model.layers.15.mlp.experts.16.down_proj", "model.language_model.layers.15.mlp.experts.17.down_proj", "model.language_model.layers.15.mlp.experts.18.down_proj", "model.language_model.layers.15.mlp.experts.19.down_proj", "model.language_model.layers.15.mlp.experts.20.down_proj", "model.language_model.layers.15.mlp.experts.21.down_proj", "model.language_model.layers.15.mlp.experts.22.down_proj", "model.language_model.layers.15.mlp.experts.23.down_proj", "model.language_model.layers.15.mlp.experts.24.down_proj", "model.language_model.layers.15.mlp.experts.25.down_proj", "model.language_model.layers.15.mlp.experts.26.down_proj", "model.language_model.layers.15.mlp.experts.27.down_proj", "model.language_model.layers.15.mlp.experts.28.down_proj", "model.language_model.layers.15.mlp.experts.29.down_proj", "model.language_model.layers.15.mlp.experts.30.down_proj", "model.language_model.layers.15.mlp.experts.31.down_proj", "model.language_model.layers.15.mlp.experts.32.down_proj", "model.language_model.layers.15.mlp.experts.33.down_proj", "model.language_model.layers.15.mlp.experts.34.down_proj", "model.language_model.layers.15.mlp.experts.35.down_proj", "model.language_model.layers.15.mlp.experts.36.down_proj", "model.language_model.layers.15.mlp.experts.37.down_proj", "model.language_model.layers.15.mlp.experts.38.down_proj", "model.language_model.layers.15.mlp.experts.39.down_proj", "model.language_model.layers.15.mlp.experts.40.down_proj", "model.language_model.layers.15.mlp.experts.41.down_proj", "model.language_model.layers.15.mlp.experts.42.down_proj", "model.language_model.layers.15.mlp.experts.43.down_proj", "model.language_model.layers.15.mlp.experts.44.down_proj", "model.language_model.layers.15.mlp.experts.45.down_proj", "model.language_model.layers.15.mlp.experts.46.down_proj", "model.language_model.layers.15.mlp.experts.47.down_proj", "model.language_model.layers.15.mlp.experts.48.down_proj", "model.language_model.layers.15.mlp.experts.49.down_proj", "model.language_model.layers.15.mlp.experts.50.down_proj", "model.language_model.layers.15.mlp.experts.51.down_proj", "model.language_model.layers.15.mlp.experts.52.down_proj", "model.language_model.layers.15.mlp.experts.53.down_proj", "model.language_model.layers.15.mlp.experts.54.down_proj", "model.language_model.layers.15.mlp.experts.55.down_proj", "model.language_model.layers.15.mlp.experts.56.down_proj", "model.language_model.layers.15.mlp.experts.57.down_proj", "model.language_model.layers.15.mlp.experts.58.down_proj", "model.language_model.layers.15.mlp.experts.59.down_proj", "model.language_model.layers.15.mlp.experts.60.down_proj", "model.language_model.layers.15.mlp.experts.61.down_proj", "model.language_model.layers.15.mlp.experts.62.down_proj", "model.language_model.layers.15.mlp.experts.63.down_proj", "model.language_model.layers.15.mlp.experts.64.down_proj", "model.language_model.layers.15.mlp.experts.65.down_proj", "model.language_model.layers.15.mlp.experts.66.down_proj", "model.language_model.layers.15.mlp.experts.67.down_proj", "model.language_model.layers.15.mlp.experts.68.down_proj", "model.language_model.layers.15.mlp.experts.69.down_proj", "model.language_model.layers.15.mlp.experts.70.down_proj", "model.language_model.layers.15.mlp.experts.71.down_proj", "model.language_model.layers.15.mlp.experts.72.down_proj", "model.language_model.layers.15.mlp.experts.73.down_proj", "model.language_model.layers.15.mlp.experts.74.down_proj", "model.language_model.layers.15.mlp.experts.75.down_proj", "model.language_model.layers.15.mlp.experts.76.down_proj", "model.language_model.layers.15.mlp.experts.77.down_proj", "model.language_model.layers.15.mlp.experts.78.down_proj", "model.language_model.layers.15.mlp.experts.79.down_proj", "model.language_model.layers.15.mlp.experts.80.down_proj", "model.language_model.layers.15.mlp.experts.81.down_proj", "model.language_model.layers.15.mlp.experts.82.down_proj", "model.language_model.layers.15.mlp.experts.83.down_proj", "model.language_model.layers.15.mlp.experts.84.down_proj", "model.language_model.layers.15.mlp.experts.85.down_proj", "model.language_model.layers.15.mlp.experts.86.down_proj", "model.language_model.layers.15.mlp.experts.87.down_proj", "model.language_model.layers.15.mlp.experts.88.down_proj", "model.language_model.layers.15.mlp.experts.89.down_proj", "model.language_model.layers.15.mlp.experts.90.down_proj", "model.language_model.layers.15.mlp.experts.91.down_proj", "model.language_model.layers.15.mlp.experts.92.down_proj", "model.language_model.layers.15.mlp.experts.93.down_proj", "model.language_model.layers.15.mlp.experts.94.down_proj", "model.language_model.layers.15.mlp.experts.95.down_proj", "model.language_model.layers.15.mlp.experts.96.down_proj", "model.language_model.layers.15.mlp.experts.97.down_proj", "model.language_model.layers.15.mlp.experts.98.down_proj", "model.language_model.layers.15.mlp.experts.99.down_proj", "model.language_model.layers.15.mlp.experts.100.down_proj", "model.language_model.layers.15.mlp.experts.101.down_proj", "model.language_model.layers.15.mlp.experts.102.down_proj", "model.language_model.layers.15.mlp.experts.103.down_proj", "model.language_model.layers.15.mlp.experts.104.down_proj", "model.language_model.layers.15.mlp.experts.105.down_proj", "model.language_model.layers.15.mlp.experts.106.down_proj", "model.language_model.layers.15.mlp.experts.107.down_proj", "model.language_model.layers.15.mlp.experts.108.down_proj", "model.language_model.layers.15.mlp.experts.109.down_proj", "model.language_model.layers.15.mlp.experts.110.down_proj", "model.language_model.layers.15.mlp.experts.111.down_proj", "model.language_model.layers.15.mlp.experts.112.down_proj", "model.language_model.layers.15.mlp.experts.113.down_proj", "model.language_model.layers.15.mlp.experts.114.down_proj", "model.language_model.layers.15.mlp.experts.115.down_proj", "model.language_model.layers.15.mlp.experts.116.down_proj", "model.language_model.layers.15.mlp.experts.117.down_proj", "model.language_model.layers.15.mlp.experts.118.down_proj", "model.language_model.layers.15.mlp.experts.119.down_proj", "model.language_model.layers.15.mlp.experts.120.down_proj", "model.language_model.layers.15.mlp.experts.121.down_proj", "model.language_model.layers.15.mlp.experts.122.down_proj", "model.language_model.layers.15.mlp.experts.123.down_proj", "model.language_model.layers.15.mlp.experts.124.down_proj", "model.language_model.layers.15.mlp.experts.125.down_proj", "model.language_model.layers.15.mlp.experts.126.down_proj", "model.language_model.layers.15.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.001403581909835333, "dbits": 603979776 } ] }, { "idx": 32, "layers": [ "model.language_model.layers.16.self_attn.q_proj", "model.language_model.layers.16.self_attn.k_proj", "model.language_model.layers.16.self_attn.v_proj", "model.language_model.layers.16.self_attn.o_proj" ], "candidates": [ { "dkld": -1.3752654194829073e-05, "dbits": 18874368 } ] }, { "idx": 33, "layers": [ "model.language_model.layers.16.mlp.experts.0.gate_proj", "model.language_model.layers.16.mlp.experts.1.gate_proj", "model.language_model.layers.16.mlp.experts.2.gate_proj", "model.language_model.layers.16.mlp.experts.3.gate_proj", "model.language_model.layers.16.mlp.experts.4.gate_proj", "model.language_model.layers.16.mlp.experts.5.gate_proj", "model.language_model.layers.16.mlp.experts.6.gate_proj", "model.language_model.layers.16.mlp.experts.7.gate_proj", "model.language_model.layers.16.mlp.experts.8.gate_proj", "model.language_model.layers.16.mlp.experts.9.gate_proj", "model.language_model.layers.16.mlp.experts.10.gate_proj", "model.language_model.layers.16.mlp.experts.11.gate_proj", "model.language_model.layers.16.mlp.experts.12.gate_proj", "model.language_model.layers.16.mlp.experts.13.gate_proj", "model.language_model.layers.16.mlp.experts.14.gate_proj", "model.language_model.layers.16.mlp.experts.15.gate_proj", "model.language_model.layers.16.mlp.experts.16.gate_proj", "model.language_model.layers.16.mlp.experts.17.gate_proj", "model.language_model.layers.16.mlp.experts.18.gate_proj", "model.language_model.layers.16.mlp.experts.19.gate_proj", "model.language_model.layers.16.mlp.experts.20.gate_proj", "model.language_model.layers.16.mlp.experts.21.gate_proj", "model.language_model.layers.16.mlp.experts.22.gate_proj", "model.language_model.layers.16.mlp.experts.23.gate_proj", "model.language_model.layers.16.mlp.experts.24.gate_proj", "model.language_model.layers.16.mlp.experts.25.gate_proj", "model.language_model.layers.16.mlp.experts.26.gate_proj", "model.language_model.layers.16.mlp.experts.27.gate_proj", "model.language_model.layers.16.mlp.experts.28.gate_proj", "model.language_model.layers.16.mlp.experts.29.gate_proj", "model.language_model.layers.16.mlp.experts.30.gate_proj", "model.language_model.layers.16.mlp.experts.31.gate_proj", "model.language_model.layers.16.mlp.experts.32.gate_proj", "model.language_model.layers.16.mlp.experts.33.gate_proj", "model.language_model.layers.16.mlp.experts.34.gate_proj", "model.language_model.layers.16.mlp.experts.35.gate_proj", "model.language_model.layers.16.mlp.experts.36.gate_proj", "model.language_model.layers.16.mlp.experts.37.gate_proj", "model.language_model.layers.16.mlp.experts.38.gate_proj", "model.language_model.layers.16.mlp.experts.39.gate_proj", "model.language_model.layers.16.mlp.experts.40.gate_proj", "model.language_model.layers.16.mlp.experts.41.gate_proj", "model.language_model.layers.16.mlp.experts.42.gate_proj", "model.language_model.layers.16.mlp.experts.43.gate_proj", "model.language_model.layers.16.mlp.experts.44.gate_proj", "model.language_model.layers.16.mlp.experts.45.gate_proj", "model.language_model.layers.16.mlp.experts.46.gate_proj", "model.language_model.layers.16.mlp.experts.47.gate_proj", "model.language_model.layers.16.mlp.experts.48.gate_proj", "model.language_model.layers.16.mlp.experts.49.gate_proj", "model.language_model.layers.16.mlp.experts.50.gate_proj", "model.language_model.layers.16.mlp.experts.51.gate_proj", "model.language_model.layers.16.mlp.experts.52.gate_proj", "model.language_model.layers.16.mlp.experts.53.gate_proj", "model.language_model.layers.16.mlp.experts.54.gate_proj", "model.language_model.layers.16.mlp.experts.55.gate_proj", "model.language_model.layers.16.mlp.experts.56.gate_proj", "model.language_model.layers.16.mlp.experts.57.gate_proj", "model.language_model.layers.16.mlp.experts.58.gate_proj", "model.language_model.layers.16.mlp.experts.59.gate_proj", "model.language_model.layers.16.mlp.experts.60.gate_proj", "model.language_model.layers.16.mlp.experts.61.gate_proj", "model.language_model.layers.16.mlp.experts.62.gate_proj", "model.language_model.layers.16.mlp.experts.63.gate_proj", "model.language_model.layers.16.mlp.experts.64.gate_proj", "model.language_model.layers.16.mlp.experts.65.gate_proj", "model.language_model.layers.16.mlp.experts.66.gate_proj", "model.language_model.layers.16.mlp.experts.67.gate_proj", "model.language_model.layers.16.mlp.experts.68.gate_proj", "model.language_model.layers.16.mlp.experts.69.gate_proj", "model.language_model.layers.16.mlp.experts.70.gate_proj", "model.language_model.layers.16.mlp.experts.71.gate_proj", "model.language_model.layers.16.mlp.experts.72.gate_proj", "model.language_model.layers.16.mlp.experts.73.gate_proj", "model.language_model.layers.16.mlp.experts.74.gate_proj", "model.language_model.layers.16.mlp.experts.75.gate_proj", "model.language_model.layers.16.mlp.experts.76.gate_proj", "model.language_model.layers.16.mlp.experts.77.gate_proj", "model.language_model.layers.16.mlp.experts.78.gate_proj", "model.language_model.layers.16.mlp.experts.79.gate_proj", "model.language_model.layers.16.mlp.experts.80.gate_proj", "model.language_model.layers.16.mlp.experts.81.gate_proj", "model.language_model.layers.16.mlp.experts.82.gate_proj", "model.language_model.layers.16.mlp.experts.83.gate_proj", "model.language_model.layers.16.mlp.experts.84.gate_proj", "model.language_model.layers.16.mlp.experts.85.gate_proj", "model.language_model.layers.16.mlp.experts.86.gate_proj", "model.language_model.layers.16.mlp.experts.87.gate_proj", "model.language_model.layers.16.mlp.experts.88.gate_proj", "model.language_model.layers.16.mlp.experts.89.gate_proj", "model.language_model.layers.16.mlp.experts.90.gate_proj", "model.language_model.layers.16.mlp.experts.91.gate_proj", "model.language_model.layers.16.mlp.experts.92.gate_proj", "model.language_model.layers.16.mlp.experts.93.gate_proj", "model.language_model.layers.16.mlp.experts.94.gate_proj", "model.language_model.layers.16.mlp.experts.95.gate_proj", "model.language_model.layers.16.mlp.experts.96.gate_proj", "model.language_model.layers.16.mlp.experts.97.gate_proj", "model.language_model.layers.16.mlp.experts.98.gate_proj", "model.language_model.layers.16.mlp.experts.99.gate_proj", "model.language_model.layers.16.mlp.experts.100.gate_proj", "model.language_model.layers.16.mlp.experts.101.gate_proj", "model.language_model.layers.16.mlp.experts.102.gate_proj", "model.language_model.layers.16.mlp.experts.103.gate_proj", "model.language_model.layers.16.mlp.experts.104.gate_proj", "model.language_model.layers.16.mlp.experts.105.gate_proj", "model.language_model.layers.16.mlp.experts.106.gate_proj", "model.language_model.layers.16.mlp.experts.107.gate_proj", "model.language_model.layers.16.mlp.experts.108.gate_proj", "model.language_model.layers.16.mlp.experts.109.gate_proj", "model.language_model.layers.16.mlp.experts.110.gate_proj", "model.language_model.layers.16.mlp.experts.111.gate_proj", "model.language_model.layers.16.mlp.experts.112.gate_proj", "model.language_model.layers.16.mlp.experts.113.gate_proj", "model.language_model.layers.16.mlp.experts.114.gate_proj", "model.language_model.layers.16.mlp.experts.115.gate_proj", "model.language_model.layers.16.mlp.experts.116.gate_proj", "model.language_model.layers.16.mlp.experts.117.gate_proj", "model.language_model.layers.16.mlp.experts.118.gate_proj", "model.language_model.layers.16.mlp.experts.119.gate_proj", "model.language_model.layers.16.mlp.experts.120.gate_proj", "model.language_model.layers.16.mlp.experts.121.gate_proj", "model.language_model.layers.16.mlp.experts.122.gate_proj", "model.language_model.layers.16.mlp.experts.123.gate_proj", "model.language_model.layers.16.mlp.experts.124.gate_proj", "model.language_model.layers.16.mlp.experts.125.gate_proj", "model.language_model.layers.16.mlp.experts.126.gate_proj", "model.language_model.layers.16.mlp.experts.127.gate_proj", "model.language_model.layers.16.mlp.experts.0.up_proj", "model.language_model.layers.16.mlp.experts.1.up_proj", "model.language_model.layers.16.mlp.experts.2.up_proj", "model.language_model.layers.16.mlp.experts.3.up_proj", "model.language_model.layers.16.mlp.experts.4.up_proj", "model.language_model.layers.16.mlp.experts.5.up_proj", "model.language_model.layers.16.mlp.experts.6.up_proj", "model.language_model.layers.16.mlp.experts.7.up_proj", "model.language_model.layers.16.mlp.experts.8.up_proj", "model.language_model.layers.16.mlp.experts.9.up_proj", "model.language_model.layers.16.mlp.experts.10.up_proj", "model.language_model.layers.16.mlp.experts.11.up_proj", "model.language_model.layers.16.mlp.experts.12.up_proj", "model.language_model.layers.16.mlp.experts.13.up_proj", "model.language_model.layers.16.mlp.experts.14.up_proj", "model.language_model.layers.16.mlp.experts.15.up_proj", "model.language_model.layers.16.mlp.experts.16.up_proj", "model.language_model.layers.16.mlp.experts.17.up_proj", "model.language_model.layers.16.mlp.experts.18.up_proj", "model.language_model.layers.16.mlp.experts.19.up_proj", "model.language_model.layers.16.mlp.experts.20.up_proj", "model.language_model.layers.16.mlp.experts.21.up_proj", "model.language_model.layers.16.mlp.experts.22.up_proj", "model.language_model.layers.16.mlp.experts.23.up_proj", "model.language_model.layers.16.mlp.experts.24.up_proj", "model.language_model.layers.16.mlp.experts.25.up_proj", "model.language_model.layers.16.mlp.experts.26.up_proj", "model.language_model.layers.16.mlp.experts.27.up_proj", "model.language_model.layers.16.mlp.experts.28.up_proj", "model.language_model.layers.16.mlp.experts.29.up_proj", "model.language_model.layers.16.mlp.experts.30.up_proj", "model.language_model.layers.16.mlp.experts.31.up_proj", "model.language_model.layers.16.mlp.experts.32.up_proj", "model.language_model.layers.16.mlp.experts.33.up_proj", "model.language_model.layers.16.mlp.experts.34.up_proj", "model.language_model.layers.16.mlp.experts.35.up_proj", "model.language_model.layers.16.mlp.experts.36.up_proj", "model.language_model.layers.16.mlp.experts.37.up_proj", "model.language_model.layers.16.mlp.experts.38.up_proj", "model.language_model.layers.16.mlp.experts.39.up_proj", "model.language_model.layers.16.mlp.experts.40.up_proj", "model.language_model.layers.16.mlp.experts.41.up_proj", "model.language_model.layers.16.mlp.experts.42.up_proj", "model.language_model.layers.16.mlp.experts.43.up_proj", "model.language_model.layers.16.mlp.experts.44.up_proj", "model.language_model.layers.16.mlp.experts.45.up_proj", "model.language_model.layers.16.mlp.experts.46.up_proj", "model.language_model.layers.16.mlp.experts.47.up_proj", "model.language_model.layers.16.mlp.experts.48.up_proj", "model.language_model.layers.16.mlp.experts.49.up_proj", "model.language_model.layers.16.mlp.experts.50.up_proj", "model.language_model.layers.16.mlp.experts.51.up_proj", "model.language_model.layers.16.mlp.experts.52.up_proj", "model.language_model.layers.16.mlp.experts.53.up_proj", "model.language_model.layers.16.mlp.experts.54.up_proj", "model.language_model.layers.16.mlp.experts.55.up_proj", "model.language_model.layers.16.mlp.experts.56.up_proj", "model.language_model.layers.16.mlp.experts.57.up_proj", "model.language_model.layers.16.mlp.experts.58.up_proj", "model.language_model.layers.16.mlp.experts.59.up_proj", "model.language_model.layers.16.mlp.experts.60.up_proj", "model.language_model.layers.16.mlp.experts.61.up_proj", "model.language_model.layers.16.mlp.experts.62.up_proj", "model.language_model.layers.16.mlp.experts.63.up_proj", "model.language_model.layers.16.mlp.experts.64.up_proj", "model.language_model.layers.16.mlp.experts.65.up_proj", "model.language_model.layers.16.mlp.experts.66.up_proj", "model.language_model.layers.16.mlp.experts.67.up_proj", "model.language_model.layers.16.mlp.experts.68.up_proj", "model.language_model.layers.16.mlp.experts.69.up_proj", "model.language_model.layers.16.mlp.experts.70.up_proj", "model.language_model.layers.16.mlp.experts.71.up_proj", "model.language_model.layers.16.mlp.experts.72.up_proj", "model.language_model.layers.16.mlp.experts.73.up_proj", "model.language_model.layers.16.mlp.experts.74.up_proj", "model.language_model.layers.16.mlp.experts.75.up_proj", "model.language_model.layers.16.mlp.experts.76.up_proj", "model.language_model.layers.16.mlp.experts.77.up_proj", "model.language_model.layers.16.mlp.experts.78.up_proj", "model.language_model.layers.16.mlp.experts.79.up_proj", "model.language_model.layers.16.mlp.experts.80.up_proj", "model.language_model.layers.16.mlp.experts.81.up_proj", "model.language_model.layers.16.mlp.experts.82.up_proj", "model.language_model.layers.16.mlp.experts.83.up_proj", "model.language_model.layers.16.mlp.experts.84.up_proj", "model.language_model.layers.16.mlp.experts.85.up_proj", "model.language_model.layers.16.mlp.experts.86.up_proj", "model.language_model.layers.16.mlp.experts.87.up_proj", "model.language_model.layers.16.mlp.experts.88.up_proj", "model.language_model.layers.16.mlp.experts.89.up_proj", "model.language_model.layers.16.mlp.experts.90.up_proj", "model.language_model.layers.16.mlp.experts.91.up_proj", "model.language_model.layers.16.mlp.experts.92.up_proj", "model.language_model.layers.16.mlp.experts.93.up_proj", "model.language_model.layers.16.mlp.experts.94.up_proj", "model.language_model.layers.16.mlp.experts.95.up_proj", "model.language_model.layers.16.mlp.experts.96.up_proj", "model.language_model.layers.16.mlp.experts.97.up_proj", "model.language_model.layers.16.mlp.experts.98.up_proj", "model.language_model.layers.16.mlp.experts.99.up_proj", "model.language_model.layers.16.mlp.experts.100.up_proj", "model.language_model.layers.16.mlp.experts.101.up_proj", "model.language_model.layers.16.mlp.experts.102.up_proj", "model.language_model.layers.16.mlp.experts.103.up_proj", "model.language_model.layers.16.mlp.experts.104.up_proj", "model.language_model.layers.16.mlp.experts.105.up_proj", "model.language_model.layers.16.mlp.experts.106.up_proj", "model.language_model.layers.16.mlp.experts.107.up_proj", "model.language_model.layers.16.mlp.experts.108.up_proj", "model.language_model.layers.16.mlp.experts.109.up_proj", "model.language_model.layers.16.mlp.experts.110.up_proj", "model.language_model.layers.16.mlp.experts.111.up_proj", "model.language_model.layers.16.mlp.experts.112.up_proj", "model.language_model.layers.16.mlp.experts.113.up_proj", "model.language_model.layers.16.mlp.experts.114.up_proj", "model.language_model.layers.16.mlp.experts.115.up_proj", "model.language_model.layers.16.mlp.experts.116.up_proj", "model.language_model.layers.16.mlp.experts.117.up_proj", "model.language_model.layers.16.mlp.experts.118.up_proj", "model.language_model.layers.16.mlp.experts.119.up_proj", "model.language_model.layers.16.mlp.experts.120.up_proj", "model.language_model.layers.16.mlp.experts.121.up_proj", "model.language_model.layers.16.mlp.experts.122.up_proj", "model.language_model.layers.16.mlp.experts.123.up_proj", "model.language_model.layers.16.mlp.experts.124.up_proj", "model.language_model.layers.16.mlp.experts.125.up_proj", "model.language_model.layers.16.mlp.experts.126.up_proj", "model.language_model.layers.16.mlp.experts.127.up_proj", "model.language_model.layers.16.mlp.experts.0.down_proj", "model.language_model.layers.16.mlp.experts.1.down_proj", "model.language_model.layers.16.mlp.experts.2.down_proj", "model.language_model.layers.16.mlp.experts.3.down_proj", "model.language_model.layers.16.mlp.experts.4.down_proj", "model.language_model.layers.16.mlp.experts.5.down_proj", "model.language_model.layers.16.mlp.experts.6.down_proj", "model.language_model.layers.16.mlp.experts.7.down_proj", "model.language_model.layers.16.mlp.experts.8.down_proj", "model.language_model.layers.16.mlp.experts.9.down_proj", "model.language_model.layers.16.mlp.experts.10.down_proj", "model.language_model.layers.16.mlp.experts.11.down_proj", "model.language_model.layers.16.mlp.experts.12.down_proj", "model.language_model.layers.16.mlp.experts.13.down_proj", "model.language_model.layers.16.mlp.experts.14.down_proj", "model.language_model.layers.16.mlp.experts.15.down_proj", "model.language_model.layers.16.mlp.experts.16.down_proj", "model.language_model.layers.16.mlp.experts.17.down_proj", "model.language_model.layers.16.mlp.experts.18.down_proj", "model.language_model.layers.16.mlp.experts.19.down_proj", "model.language_model.layers.16.mlp.experts.20.down_proj", "model.language_model.layers.16.mlp.experts.21.down_proj", "model.language_model.layers.16.mlp.experts.22.down_proj", "model.language_model.layers.16.mlp.experts.23.down_proj", "model.language_model.layers.16.mlp.experts.24.down_proj", "model.language_model.layers.16.mlp.experts.25.down_proj", "model.language_model.layers.16.mlp.experts.26.down_proj", "model.language_model.layers.16.mlp.experts.27.down_proj", "model.language_model.layers.16.mlp.experts.28.down_proj", "model.language_model.layers.16.mlp.experts.29.down_proj", "model.language_model.layers.16.mlp.experts.30.down_proj", "model.language_model.layers.16.mlp.experts.31.down_proj", "model.language_model.layers.16.mlp.experts.32.down_proj", "model.language_model.layers.16.mlp.experts.33.down_proj", "model.language_model.layers.16.mlp.experts.34.down_proj", "model.language_model.layers.16.mlp.experts.35.down_proj", "model.language_model.layers.16.mlp.experts.36.down_proj", "model.language_model.layers.16.mlp.experts.37.down_proj", "model.language_model.layers.16.mlp.experts.38.down_proj", "model.language_model.layers.16.mlp.experts.39.down_proj", "model.language_model.layers.16.mlp.experts.40.down_proj", "model.language_model.layers.16.mlp.experts.41.down_proj", "model.language_model.layers.16.mlp.experts.42.down_proj", "model.language_model.layers.16.mlp.experts.43.down_proj", "model.language_model.layers.16.mlp.experts.44.down_proj", "model.language_model.layers.16.mlp.experts.45.down_proj", "model.language_model.layers.16.mlp.experts.46.down_proj", "model.language_model.layers.16.mlp.experts.47.down_proj", "model.language_model.layers.16.mlp.experts.48.down_proj", "model.language_model.layers.16.mlp.experts.49.down_proj", "model.language_model.layers.16.mlp.experts.50.down_proj", "model.language_model.layers.16.mlp.experts.51.down_proj", "model.language_model.layers.16.mlp.experts.52.down_proj", "model.language_model.layers.16.mlp.experts.53.down_proj", "model.language_model.layers.16.mlp.experts.54.down_proj", "model.language_model.layers.16.mlp.experts.55.down_proj", "model.language_model.layers.16.mlp.experts.56.down_proj", "model.language_model.layers.16.mlp.experts.57.down_proj", "model.language_model.layers.16.mlp.experts.58.down_proj", "model.language_model.layers.16.mlp.experts.59.down_proj", "model.language_model.layers.16.mlp.experts.60.down_proj", "model.language_model.layers.16.mlp.experts.61.down_proj", "model.language_model.layers.16.mlp.experts.62.down_proj", "model.language_model.layers.16.mlp.experts.63.down_proj", "model.language_model.layers.16.mlp.experts.64.down_proj", "model.language_model.layers.16.mlp.experts.65.down_proj", "model.language_model.layers.16.mlp.experts.66.down_proj", "model.language_model.layers.16.mlp.experts.67.down_proj", "model.language_model.layers.16.mlp.experts.68.down_proj", "model.language_model.layers.16.mlp.experts.69.down_proj", "model.language_model.layers.16.mlp.experts.70.down_proj", "model.language_model.layers.16.mlp.experts.71.down_proj", "model.language_model.layers.16.mlp.experts.72.down_proj", "model.language_model.layers.16.mlp.experts.73.down_proj", "model.language_model.layers.16.mlp.experts.74.down_proj", "model.language_model.layers.16.mlp.experts.75.down_proj", "model.language_model.layers.16.mlp.experts.76.down_proj", "model.language_model.layers.16.mlp.experts.77.down_proj", "model.language_model.layers.16.mlp.experts.78.down_proj", "model.language_model.layers.16.mlp.experts.79.down_proj", "model.language_model.layers.16.mlp.experts.80.down_proj", "model.language_model.layers.16.mlp.experts.81.down_proj", "model.language_model.layers.16.mlp.experts.82.down_proj", "model.language_model.layers.16.mlp.experts.83.down_proj", "model.language_model.layers.16.mlp.experts.84.down_proj", "model.language_model.layers.16.mlp.experts.85.down_proj", "model.language_model.layers.16.mlp.experts.86.down_proj", "model.language_model.layers.16.mlp.experts.87.down_proj", "model.language_model.layers.16.mlp.experts.88.down_proj", "model.language_model.layers.16.mlp.experts.89.down_proj", "model.language_model.layers.16.mlp.experts.90.down_proj", "model.language_model.layers.16.mlp.experts.91.down_proj", "model.language_model.layers.16.mlp.experts.92.down_proj", "model.language_model.layers.16.mlp.experts.93.down_proj", "model.language_model.layers.16.mlp.experts.94.down_proj", "model.language_model.layers.16.mlp.experts.95.down_proj", "model.language_model.layers.16.mlp.experts.96.down_proj", "model.language_model.layers.16.mlp.experts.97.down_proj", "model.language_model.layers.16.mlp.experts.98.down_proj", "model.language_model.layers.16.mlp.experts.99.down_proj", "model.language_model.layers.16.mlp.experts.100.down_proj", "model.language_model.layers.16.mlp.experts.101.down_proj", "model.language_model.layers.16.mlp.experts.102.down_proj", "model.language_model.layers.16.mlp.experts.103.down_proj", "model.language_model.layers.16.mlp.experts.104.down_proj", "model.language_model.layers.16.mlp.experts.105.down_proj", "model.language_model.layers.16.mlp.experts.106.down_proj", "model.language_model.layers.16.mlp.experts.107.down_proj", "model.language_model.layers.16.mlp.experts.108.down_proj", "model.language_model.layers.16.mlp.experts.109.down_proj", "model.language_model.layers.16.mlp.experts.110.down_proj", "model.language_model.layers.16.mlp.experts.111.down_proj", "model.language_model.layers.16.mlp.experts.112.down_proj", "model.language_model.layers.16.mlp.experts.113.down_proj", "model.language_model.layers.16.mlp.experts.114.down_proj", "model.language_model.layers.16.mlp.experts.115.down_proj", "model.language_model.layers.16.mlp.experts.116.down_proj", "model.language_model.layers.16.mlp.experts.117.down_proj", "model.language_model.layers.16.mlp.experts.118.down_proj", "model.language_model.layers.16.mlp.experts.119.down_proj", "model.language_model.layers.16.mlp.experts.120.down_proj", "model.language_model.layers.16.mlp.experts.121.down_proj", "model.language_model.layers.16.mlp.experts.122.down_proj", "model.language_model.layers.16.mlp.experts.123.down_proj", "model.language_model.layers.16.mlp.experts.124.down_proj", "model.language_model.layers.16.mlp.experts.125.down_proj", "model.language_model.layers.16.mlp.experts.126.down_proj", "model.language_model.layers.16.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.0010460568591952324, "dbits": 603979776 } ] }, { "idx": 34, "layers": [ "model.language_model.layers.17.self_attn.q_proj", "model.language_model.layers.17.self_attn.k_proj", "model.language_model.layers.17.self_attn.v_proj", "model.language_model.layers.17.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0004175770096480902, "dbits": 18874368 } ] }, { "idx": 35, "layers": [ "model.language_model.layers.17.mlp.experts.0.gate_proj", "model.language_model.layers.17.mlp.experts.1.gate_proj", "model.language_model.layers.17.mlp.experts.2.gate_proj", "model.language_model.layers.17.mlp.experts.3.gate_proj", "model.language_model.layers.17.mlp.experts.4.gate_proj", "model.language_model.layers.17.mlp.experts.5.gate_proj", "model.language_model.layers.17.mlp.experts.6.gate_proj", "model.language_model.layers.17.mlp.experts.7.gate_proj", "model.language_model.layers.17.mlp.experts.8.gate_proj", "model.language_model.layers.17.mlp.experts.9.gate_proj", "model.language_model.layers.17.mlp.experts.10.gate_proj", "model.language_model.layers.17.mlp.experts.11.gate_proj", "model.language_model.layers.17.mlp.experts.12.gate_proj", "model.language_model.layers.17.mlp.experts.13.gate_proj", "model.language_model.layers.17.mlp.experts.14.gate_proj", "model.language_model.layers.17.mlp.experts.15.gate_proj", "model.language_model.layers.17.mlp.experts.16.gate_proj", "model.language_model.layers.17.mlp.experts.17.gate_proj", "model.language_model.layers.17.mlp.experts.18.gate_proj", "model.language_model.layers.17.mlp.experts.19.gate_proj", "model.language_model.layers.17.mlp.experts.20.gate_proj", "model.language_model.layers.17.mlp.experts.21.gate_proj", "model.language_model.layers.17.mlp.experts.22.gate_proj", "model.language_model.layers.17.mlp.experts.23.gate_proj", "model.language_model.layers.17.mlp.experts.24.gate_proj", "model.language_model.layers.17.mlp.experts.25.gate_proj", "model.language_model.layers.17.mlp.experts.26.gate_proj", "model.language_model.layers.17.mlp.experts.27.gate_proj", "model.language_model.layers.17.mlp.experts.28.gate_proj", "model.language_model.layers.17.mlp.experts.29.gate_proj", "model.language_model.layers.17.mlp.experts.30.gate_proj", "model.language_model.layers.17.mlp.experts.31.gate_proj", "model.language_model.layers.17.mlp.experts.32.gate_proj", "model.language_model.layers.17.mlp.experts.33.gate_proj", "model.language_model.layers.17.mlp.experts.34.gate_proj", "model.language_model.layers.17.mlp.experts.35.gate_proj", "model.language_model.layers.17.mlp.experts.36.gate_proj", "model.language_model.layers.17.mlp.experts.37.gate_proj", "model.language_model.layers.17.mlp.experts.38.gate_proj", "model.language_model.layers.17.mlp.experts.39.gate_proj", "model.language_model.layers.17.mlp.experts.40.gate_proj", "model.language_model.layers.17.mlp.experts.41.gate_proj", "model.language_model.layers.17.mlp.experts.42.gate_proj", "model.language_model.layers.17.mlp.experts.43.gate_proj", "model.language_model.layers.17.mlp.experts.44.gate_proj", "model.language_model.layers.17.mlp.experts.45.gate_proj", "model.language_model.layers.17.mlp.experts.46.gate_proj", "model.language_model.layers.17.mlp.experts.47.gate_proj", "model.language_model.layers.17.mlp.experts.48.gate_proj", "model.language_model.layers.17.mlp.experts.49.gate_proj", "model.language_model.layers.17.mlp.experts.50.gate_proj", "model.language_model.layers.17.mlp.experts.51.gate_proj", "model.language_model.layers.17.mlp.experts.52.gate_proj", "model.language_model.layers.17.mlp.experts.53.gate_proj", "model.language_model.layers.17.mlp.experts.54.gate_proj", "model.language_model.layers.17.mlp.experts.55.gate_proj", "model.language_model.layers.17.mlp.experts.56.gate_proj", "model.language_model.layers.17.mlp.experts.57.gate_proj", "model.language_model.layers.17.mlp.experts.58.gate_proj", "model.language_model.layers.17.mlp.experts.59.gate_proj", "model.language_model.layers.17.mlp.experts.60.gate_proj", "model.language_model.layers.17.mlp.experts.61.gate_proj", "model.language_model.layers.17.mlp.experts.62.gate_proj", "model.language_model.layers.17.mlp.experts.63.gate_proj", "model.language_model.layers.17.mlp.experts.64.gate_proj", "model.language_model.layers.17.mlp.experts.65.gate_proj", "model.language_model.layers.17.mlp.experts.66.gate_proj", "model.language_model.layers.17.mlp.experts.67.gate_proj", "model.language_model.layers.17.mlp.experts.68.gate_proj", "model.language_model.layers.17.mlp.experts.69.gate_proj", "model.language_model.layers.17.mlp.experts.70.gate_proj", "model.language_model.layers.17.mlp.experts.71.gate_proj", "model.language_model.layers.17.mlp.experts.72.gate_proj", "model.language_model.layers.17.mlp.experts.73.gate_proj", "model.language_model.layers.17.mlp.experts.74.gate_proj", "model.language_model.layers.17.mlp.experts.75.gate_proj", "model.language_model.layers.17.mlp.experts.76.gate_proj", "model.language_model.layers.17.mlp.experts.77.gate_proj", "model.language_model.layers.17.mlp.experts.78.gate_proj", "model.language_model.layers.17.mlp.experts.79.gate_proj", "model.language_model.layers.17.mlp.experts.80.gate_proj", "model.language_model.layers.17.mlp.experts.81.gate_proj", "model.language_model.layers.17.mlp.experts.82.gate_proj", "model.language_model.layers.17.mlp.experts.83.gate_proj", "model.language_model.layers.17.mlp.experts.84.gate_proj", "model.language_model.layers.17.mlp.experts.85.gate_proj", "model.language_model.layers.17.mlp.experts.86.gate_proj", "model.language_model.layers.17.mlp.experts.87.gate_proj", "model.language_model.layers.17.mlp.experts.88.gate_proj", "model.language_model.layers.17.mlp.experts.89.gate_proj", "model.language_model.layers.17.mlp.experts.90.gate_proj", "model.language_model.layers.17.mlp.experts.91.gate_proj", "model.language_model.layers.17.mlp.experts.92.gate_proj", "model.language_model.layers.17.mlp.experts.93.gate_proj", "model.language_model.layers.17.mlp.experts.94.gate_proj", "model.language_model.layers.17.mlp.experts.95.gate_proj", "model.language_model.layers.17.mlp.experts.96.gate_proj", "model.language_model.layers.17.mlp.experts.97.gate_proj", "model.language_model.layers.17.mlp.experts.98.gate_proj", "model.language_model.layers.17.mlp.experts.99.gate_proj", "model.language_model.layers.17.mlp.experts.100.gate_proj", "model.language_model.layers.17.mlp.experts.101.gate_proj", "model.language_model.layers.17.mlp.experts.102.gate_proj", "model.language_model.layers.17.mlp.experts.103.gate_proj", "model.language_model.layers.17.mlp.experts.104.gate_proj", "model.language_model.layers.17.mlp.experts.105.gate_proj", "model.language_model.layers.17.mlp.experts.106.gate_proj", "model.language_model.layers.17.mlp.experts.107.gate_proj", "model.language_model.layers.17.mlp.experts.108.gate_proj", "model.language_model.layers.17.mlp.experts.109.gate_proj", "model.language_model.layers.17.mlp.experts.110.gate_proj", "model.language_model.layers.17.mlp.experts.111.gate_proj", "model.language_model.layers.17.mlp.experts.112.gate_proj", "model.language_model.layers.17.mlp.experts.113.gate_proj", "model.language_model.layers.17.mlp.experts.114.gate_proj", "model.language_model.layers.17.mlp.experts.115.gate_proj", "model.language_model.layers.17.mlp.experts.116.gate_proj", "model.language_model.layers.17.mlp.experts.117.gate_proj", "model.language_model.layers.17.mlp.experts.118.gate_proj", "model.language_model.layers.17.mlp.experts.119.gate_proj", "model.language_model.layers.17.mlp.experts.120.gate_proj", "model.language_model.layers.17.mlp.experts.121.gate_proj", "model.language_model.layers.17.mlp.experts.122.gate_proj", "model.language_model.layers.17.mlp.experts.123.gate_proj", "model.language_model.layers.17.mlp.experts.124.gate_proj", "model.language_model.layers.17.mlp.experts.125.gate_proj", "model.language_model.layers.17.mlp.experts.126.gate_proj", "model.language_model.layers.17.mlp.experts.127.gate_proj", "model.language_model.layers.17.mlp.experts.0.up_proj", "model.language_model.layers.17.mlp.experts.1.up_proj", "model.language_model.layers.17.mlp.experts.2.up_proj", "model.language_model.layers.17.mlp.experts.3.up_proj", "model.language_model.layers.17.mlp.experts.4.up_proj", "model.language_model.layers.17.mlp.experts.5.up_proj", "model.language_model.layers.17.mlp.experts.6.up_proj", "model.language_model.layers.17.mlp.experts.7.up_proj", "model.language_model.layers.17.mlp.experts.8.up_proj", "model.language_model.layers.17.mlp.experts.9.up_proj", "model.language_model.layers.17.mlp.experts.10.up_proj", "model.language_model.layers.17.mlp.experts.11.up_proj", "model.language_model.layers.17.mlp.experts.12.up_proj", "model.language_model.layers.17.mlp.experts.13.up_proj", "model.language_model.layers.17.mlp.experts.14.up_proj", "model.language_model.layers.17.mlp.experts.15.up_proj", "model.language_model.layers.17.mlp.experts.16.up_proj", "model.language_model.layers.17.mlp.experts.17.up_proj", "model.language_model.layers.17.mlp.experts.18.up_proj", "model.language_model.layers.17.mlp.experts.19.up_proj", "model.language_model.layers.17.mlp.experts.20.up_proj", "model.language_model.layers.17.mlp.experts.21.up_proj", "model.language_model.layers.17.mlp.experts.22.up_proj", "model.language_model.layers.17.mlp.experts.23.up_proj", "model.language_model.layers.17.mlp.experts.24.up_proj", "model.language_model.layers.17.mlp.experts.25.up_proj", "model.language_model.layers.17.mlp.experts.26.up_proj", "model.language_model.layers.17.mlp.experts.27.up_proj", "model.language_model.layers.17.mlp.experts.28.up_proj", "model.language_model.layers.17.mlp.experts.29.up_proj", "model.language_model.layers.17.mlp.experts.30.up_proj", "model.language_model.layers.17.mlp.experts.31.up_proj", "model.language_model.layers.17.mlp.experts.32.up_proj", "model.language_model.layers.17.mlp.experts.33.up_proj", "model.language_model.layers.17.mlp.experts.34.up_proj", "model.language_model.layers.17.mlp.experts.35.up_proj", "model.language_model.layers.17.mlp.experts.36.up_proj", "model.language_model.layers.17.mlp.experts.37.up_proj", "model.language_model.layers.17.mlp.experts.38.up_proj", "model.language_model.layers.17.mlp.experts.39.up_proj", "model.language_model.layers.17.mlp.experts.40.up_proj", "model.language_model.layers.17.mlp.experts.41.up_proj", "model.language_model.layers.17.mlp.experts.42.up_proj", "model.language_model.layers.17.mlp.experts.43.up_proj", "model.language_model.layers.17.mlp.experts.44.up_proj", "model.language_model.layers.17.mlp.experts.45.up_proj", "model.language_model.layers.17.mlp.experts.46.up_proj", "model.language_model.layers.17.mlp.experts.47.up_proj", "model.language_model.layers.17.mlp.experts.48.up_proj", "model.language_model.layers.17.mlp.experts.49.up_proj", "model.language_model.layers.17.mlp.experts.50.up_proj", "model.language_model.layers.17.mlp.experts.51.up_proj", "model.language_model.layers.17.mlp.experts.52.up_proj", "model.language_model.layers.17.mlp.experts.53.up_proj", "model.language_model.layers.17.mlp.experts.54.up_proj", "model.language_model.layers.17.mlp.experts.55.up_proj", "model.language_model.layers.17.mlp.experts.56.up_proj", "model.language_model.layers.17.mlp.experts.57.up_proj", "model.language_model.layers.17.mlp.experts.58.up_proj", "model.language_model.layers.17.mlp.experts.59.up_proj", "model.language_model.layers.17.mlp.experts.60.up_proj", "model.language_model.layers.17.mlp.experts.61.up_proj", "model.language_model.layers.17.mlp.experts.62.up_proj", "model.language_model.layers.17.mlp.experts.63.up_proj", "model.language_model.layers.17.mlp.experts.64.up_proj", "model.language_model.layers.17.mlp.experts.65.up_proj", "model.language_model.layers.17.mlp.experts.66.up_proj", "model.language_model.layers.17.mlp.experts.67.up_proj", "model.language_model.layers.17.mlp.experts.68.up_proj", "model.language_model.layers.17.mlp.experts.69.up_proj", "model.language_model.layers.17.mlp.experts.70.up_proj", "model.language_model.layers.17.mlp.experts.71.up_proj", "model.language_model.layers.17.mlp.experts.72.up_proj", "model.language_model.layers.17.mlp.experts.73.up_proj", "model.language_model.layers.17.mlp.experts.74.up_proj", "model.language_model.layers.17.mlp.experts.75.up_proj", "model.language_model.layers.17.mlp.experts.76.up_proj", "model.language_model.layers.17.mlp.experts.77.up_proj", "model.language_model.layers.17.mlp.experts.78.up_proj", "model.language_model.layers.17.mlp.experts.79.up_proj", "model.language_model.layers.17.mlp.experts.80.up_proj", "model.language_model.layers.17.mlp.experts.81.up_proj", "model.language_model.layers.17.mlp.experts.82.up_proj", "model.language_model.layers.17.mlp.experts.83.up_proj", "model.language_model.layers.17.mlp.experts.84.up_proj", "model.language_model.layers.17.mlp.experts.85.up_proj", "model.language_model.layers.17.mlp.experts.86.up_proj", "model.language_model.layers.17.mlp.experts.87.up_proj", "model.language_model.layers.17.mlp.experts.88.up_proj", "model.language_model.layers.17.mlp.experts.89.up_proj", "model.language_model.layers.17.mlp.experts.90.up_proj", "model.language_model.layers.17.mlp.experts.91.up_proj", "model.language_model.layers.17.mlp.experts.92.up_proj", "model.language_model.layers.17.mlp.experts.93.up_proj", "model.language_model.layers.17.mlp.experts.94.up_proj", "model.language_model.layers.17.mlp.experts.95.up_proj", "model.language_model.layers.17.mlp.experts.96.up_proj", "model.language_model.layers.17.mlp.experts.97.up_proj", "model.language_model.layers.17.mlp.experts.98.up_proj", "model.language_model.layers.17.mlp.experts.99.up_proj", "model.language_model.layers.17.mlp.experts.100.up_proj", "model.language_model.layers.17.mlp.experts.101.up_proj", "model.language_model.layers.17.mlp.experts.102.up_proj", "model.language_model.layers.17.mlp.experts.103.up_proj", "model.language_model.layers.17.mlp.experts.104.up_proj", "model.language_model.layers.17.mlp.experts.105.up_proj", "model.language_model.layers.17.mlp.experts.106.up_proj", "model.language_model.layers.17.mlp.experts.107.up_proj", "model.language_model.layers.17.mlp.experts.108.up_proj", "model.language_model.layers.17.mlp.experts.109.up_proj", "model.language_model.layers.17.mlp.experts.110.up_proj", "model.language_model.layers.17.mlp.experts.111.up_proj", "model.language_model.layers.17.mlp.experts.112.up_proj", "model.language_model.layers.17.mlp.experts.113.up_proj", "model.language_model.layers.17.mlp.experts.114.up_proj", "model.language_model.layers.17.mlp.experts.115.up_proj", "model.language_model.layers.17.mlp.experts.116.up_proj", "model.language_model.layers.17.mlp.experts.117.up_proj", "model.language_model.layers.17.mlp.experts.118.up_proj", "model.language_model.layers.17.mlp.experts.119.up_proj", "model.language_model.layers.17.mlp.experts.120.up_proj", "model.language_model.layers.17.mlp.experts.121.up_proj", "model.language_model.layers.17.mlp.experts.122.up_proj", "model.language_model.layers.17.mlp.experts.123.up_proj", "model.language_model.layers.17.mlp.experts.124.up_proj", "model.language_model.layers.17.mlp.experts.125.up_proj", "model.language_model.layers.17.mlp.experts.126.up_proj", "model.language_model.layers.17.mlp.experts.127.up_proj", "model.language_model.layers.17.mlp.experts.0.down_proj", "model.language_model.layers.17.mlp.experts.1.down_proj", "model.language_model.layers.17.mlp.experts.2.down_proj", "model.language_model.layers.17.mlp.experts.3.down_proj", "model.language_model.layers.17.mlp.experts.4.down_proj", "model.language_model.layers.17.mlp.experts.5.down_proj", "model.language_model.layers.17.mlp.experts.6.down_proj", "model.language_model.layers.17.mlp.experts.7.down_proj", "model.language_model.layers.17.mlp.experts.8.down_proj", "model.language_model.layers.17.mlp.experts.9.down_proj", "model.language_model.layers.17.mlp.experts.10.down_proj", "model.language_model.layers.17.mlp.experts.11.down_proj", "model.language_model.layers.17.mlp.experts.12.down_proj", "model.language_model.layers.17.mlp.experts.13.down_proj", "model.language_model.layers.17.mlp.experts.14.down_proj", "model.language_model.layers.17.mlp.experts.15.down_proj", "model.language_model.layers.17.mlp.experts.16.down_proj", "model.language_model.layers.17.mlp.experts.17.down_proj", "model.language_model.layers.17.mlp.experts.18.down_proj", "model.language_model.layers.17.mlp.experts.19.down_proj", "model.language_model.layers.17.mlp.experts.20.down_proj", "model.language_model.layers.17.mlp.experts.21.down_proj", "model.language_model.layers.17.mlp.experts.22.down_proj", "model.language_model.layers.17.mlp.experts.23.down_proj", "model.language_model.layers.17.mlp.experts.24.down_proj", "model.language_model.layers.17.mlp.experts.25.down_proj", "model.language_model.layers.17.mlp.experts.26.down_proj", "model.language_model.layers.17.mlp.experts.27.down_proj", "model.language_model.layers.17.mlp.experts.28.down_proj", "model.language_model.layers.17.mlp.experts.29.down_proj", "model.language_model.layers.17.mlp.experts.30.down_proj", "model.language_model.layers.17.mlp.experts.31.down_proj", "model.language_model.layers.17.mlp.experts.32.down_proj", "model.language_model.layers.17.mlp.experts.33.down_proj", "model.language_model.layers.17.mlp.experts.34.down_proj", "model.language_model.layers.17.mlp.experts.35.down_proj", "model.language_model.layers.17.mlp.experts.36.down_proj", "model.language_model.layers.17.mlp.experts.37.down_proj", "model.language_model.layers.17.mlp.experts.38.down_proj", "model.language_model.layers.17.mlp.experts.39.down_proj", "model.language_model.layers.17.mlp.experts.40.down_proj", "model.language_model.layers.17.mlp.experts.41.down_proj", "model.language_model.layers.17.mlp.experts.42.down_proj", "model.language_model.layers.17.mlp.experts.43.down_proj", "model.language_model.layers.17.mlp.experts.44.down_proj", "model.language_model.layers.17.mlp.experts.45.down_proj", "model.language_model.layers.17.mlp.experts.46.down_proj", "model.language_model.layers.17.mlp.experts.47.down_proj", "model.language_model.layers.17.mlp.experts.48.down_proj", "model.language_model.layers.17.mlp.experts.49.down_proj", "model.language_model.layers.17.mlp.experts.50.down_proj", "model.language_model.layers.17.mlp.experts.51.down_proj", "model.language_model.layers.17.mlp.experts.52.down_proj", "model.language_model.layers.17.mlp.experts.53.down_proj", "model.language_model.layers.17.mlp.experts.54.down_proj", "model.language_model.layers.17.mlp.experts.55.down_proj", "model.language_model.layers.17.mlp.experts.56.down_proj", "model.language_model.layers.17.mlp.experts.57.down_proj", "model.language_model.layers.17.mlp.experts.58.down_proj", "model.language_model.layers.17.mlp.experts.59.down_proj", "model.language_model.layers.17.mlp.experts.60.down_proj", "model.language_model.layers.17.mlp.experts.61.down_proj", "model.language_model.layers.17.mlp.experts.62.down_proj", "model.language_model.layers.17.mlp.experts.63.down_proj", "model.language_model.layers.17.mlp.experts.64.down_proj", "model.language_model.layers.17.mlp.experts.65.down_proj", "model.language_model.layers.17.mlp.experts.66.down_proj", "model.language_model.layers.17.mlp.experts.67.down_proj", "model.language_model.layers.17.mlp.experts.68.down_proj", "model.language_model.layers.17.mlp.experts.69.down_proj", "model.language_model.layers.17.mlp.experts.70.down_proj", "model.language_model.layers.17.mlp.experts.71.down_proj", "model.language_model.layers.17.mlp.experts.72.down_proj", "model.language_model.layers.17.mlp.experts.73.down_proj", "model.language_model.layers.17.mlp.experts.74.down_proj", "model.language_model.layers.17.mlp.experts.75.down_proj", "model.language_model.layers.17.mlp.experts.76.down_proj", "model.language_model.layers.17.mlp.experts.77.down_proj", "model.language_model.layers.17.mlp.experts.78.down_proj", "model.language_model.layers.17.mlp.experts.79.down_proj", "model.language_model.layers.17.mlp.experts.80.down_proj", "model.language_model.layers.17.mlp.experts.81.down_proj", "model.language_model.layers.17.mlp.experts.82.down_proj", "model.language_model.layers.17.mlp.experts.83.down_proj", "model.language_model.layers.17.mlp.experts.84.down_proj", "model.language_model.layers.17.mlp.experts.85.down_proj", "model.language_model.layers.17.mlp.experts.86.down_proj", "model.language_model.layers.17.mlp.experts.87.down_proj", "model.language_model.layers.17.mlp.experts.88.down_proj", "model.language_model.layers.17.mlp.experts.89.down_proj", "model.language_model.layers.17.mlp.experts.90.down_proj", "model.language_model.layers.17.mlp.experts.91.down_proj", "model.language_model.layers.17.mlp.experts.92.down_proj", "model.language_model.layers.17.mlp.experts.93.down_proj", "model.language_model.layers.17.mlp.experts.94.down_proj", "model.language_model.layers.17.mlp.experts.95.down_proj", "model.language_model.layers.17.mlp.experts.96.down_proj", "model.language_model.layers.17.mlp.experts.97.down_proj", "model.language_model.layers.17.mlp.experts.98.down_proj", "model.language_model.layers.17.mlp.experts.99.down_proj", "model.language_model.layers.17.mlp.experts.100.down_proj", "model.language_model.layers.17.mlp.experts.101.down_proj", "model.language_model.layers.17.mlp.experts.102.down_proj", "model.language_model.layers.17.mlp.experts.103.down_proj", "model.language_model.layers.17.mlp.experts.104.down_proj", "model.language_model.layers.17.mlp.experts.105.down_proj", "model.language_model.layers.17.mlp.experts.106.down_proj", "model.language_model.layers.17.mlp.experts.107.down_proj", "model.language_model.layers.17.mlp.experts.108.down_proj", "model.language_model.layers.17.mlp.experts.109.down_proj", "model.language_model.layers.17.mlp.experts.110.down_proj", "model.language_model.layers.17.mlp.experts.111.down_proj", "model.language_model.layers.17.mlp.experts.112.down_proj", "model.language_model.layers.17.mlp.experts.113.down_proj", "model.language_model.layers.17.mlp.experts.114.down_proj", "model.language_model.layers.17.mlp.experts.115.down_proj", "model.language_model.layers.17.mlp.experts.116.down_proj", "model.language_model.layers.17.mlp.experts.117.down_proj", "model.language_model.layers.17.mlp.experts.118.down_proj", "model.language_model.layers.17.mlp.experts.119.down_proj", "model.language_model.layers.17.mlp.experts.120.down_proj", "model.language_model.layers.17.mlp.experts.121.down_proj", "model.language_model.layers.17.mlp.experts.122.down_proj", "model.language_model.layers.17.mlp.experts.123.down_proj", "model.language_model.layers.17.mlp.experts.124.down_proj", "model.language_model.layers.17.mlp.experts.125.down_proj", "model.language_model.layers.17.mlp.experts.126.down_proj", "model.language_model.layers.17.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.0009451342746615438, "dbits": 603979776 } ] }, { "idx": 36, "layers": [ "model.language_model.layers.18.self_attn.q_proj", "model.language_model.layers.18.self_attn.k_proj", "model.language_model.layers.18.self_attn.v_proj", "model.language_model.layers.18.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0010105229914188357, "dbits": 18874368 } ] }, { "idx": 37, "layers": [ "model.language_model.layers.18.mlp.experts.0.gate_proj", "model.language_model.layers.18.mlp.experts.1.gate_proj", "model.language_model.layers.18.mlp.experts.2.gate_proj", "model.language_model.layers.18.mlp.experts.3.gate_proj", "model.language_model.layers.18.mlp.experts.4.gate_proj", "model.language_model.layers.18.mlp.experts.5.gate_proj", "model.language_model.layers.18.mlp.experts.6.gate_proj", "model.language_model.layers.18.mlp.experts.7.gate_proj", "model.language_model.layers.18.mlp.experts.8.gate_proj", "model.language_model.layers.18.mlp.experts.9.gate_proj", "model.language_model.layers.18.mlp.experts.10.gate_proj", "model.language_model.layers.18.mlp.experts.11.gate_proj", "model.language_model.layers.18.mlp.experts.12.gate_proj", "model.language_model.layers.18.mlp.experts.13.gate_proj", "model.language_model.layers.18.mlp.experts.14.gate_proj", "model.language_model.layers.18.mlp.experts.15.gate_proj", "model.language_model.layers.18.mlp.experts.16.gate_proj", "model.language_model.layers.18.mlp.experts.17.gate_proj", "model.language_model.layers.18.mlp.experts.18.gate_proj", "model.language_model.layers.18.mlp.experts.19.gate_proj", "model.language_model.layers.18.mlp.experts.20.gate_proj", "model.language_model.layers.18.mlp.experts.21.gate_proj", "model.language_model.layers.18.mlp.experts.22.gate_proj", "model.language_model.layers.18.mlp.experts.23.gate_proj", "model.language_model.layers.18.mlp.experts.24.gate_proj", "model.language_model.layers.18.mlp.experts.25.gate_proj", "model.language_model.layers.18.mlp.experts.26.gate_proj", "model.language_model.layers.18.mlp.experts.27.gate_proj", "model.language_model.layers.18.mlp.experts.28.gate_proj", "model.language_model.layers.18.mlp.experts.29.gate_proj", "model.language_model.layers.18.mlp.experts.30.gate_proj", "model.language_model.layers.18.mlp.experts.31.gate_proj", "model.language_model.layers.18.mlp.experts.32.gate_proj", "model.language_model.layers.18.mlp.experts.33.gate_proj", "model.language_model.layers.18.mlp.experts.34.gate_proj", "model.language_model.layers.18.mlp.experts.35.gate_proj", "model.language_model.layers.18.mlp.experts.36.gate_proj", "model.language_model.layers.18.mlp.experts.37.gate_proj", "model.language_model.layers.18.mlp.experts.38.gate_proj", "model.language_model.layers.18.mlp.experts.39.gate_proj", "model.language_model.layers.18.mlp.experts.40.gate_proj", "model.language_model.layers.18.mlp.experts.41.gate_proj", "model.language_model.layers.18.mlp.experts.42.gate_proj", "model.language_model.layers.18.mlp.experts.43.gate_proj", "model.language_model.layers.18.mlp.experts.44.gate_proj", "model.language_model.layers.18.mlp.experts.45.gate_proj", "model.language_model.layers.18.mlp.experts.46.gate_proj", "model.language_model.layers.18.mlp.experts.47.gate_proj", "model.language_model.layers.18.mlp.experts.48.gate_proj", "model.language_model.layers.18.mlp.experts.49.gate_proj", "model.language_model.layers.18.mlp.experts.50.gate_proj", "model.language_model.layers.18.mlp.experts.51.gate_proj", "model.language_model.layers.18.mlp.experts.52.gate_proj", "model.language_model.layers.18.mlp.experts.53.gate_proj", "model.language_model.layers.18.mlp.experts.54.gate_proj", "model.language_model.layers.18.mlp.experts.55.gate_proj", "model.language_model.layers.18.mlp.experts.56.gate_proj", "model.language_model.layers.18.mlp.experts.57.gate_proj", "model.language_model.layers.18.mlp.experts.58.gate_proj", "model.language_model.layers.18.mlp.experts.59.gate_proj", "model.language_model.layers.18.mlp.experts.60.gate_proj", "model.language_model.layers.18.mlp.experts.61.gate_proj", "model.language_model.layers.18.mlp.experts.62.gate_proj", "model.language_model.layers.18.mlp.experts.63.gate_proj", "model.language_model.layers.18.mlp.experts.64.gate_proj", "model.language_model.layers.18.mlp.experts.65.gate_proj", "model.language_model.layers.18.mlp.experts.66.gate_proj", "model.language_model.layers.18.mlp.experts.67.gate_proj", "model.language_model.layers.18.mlp.experts.68.gate_proj", "model.language_model.layers.18.mlp.experts.69.gate_proj", "model.language_model.layers.18.mlp.experts.70.gate_proj", "model.language_model.layers.18.mlp.experts.71.gate_proj", "model.language_model.layers.18.mlp.experts.72.gate_proj", "model.language_model.layers.18.mlp.experts.73.gate_proj", "model.language_model.layers.18.mlp.experts.74.gate_proj", "model.language_model.layers.18.mlp.experts.75.gate_proj", "model.language_model.layers.18.mlp.experts.76.gate_proj", "model.language_model.layers.18.mlp.experts.77.gate_proj", "model.language_model.layers.18.mlp.experts.78.gate_proj", "model.language_model.layers.18.mlp.experts.79.gate_proj", "model.language_model.layers.18.mlp.experts.80.gate_proj", "model.language_model.layers.18.mlp.experts.81.gate_proj", "model.language_model.layers.18.mlp.experts.82.gate_proj", "model.language_model.layers.18.mlp.experts.83.gate_proj", "model.language_model.layers.18.mlp.experts.84.gate_proj", "model.language_model.layers.18.mlp.experts.85.gate_proj", "model.language_model.layers.18.mlp.experts.86.gate_proj", "model.language_model.layers.18.mlp.experts.87.gate_proj", "model.language_model.layers.18.mlp.experts.88.gate_proj", "model.language_model.layers.18.mlp.experts.89.gate_proj", "model.language_model.layers.18.mlp.experts.90.gate_proj", "model.language_model.layers.18.mlp.experts.91.gate_proj", "model.language_model.layers.18.mlp.experts.92.gate_proj", "model.language_model.layers.18.mlp.experts.93.gate_proj", "model.language_model.layers.18.mlp.experts.94.gate_proj", "model.language_model.layers.18.mlp.experts.95.gate_proj", "model.language_model.layers.18.mlp.experts.96.gate_proj", "model.language_model.layers.18.mlp.experts.97.gate_proj", "model.language_model.layers.18.mlp.experts.98.gate_proj", "model.language_model.layers.18.mlp.experts.99.gate_proj", "model.language_model.layers.18.mlp.experts.100.gate_proj", "model.language_model.layers.18.mlp.experts.101.gate_proj", "model.language_model.layers.18.mlp.experts.102.gate_proj", "model.language_model.layers.18.mlp.experts.103.gate_proj", "model.language_model.layers.18.mlp.experts.104.gate_proj", "model.language_model.layers.18.mlp.experts.105.gate_proj", "model.language_model.layers.18.mlp.experts.106.gate_proj", "model.language_model.layers.18.mlp.experts.107.gate_proj", "model.language_model.layers.18.mlp.experts.108.gate_proj", "model.language_model.layers.18.mlp.experts.109.gate_proj", "model.language_model.layers.18.mlp.experts.110.gate_proj", "model.language_model.layers.18.mlp.experts.111.gate_proj", "model.language_model.layers.18.mlp.experts.112.gate_proj", "model.language_model.layers.18.mlp.experts.113.gate_proj", "model.language_model.layers.18.mlp.experts.114.gate_proj", "model.language_model.layers.18.mlp.experts.115.gate_proj", "model.language_model.layers.18.mlp.experts.116.gate_proj", "model.language_model.layers.18.mlp.experts.117.gate_proj", "model.language_model.layers.18.mlp.experts.118.gate_proj", "model.language_model.layers.18.mlp.experts.119.gate_proj", "model.language_model.layers.18.mlp.experts.120.gate_proj", "model.language_model.layers.18.mlp.experts.121.gate_proj", "model.language_model.layers.18.mlp.experts.122.gate_proj", "model.language_model.layers.18.mlp.experts.123.gate_proj", "model.language_model.layers.18.mlp.experts.124.gate_proj", "model.language_model.layers.18.mlp.experts.125.gate_proj", "model.language_model.layers.18.mlp.experts.126.gate_proj", "model.language_model.layers.18.mlp.experts.127.gate_proj", "model.language_model.layers.18.mlp.experts.0.up_proj", "model.language_model.layers.18.mlp.experts.1.up_proj", "model.language_model.layers.18.mlp.experts.2.up_proj", "model.language_model.layers.18.mlp.experts.3.up_proj", "model.language_model.layers.18.mlp.experts.4.up_proj", "model.language_model.layers.18.mlp.experts.5.up_proj", "model.language_model.layers.18.mlp.experts.6.up_proj", "model.language_model.layers.18.mlp.experts.7.up_proj", "model.language_model.layers.18.mlp.experts.8.up_proj", "model.language_model.layers.18.mlp.experts.9.up_proj", "model.language_model.layers.18.mlp.experts.10.up_proj", "model.language_model.layers.18.mlp.experts.11.up_proj", "model.language_model.layers.18.mlp.experts.12.up_proj", "model.language_model.layers.18.mlp.experts.13.up_proj", "model.language_model.layers.18.mlp.experts.14.up_proj", "model.language_model.layers.18.mlp.experts.15.up_proj", "model.language_model.layers.18.mlp.experts.16.up_proj", "model.language_model.layers.18.mlp.experts.17.up_proj", "model.language_model.layers.18.mlp.experts.18.up_proj", "model.language_model.layers.18.mlp.experts.19.up_proj", "model.language_model.layers.18.mlp.experts.20.up_proj", "model.language_model.layers.18.mlp.experts.21.up_proj", "model.language_model.layers.18.mlp.experts.22.up_proj", "model.language_model.layers.18.mlp.experts.23.up_proj", "model.language_model.layers.18.mlp.experts.24.up_proj", "model.language_model.layers.18.mlp.experts.25.up_proj", "model.language_model.layers.18.mlp.experts.26.up_proj", "model.language_model.layers.18.mlp.experts.27.up_proj", "model.language_model.layers.18.mlp.experts.28.up_proj", "model.language_model.layers.18.mlp.experts.29.up_proj", "model.language_model.layers.18.mlp.experts.30.up_proj", "model.language_model.layers.18.mlp.experts.31.up_proj", "model.language_model.layers.18.mlp.experts.32.up_proj", "model.language_model.layers.18.mlp.experts.33.up_proj", "model.language_model.layers.18.mlp.experts.34.up_proj", "model.language_model.layers.18.mlp.experts.35.up_proj", "model.language_model.layers.18.mlp.experts.36.up_proj", "model.language_model.layers.18.mlp.experts.37.up_proj", "model.language_model.layers.18.mlp.experts.38.up_proj", "model.language_model.layers.18.mlp.experts.39.up_proj", "model.language_model.layers.18.mlp.experts.40.up_proj", "model.language_model.layers.18.mlp.experts.41.up_proj", "model.language_model.layers.18.mlp.experts.42.up_proj", "model.language_model.layers.18.mlp.experts.43.up_proj", "model.language_model.layers.18.mlp.experts.44.up_proj", "model.language_model.layers.18.mlp.experts.45.up_proj", "model.language_model.layers.18.mlp.experts.46.up_proj", "model.language_model.layers.18.mlp.experts.47.up_proj", "model.language_model.layers.18.mlp.experts.48.up_proj", "model.language_model.layers.18.mlp.experts.49.up_proj", "model.language_model.layers.18.mlp.experts.50.up_proj", "model.language_model.layers.18.mlp.experts.51.up_proj", "model.language_model.layers.18.mlp.experts.52.up_proj", "model.language_model.layers.18.mlp.experts.53.up_proj", "model.language_model.layers.18.mlp.experts.54.up_proj", "model.language_model.layers.18.mlp.experts.55.up_proj", "model.language_model.layers.18.mlp.experts.56.up_proj", "model.language_model.layers.18.mlp.experts.57.up_proj", "model.language_model.layers.18.mlp.experts.58.up_proj", "model.language_model.layers.18.mlp.experts.59.up_proj", "model.language_model.layers.18.mlp.experts.60.up_proj", "model.language_model.layers.18.mlp.experts.61.up_proj", "model.language_model.layers.18.mlp.experts.62.up_proj", "model.language_model.layers.18.mlp.experts.63.up_proj", "model.language_model.layers.18.mlp.experts.64.up_proj", "model.language_model.layers.18.mlp.experts.65.up_proj", "model.language_model.layers.18.mlp.experts.66.up_proj", "model.language_model.layers.18.mlp.experts.67.up_proj", "model.language_model.layers.18.mlp.experts.68.up_proj", "model.language_model.layers.18.mlp.experts.69.up_proj", "model.language_model.layers.18.mlp.experts.70.up_proj", "model.language_model.layers.18.mlp.experts.71.up_proj", "model.language_model.layers.18.mlp.experts.72.up_proj", "model.language_model.layers.18.mlp.experts.73.up_proj", "model.language_model.layers.18.mlp.experts.74.up_proj", "model.language_model.layers.18.mlp.experts.75.up_proj", "model.language_model.layers.18.mlp.experts.76.up_proj", "model.language_model.layers.18.mlp.experts.77.up_proj", "model.language_model.layers.18.mlp.experts.78.up_proj", "model.language_model.layers.18.mlp.experts.79.up_proj", "model.language_model.layers.18.mlp.experts.80.up_proj", "model.language_model.layers.18.mlp.experts.81.up_proj", "model.language_model.layers.18.mlp.experts.82.up_proj", "model.language_model.layers.18.mlp.experts.83.up_proj", "model.language_model.layers.18.mlp.experts.84.up_proj", "model.language_model.layers.18.mlp.experts.85.up_proj", "model.language_model.layers.18.mlp.experts.86.up_proj", "model.language_model.layers.18.mlp.experts.87.up_proj", "model.language_model.layers.18.mlp.experts.88.up_proj", "model.language_model.layers.18.mlp.experts.89.up_proj", "model.language_model.layers.18.mlp.experts.90.up_proj", "model.language_model.layers.18.mlp.experts.91.up_proj", "model.language_model.layers.18.mlp.experts.92.up_proj", "model.language_model.layers.18.mlp.experts.93.up_proj", "model.language_model.layers.18.mlp.experts.94.up_proj", "model.language_model.layers.18.mlp.experts.95.up_proj", "model.language_model.layers.18.mlp.experts.96.up_proj", "model.language_model.layers.18.mlp.experts.97.up_proj", "model.language_model.layers.18.mlp.experts.98.up_proj", "model.language_model.layers.18.mlp.experts.99.up_proj", "model.language_model.layers.18.mlp.experts.100.up_proj", "model.language_model.layers.18.mlp.experts.101.up_proj", "model.language_model.layers.18.mlp.experts.102.up_proj", "model.language_model.layers.18.mlp.experts.103.up_proj", "model.language_model.layers.18.mlp.experts.104.up_proj", "model.language_model.layers.18.mlp.experts.105.up_proj", "model.language_model.layers.18.mlp.experts.106.up_proj", "model.language_model.layers.18.mlp.experts.107.up_proj", "model.language_model.layers.18.mlp.experts.108.up_proj", "model.language_model.layers.18.mlp.experts.109.up_proj", "model.language_model.layers.18.mlp.experts.110.up_proj", "model.language_model.layers.18.mlp.experts.111.up_proj", "model.language_model.layers.18.mlp.experts.112.up_proj", "model.language_model.layers.18.mlp.experts.113.up_proj", "model.language_model.layers.18.mlp.experts.114.up_proj", "model.language_model.layers.18.mlp.experts.115.up_proj", "model.language_model.layers.18.mlp.experts.116.up_proj", "model.language_model.layers.18.mlp.experts.117.up_proj", "model.language_model.layers.18.mlp.experts.118.up_proj", "model.language_model.layers.18.mlp.experts.119.up_proj", "model.language_model.layers.18.mlp.experts.120.up_proj", "model.language_model.layers.18.mlp.experts.121.up_proj", "model.language_model.layers.18.mlp.experts.122.up_proj", "model.language_model.layers.18.mlp.experts.123.up_proj", "model.language_model.layers.18.mlp.experts.124.up_proj", "model.language_model.layers.18.mlp.experts.125.up_proj", "model.language_model.layers.18.mlp.experts.126.up_proj", "model.language_model.layers.18.mlp.experts.127.up_proj", "model.language_model.layers.18.mlp.experts.0.down_proj", "model.language_model.layers.18.mlp.experts.1.down_proj", "model.language_model.layers.18.mlp.experts.2.down_proj", "model.language_model.layers.18.mlp.experts.3.down_proj", "model.language_model.layers.18.mlp.experts.4.down_proj", "model.language_model.layers.18.mlp.experts.5.down_proj", "model.language_model.layers.18.mlp.experts.6.down_proj", "model.language_model.layers.18.mlp.experts.7.down_proj", "model.language_model.layers.18.mlp.experts.8.down_proj", "model.language_model.layers.18.mlp.experts.9.down_proj", "model.language_model.layers.18.mlp.experts.10.down_proj", "model.language_model.layers.18.mlp.experts.11.down_proj", "model.language_model.layers.18.mlp.experts.12.down_proj", "model.language_model.layers.18.mlp.experts.13.down_proj", "model.language_model.layers.18.mlp.experts.14.down_proj", "model.language_model.layers.18.mlp.experts.15.down_proj", "model.language_model.layers.18.mlp.experts.16.down_proj", "model.language_model.layers.18.mlp.experts.17.down_proj", "model.language_model.layers.18.mlp.experts.18.down_proj", "model.language_model.layers.18.mlp.experts.19.down_proj", "model.language_model.layers.18.mlp.experts.20.down_proj", "model.language_model.layers.18.mlp.experts.21.down_proj", "model.language_model.layers.18.mlp.experts.22.down_proj", "model.language_model.layers.18.mlp.experts.23.down_proj", "model.language_model.layers.18.mlp.experts.24.down_proj", "model.language_model.layers.18.mlp.experts.25.down_proj", "model.language_model.layers.18.mlp.experts.26.down_proj", "model.language_model.layers.18.mlp.experts.27.down_proj", "model.language_model.layers.18.mlp.experts.28.down_proj", "model.language_model.layers.18.mlp.experts.29.down_proj", "model.language_model.layers.18.mlp.experts.30.down_proj", "model.language_model.layers.18.mlp.experts.31.down_proj", "model.language_model.layers.18.mlp.experts.32.down_proj", "model.language_model.layers.18.mlp.experts.33.down_proj", "model.language_model.layers.18.mlp.experts.34.down_proj", "model.language_model.layers.18.mlp.experts.35.down_proj", "model.language_model.layers.18.mlp.experts.36.down_proj", "model.language_model.layers.18.mlp.experts.37.down_proj", "model.language_model.layers.18.mlp.experts.38.down_proj", "model.language_model.layers.18.mlp.experts.39.down_proj", "model.language_model.layers.18.mlp.experts.40.down_proj", "model.language_model.layers.18.mlp.experts.41.down_proj", "model.language_model.layers.18.mlp.experts.42.down_proj", "model.language_model.layers.18.mlp.experts.43.down_proj", "model.language_model.layers.18.mlp.experts.44.down_proj", "model.language_model.layers.18.mlp.experts.45.down_proj", "model.language_model.layers.18.mlp.experts.46.down_proj", "model.language_model.layers.18.mlp.experts.47.down_proj", "model.language_model.layers.18.mlp.experts.48.down_proj", "model.language_model.layers.18.mlp.experts.49.down_proj", "model.language_model.layers.18.mlp.experts.50.down_proj", "model.language_model.layers.18.mlp.experts.51.down_proj", "model.language_model.layers.18.mlp.experts.52.down_proj", "model.language_model.layers.18.mlp.experts.53.down_proj", "model.language_model.layers.18.mlp.experts.54.down_proj", "model.language_model.layers.18.mlp.experts.55.down_proj", "model.language_model.layers.18.mlp.experts.56.down_proj", "model.language_model.layers.18.mlp.experts.57.down_proj", "model.language_model.layers.18.mlp.experts.58.down_proj", "model.language_model.layers.18.mlp.experts.59.down_proj", "model.language_model.layers.18.mlp.experts.60.down_proj", "model.language_model.layers.18.mlp.experts.61.down_proj", "model.language_model.layers.18.mlp.experts.62.down_proj", "model.language_model.layers.18.mlp.experts.63.down_proj", "model.language_model.layers.18.mlp.experts.64.down_proj", "model.language_model.layers.18.mlp.experts.65.down_proj", "model.language_model.layers.18.mlp.experts.66.down_proj", "model.language_model.layers.18.mlp.experts.67.down_proj", "model.language_model.layers.18.mlp.experts.68.down_proj", "model.language_model.layers.18.mlp.experts.69.down_proj", "model.language_model.layers.18.mlp.experts.70.down_proj", "model.language_model.layers.18.mlp.experts.71.down_proj", "model.language_model.layers.18.mlp.experts.72.down_proj", "model.language_model.layers.18.mlp.experts.73.down_proj", "model.language_model.layers.18.mlp.experts.74.down_proj", "model.language_model.layers.18.mlp.experts.75.down_proj", "model.language_model.layers.18.mlp.experts.76.down_proj", "model.language_model.layers.18.mlp.experts.77.down_proj", "model.language_model.layers.18.mlp.experts.78.down_proj", "model.language_model.layers.18.mlp.experts.79.down_proj", "model.language_model.layers.18.mlp.experts.80.down_proj", "model.language_model.layers.18.mlp.experts.81.down_proj", "model.language_model.layers.18.mlp.experts.82.down_proj", "model.language_model.layers.18.mlp.experts.83.down_proj", "model.language_model.layers.18.mlp.experts.84.down_proj", "model.language_model.layers.18.mlp.experts.85.down_proj", "model.language_model.layers.18.mlp.experts.86.down_proj", "model.language_model.layers.18.mlp.experts.87.down_proj", "model.language_model.layers.18.mlp.experts.88.down_proj", "model.language_model.layers.18.mlp.experts.89.down_proj", "model.language_model.layers.18.mlp.experts.90.down_proj", "model.language_model.layers.18.mlp.experts.91.down_proj", "model.language_model.layers.18.mlp.experts.92.down_proj", "model.language_model.layers.18.mlp.experts.93.down_proj", "model.language_model.layers.18.mlp.experts.94.down_proj", "model.language_model.layers.18.mlp.experts.95.down_proj", "model.language_model.layers.18.mlp.experts.96.down_proj", "model.language_model.layers.18.mlp.experts.97.down_proj", "model.language_model.layers.18.mlp.experts.98.down_proj", "model.language_model.layers.18.mlp.experts.99.down_proj", "model.language_model.layers.18.mlp.experts.100.down_proj", "model.language_model.layers.18.mlp.experts.101.down_proj", "model.language_model.layers.18.mlp.experts.102.down_proj", "model.language_model.layers.18.mlp.experts.103.down_proj", "model.language_model.layers.18.mlp.experts.104.down_proj", "model.language_model.layers.18.mlp.experts.105.down_proj", "model.language_model.layers.18.mlp.experts.106.down_proj", "model.language_model.layers.18.mlp.experts.107.down_proj", "model.language_model.layers.18.mlp.experts.108.down_proj", "model.language_model.layers.18.mlp.experts.109.down_proj", "model.language_model.layers.18.mlp.experts.110.down_proj", "model.language_model.layers.18.mlp.experts.111.down_proj", "model.language_model.layers.18.mlp.experts.112.down_proj", "model.language_model.layers.18.mlp.experts.113.down_proj", "model.language_model.layers.18.mlp.experts.114.down_proj", "model.language_model.layers.18.mlp.experts.115.down_proj", "model.language_model.layers.18.mlp.experts.116.down_proj", "model.language_model.layers.18.mlp.experts.117.down_proj", "model.language_model.layers.18.mlp.experts.118.down_proj", "model.language_model.layers.18.mlp.experts.119.down_proj", "model.language_model.layers.18.mlp.experts.120.down_proj", "model.language_model.layers.18.mlp.experts.121.down_proj", "model.language_model.layers.18.mlp.experts.122.down_proj", "model.language_model.layers.18.mlp.experts.123.down_proj", "model.language_model.layers.18.mlp.experts.124.down_proj", "model.language_model.layers.18.mlp.experts.125.down_proj", "model.language_model.layers.18.mlp.experts.126.down_proj", "model.language_model.layers.18.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": 0.0013839796185493441, "dbits": 603979776 } ] }, { "idx": 38, "layers": [ "model.language_model.layers.19.self_attn.q_proj", "model.language_model.layers.19.self_attn.k_proj", "model.language_model.layers.19.self_attn.v_proj", "model.language_model.layers.19.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0011490038596093655, "dbits": 18874368 } ] }, { "idx": 39, "layers": [ "model.language_model.layers.19.mlp.experts.0.gate_proj", "model.language_model.layers.19.mlp.experts.1.gate_proj", "model.language_model.layers.19.mlp.experts.2.gate_proj", "model.language_model.layers.19.mlp.experts.3.gate_proj", "model.language_model.layers.19.mlp.experts.4.gate_proj", "model.language_model.layers.19.mlp.experts.5.gate_proj", "model.language_model.layers.19.mlp.experts.6.gate_proj", "model.language_model.layers.19.mlp.experts.7.gate_proj", "model.language_model.layers.19.mlp.experts.8.gate_proj", "model.language_model.layers.19.mlp.experts.9.gate_proj", "model.language_model.layers.19.mlp.experts.10.gate_proj", "model.language_model.layers.19.mlp.experts.11.gate_proj", "model.language_model.layers.19.mlp.experts.12.gate_proj", "model.language_model.layers.19.mlp.experts.13.gate_proj", "model.language_model.layers.19.mlp.experts.14.gate_proj", "model.language_model.layers.19.mlp.experts.15.gate_proj", "model.language_model.layers.19.mlp.experts.16.gate_proj", "model.language_model.layers.19.mlp.experts.17.gate_proj", "model.language_model.layers.19.mlp.experts.18.gate_proj", "model.language_model.layers.19.mlp.experts.19.gate_proj", "model.language_model.layers.19.mlp.experts.20.gate_proj", "model.language_model.layers.19.mlp.experts.21.gate_proj", "model.language_model.layers.19.mlp.experts.22.gate_proj", "model.language_model.layers.19.mlp.experts.23.gate_proj", "model.language_model.layers.19.mlp.experts.24.gate_proj", "model.language_model.layers.19.mlp.experts.25.gate_proj", "model.language_model.layers.19.mlp.experts.26.gate_proj", "model.language_model.layers.19.mlp.experts.27.gate_proj", "model.language_model.layers.19.mlp.experts.28.gate_proj", "model.language_model.layers.19.mlp.experts.29.gate_proj", "model.language_model.layers.19.mlp.experts.30.gate_proj", "model.language_model.layers.19.mlp.experts.31.gate_proj", "model.language_model.layers.19.mlp.experts.32.gate_proj", "model.language_model.layers.19.mlp.experts.33.gate_proj", "model.language_model.layers.19.mlp.experts.34.gate_proj", "model.language_model.layers.19.mlp.experts.35.gate_proj", "model.language_model.layers.19.mlp.experts.36.gate_proj", "model.language_model.layers.19.mlp.experts.37.gate_proj", "model.language_model.layers.19.mlp.experts.38.gate_proj", "model.language_model.layers.19.mlp.experts.39.gate_proj", "model.language_model.layers.19.mlp.experts.40.gate_proj", "model.language_model.layers.19.mlp.experts.41.gate_proj", "model.language_model.layers.19.mlp.experts.42.gate_proj", "model.language_model.layers.19.mlp.experts.43.gate_proj", "model.language_model.layers.19.mlp.experts.44.gate_proj", "model.language_model.layers.19.mlp.experts.45.gate_proj", "model.language_model.layers.19.mlp.experts.46.gate_proj", "model.language_model.layers.19.mlp.experts.47.gate_proj", "model.language_model.layers.19.mlp.experts.48.gate_proj", "model.language_model.layers.19.mlp.experts.49.gate_proj", "model.language_model.layers.19.mlp.experts.50.gate_proj", "model.language_model.layers.19.mlp.experts.51.gate_proj", "model.language_model.layers.19.mlp.experts.52.gate_proj", "model.language_model.layers.19.mlp.experts.53.gate_proj", "model.language_model.layers.19.mlp.experts.54.gate_proj", "model.language_model.layers.19.mlp.experts.55.gate_proj", "model.language_model.layers.19.mlp.experts.56.gate_proj", "model.language_model.layers.19.mlp.experts.57.gate_proj", "model.language_model.layers.19.mlp.experts.58.gate_proj", "model.language_model.layers.19.mlp.experts.59.gate_proj", "model.language_model.layers.19.mlp.experts.60.gate_proj", "model.language_model.layers.19.mlp.experts.61.gate_proj", "model.language_model.layers.19.mlp.experts.62.gate_proj", "model.language_model.layers.19.mlp.experts.63.gate_proj", "model.language_model.layers.19.mlp.experts.64.gate_proj", "model.language_model.layers.19.mlp.experts.65.gate_proj", "model.language_model.layers.19.mlp.experts.66.gate_proj", "model.language_model.layers.19.mlp.experts.67.gate_proj", "model.language_model.layers.19.mlp.experts.68.gate_proj", "model.language_model.layers.19.mlp.experts.69.gate_proj", "model.language_model.layers.19.mlp.experts.70.gate_proj", "model.language_model.layers.19.mlp.experts.71.gate_proj", "model.language_model.layers.19.mlp.experts.72.gate_proj", "model.language_model.layers.19.mlp.experts.73.gate_proj", "model.language_model.layers.19.mlp.experts.74.gate_proj", "model.language_model.layers.19.mlp.experts.75.gate_proj", "model.language_model.layers.19.mlp.experts.76.gate_proj", "model.language_model.layers.19.mlp.experts.77.gate_proj", "model.language_model.layers.19.mlp.experts.78.gate_proj", "model.language_model.layers.19.mlp.experts.79.gate_proj", "model.language_model.layers.19.mlp.experts.80.gate_proj", "model.language_model.layers.19.mlp.experts.81.gate_proj", "model.language_model.layers.19.mlp.experts.82.gate_proj", "model.language_model.layers.19.mlp.experts.83.gate_proj", "model.language_model.layers.19.mlp.experts.84.gate_proj", "model.language_model.layers.19.mlp.experts.85.gate_proj", "model.language_model.layers.19.mlp.experts.86.gate_proj", "model.language_model.layers.19.mlp.experts.87.gate_proj", "model.language_model.layers.19.mlp.experts.88.gate_proj", "model.language_model.layers.19.mlp.experts.89.gate_proj", "model.language_model.layers.19.mlp.experts.90.gate_proj", "model.language_model.layers.19.mlp.experts.91.gate_proj", "model.language_model.layers.19.mlp.experts.92.gate_proj", "model.language_model.layers.19.mlp.experts.93.gate_proj", "model.language_model.layers.19.mlp.experts.94.gate_proj", "model.language_model.layers.19.mlp.experts.95.gate_proj", "model.language_model.layers.19.mlp.experts.96.gate_proj", "model.language_model.layers.19.mlp.experts.97.gate_proj", "model.language_model.layers.19.mlp.experts.98.gate_proj", "model.language_model.layers.19.mlp.experts.99.gate_proj", "model.language_model.layers.19.mlp.experts.100.gate_proj", "model.language_model.layers.19.mlp.experts.101.gate_proj", "model.language_model.layers.19.mlp.experts.102.gate_proj", "model.language_model.layers.19.mlp.experts.103.gate_proj", "model.language_model.layers.19.mlp.experts.104.gate_proj", "model.language_model.layers.19.mlp.experts.105.gate_proj", "model.language_model.layers.19.mlp.experts.106.gate_proj", "model.language_model.layers.19.mlp.experts.107.gate_proj", "model.language_model.layers.19.mlp.experts.108.gate_proj", "model.language_model.layers.19.mlp.experts.109.gate_proj", "model.language_model.layers.19.mlp.experts.110.gate_proj", "model.language_model.layers.19.mlp.experts.111.gate_proj", "model.language_model.layers.19.mlp.experts.112.gate_proj", "model.language_model.layers.19.mlp.experts.113.gate_proj", "model.language_model.layers.19.mlp.experts.114.gate_proj", "model.language_model.layers.19.mlp.experts.115.gate_proj", "model.language_model.layers.19.mlp.experts.116.gate_proj", "model.language_model.layers.19.mlp.experts.117.gate_proj", "model.language_model.layers.19.mlp.experts.118.gate_proj", "model.language_model.layers.19.mlp.experts.119.gate_proj", "model.language_model.layers.19.mlp.experts.120.gate_proj", "model.language_model.layers.19.mlp.experts.121.gate_proj", "model.language_model.layers.19.mlp.experts.122.gate_proj", "model.language_model.layers.19.mlp.experts.123.gate_proj", "model.language_model.layers.19.mlp.experts.124.gate_proj", "model.language_model.layers.19.mlp.experts.125.gate_proj", "model.language_model.layers.19.mlp.experts.126.gate_proj", "model.language_model.layers.19.mlp.experts.127.gate_proj", "model.language_model.layers.19.mlp.experts.0.up_proj", "model.language_model.layers.19.mlp.experts.1.up_proj", "model.language_model.layers.19.mlp.experts.2.up_proj", "model.language_model.layers.19.mlp.experts.3.up_proj", "model.language_model.layers.19.mlp.experts.4.up_proj", "model.language_model.layers.19.mlp.experts.5.up_proj", "model.language_model.layers.19.mlp.experts.6.up_proj", "model.language_model.layers.19.mlp.experts.7.up_proj", "model.language_model.layers.19.mlp.experts.8.up_proj", "model.language_model.layers.19.mlp.experts.9.up_proj", "model.language_model.layers.19.mlp.experts.10.up_proj", "model.language_model.layers.19.mlp.experts.11.up_proj", "model.language_model.layers.19.mlp.experts.12.up_proj", "model.language_model.layers.19.mlp.experts.13.up_proj", "model.language_model.layers.19.mlp.experts.14.up_proj", "model.language_model.layers.19.mlp.experts.15.up_proj", "model.language_model.layers.19.mlp.experts.16.up_proj", "model.language_model.layers.19.mlp.experts.17.up_proj", "model.language_model.layers.19.mlp.experts.18.up_proj", "model.language_model.layers.19.mlp.experts.19.up_proj", "model.language_model.layers.19.mlp.experts.20.up_proj", "model.language_model.layers.19.mlp.experts.21.up_proj", "model.language_model.layers.19.mlp.experts.22.up_proj", "model.language_model.layers.19.mlp.experts.23.up_proj", "model.language_model.layers.19.mlp.experts.24.up_proj", "model.language_model.layers.19.mlp.experts.25.up_proj", "model.language_model.layers.19.mlp.experts.26.up_proj", "model.language_model.layers.19.mlp.experts.27.up_proj", "model.language_model.layers.19.mlp.experts.28.up_proj", "model.language_model.layers.19.mlp.experts.29.up_proj", "model.language_model.layers.19.mlp.experts.30.up_proj", "model.language_model.layers.19.mlp.experts.31.up_proj", "model.language_model.layers.19.mlp.experts.32.up_proj", "model.language_model.layers.19.mlp.experts.33.up_proj", "model.language_model.layers.19.mlp.experts.34.up_proj", "model.language_model.layers.19.mlp.experts.35.up_proj", "model.language_model.layers.19.mlp.experts.36.up_proj", "model.language_model.layers.19.mlp.experts.37.up_proj", "model.language_model.layers.19.mlp.experts.38.up_proj", "model.language_model.layers.19.mlp.experts.39.up_proj", "model.language_model.layers.19.mlp.experts.40.up_proj", "model.language_model.layers.19.mlp.experts.41.up_proj", "model.language_model.layers.19.mlp.experts.42.up_proj", "model.language_model.layers.19.mlp.experts.43.up_proj", "model.language_model.layers.19.mlp.experts.44.up_proj", "model.language_model.layers.19.mlp.experts.45.up_proj", "model.language_model.layers.19.mlp.experts.46.up_proj", "model.language_model.layers.19.mlp.experts.47.up_proj", "model.language_model.layers.19.mlp.experts.48.up_proj", "model.language_model.layers.19.mlp.experts.49.up_proj", "model.language_model.layers.19.mlp.experts.50.up_proj", "model.language_model.layers.19.mlp.experts.51.up_proj", "model.language_model.layers.19.mlp.experts.52.up_proj", "model.language_model.layers.19.mlp.experts.53.up_proj", "model.language_model.layers.19.mlp.experts.54.up_proj", "model.language_model.layers.19.mlp.experts.55.up_proj", "model.language_model.layers.19.mlp.experts.56.up_proj", "model.language_model.layers.19.mlp.experts.57.up_proj", "model.language_model.layers.19.mlp.experts.58.up_proj", "model.language_model.layers.19.mlp.experts.59.up_proj", "model.language_model.layers.19.mlp.experts.60.up_proj", "model.language_model.layers.19.mlp.experts.61.up_proj", "model.language_model.layers.19.mlp.experts.62.up_proj", "model.language_model.layers.19.mlp.experts.63.up_proj", "model.language_model.layers.19.mlp.experts.64.up_proj", "model.language_model.layers.19.mlp.experts.65.up_proj", "model.language_model.layers.19.mlp.experts.66.up_proj", "model.language_model.layers.19.mlp.experts.67.up_proj", "model.language_model.layers.19.mlp.experts.68.up_proj", "model.language_model.layers.19.mlp.experts.69.up_proj", "model.language_model.layers.19.mlp.experts.70.up_proj", "model.language_model.layers.19.mlp.experts.71.up_proj", "model.language_model.layers.19.mlp.experts.72.up_proj", "model.language_model.layers.19.mlp.experts.73.up_proj", "model.language_model.layers.19.mlp.experts.74.up_proj", "model.language_model.layers.19.mlp.experts.75.up_proj", "model.language_model.layers.19.mlp.experts.76.up_proj", "model.language_model.layers.19.mlp.experts.77.up_proj", "model.language_model.layers.19.mlp.experts.78.up_proj", "model.language_model.layers.19.mlp.experts.79.up_proj", "model.language_model.layers.19.mlp.experts.80.up_proj", "model.language_model.layers.19.mlp.experts.81.up_proj", "model.language_model.layers.19.mlp.experts.82.up_proj", "model.language_model.layers.19.mlp.experts.83.up_proj", "model.language_model.layers.19.mlp.experts.84.up_proj", "model.language_model.layers.19.mlp.experts.85.up_proj", "model.language_model.layers.19.mlp.experts.86.up_proj", "model.language_model.layers.19.mlp.experts.87.up_proj", "model.language_model.layers.19.mlp.experts.88.up_proj", "model.language_model.layers.19.mlp.experts.89.up_proj", "model.language_model.layers.19.mlp.experts.90.up_proj", "model.language_model.layers.19.mlp.experts.91.up_proj", "model.language_model.layers.19.mlp.experts.92.up_proj", "model.language_model.layers.19.mlp.experts.93.up_proj", "model.language_model.layers.19.mlp.experts.94.up_proj", "model.language_model.layers.19.mlp.experts.95.up_proj", "model.language_model.layers.19.mlp.experts.96.up_proj", "model.language_model.layers.19.mlp.experts.97.up_proj", "model.language_model.layers.19.mlp.experts.98.up_proj", "model.language_model.layers.19.mlp.experts.99.up_proj", "model.language_model.layers.19.mlp.experts.100.up_proj", "model.language_model.layers.19.mlp.experts.101.up_proj", "model.language_model.layers.19.mlp.experts.102.up_proj", "model.language_model.layers.19.mlp.experts.103.up_proj", "model.language_model.layers.19.mlp.experts.104.up_proj", "model.language_model.layers.19.mlp.experts.105.up_proj", "model.language_model.layers.19.mlp.experts.106.up_proj", "model.language_model.layers.19.mlp.experts.107.up_proj", "model.language_model.layers.19.mlp.experts.108.up_proj", "model.language_model.layers.19.mlp.experts.109.up_proj", "model.language_model.layers.19.mlp.experts.110.up_proj", "model.language_model.layers.19.mlp.experts.111.up_proj", "model.language_model.layers.19.mlp.experts.112.up_proj", "model.language_model.layers.19.mlp.experts.113.up_proj", "model.language_model.layers.19.mlp.experts.114.up_proj", "model.language_model.layers.19.mlp.experts.115.up_proj", "model.language_model.layers.19.mlp.experts.116.up_proj", "model.language_model.layers.19.mlp.experts.117.up_proj", "model.language_model.layers.19.mlp.experts.118.up_proj", "model.language_model.layers.19.mlp.experts.119.up_proj", "model.language_model.layers.19.mlp.experts.120.up_proj", "model.language_model.layers.19.mlp.experts.121.up_proj", "model.language_model.layers.19.mlp.experts.122.up_proj", "model.language_model.layers.19.mlp.experts.123.up_proj", "model.language_model.layers.19.mlp.experts.124.up_proj", "model.language_model.layers.19.mlp.experts.125.up_proj", "model.language_model.layers.19.mlp.experts.126.up_proj", "model.language_model.layers.19.mlp.experts.127.up_proj", "model.language_model.layers.19.mlp.experts.0.down_proj", "model.language_model.layers.19.mlp.experts.1.down_proj", "model.language_model.layers.19.mlp.experts.2.down_proj", "model.language_model.layers.19.mlp.experts.3.down_proj", "model.language_model.layers.19.mlp.experts.4.down_proj", "model.language_model.layers.19.mlp.experts.5.down_proj", "model.language_model.layers.19.mlp.experts.6.down_proj", "model.language_model.layers.19.mlp.experts.7.down_proj", "model.language_model.layers.19.mlp.experts.8.down_proj", "model.language_model.layers.19.mlp.experts.9.down_proj", "model.language_model.layers.19.mlp.experts.10.down_proj", "model.language_model.layers.19.mlp.experts.11.down_proj", "model.language_model.layers.19.mlp.experts.12.down_proj", "model.language_model.layers.19.mlp.experts.13.down_proj", "model.language_model.layers.19.mlp.experts.14.down_proj", "model.language_model.layers.19.mlp.experts.15.down_proj", "model.language_model.layers.19.mlp.experts.16.down_proj", "model.language_model.layers.19.mlp.experts.17.down_proj", "model.language_model.layers.19.mlp.experts.18.down_proj", "model.language_model.layers.19.mlp.experts.19.down_proj", "model.language_model.layers.19.mlp.experts.20.down_proj", "model.language_model.layers.19.mlp.experts.21.down_proj", "model.language_model.layers.19.mlp.experts.22.down_proj", "model.language_model.layers.19.mlp.experts.23.down_proj", "model.language_model.layers.19.mlp.experts.24.down_proj", "model.language_model.layers.19.mlp.experts.25.down_proj", "model.language_model.layers.19.mlp.experts.26.down_proj", "model.language_model.layers.19.mlp.experts.27.down_proj", "model.language_model.layers.19.mlp.experts.28.down_proj", "model.language_model.layers.19.mlp.experts.29.down_proj", "model.language_model.layers.19.mlp.experts.30.down_proj", "model.language_model.layers.19.mlp.experts.31.down_proj", "model.language_model.layers.19.mlp.experts.32.down_proj", "model.language_model.layers.19.mlp.experts.33.down_proj", "model.language_model.layers.19.mlp.experts.34.down_proj", "model.language_model.layers.19.mlp.experts.35.down_proj", "model.language_model.layers.19.mlp.experts.36.down_proj", "model.language_model.layers.19.mlp.experts.37.down_proj", "model.language_model.layers.19.mlp.experts.38.down_proj", "model.language_model.layers.19.mlp.experts.39.down_proj", "model.language_model.layers.19.mlp.experts.40.down_proj", "model.language_model.layers.19.mlp.experts.41.down_proj", "model.language_model.layers.19.mlp.experts.42.down_proj", "model.language_model.layers.19.mlp.experts.43.down_proj", "model.language_model.layers.19.mlp.experts.44.down_proj", "model.language_model.layers.19.mlp.experts.45.down_proj", "model.language_model.layers.19.mlp.experts.46.down_proj", "model.language_model.layers.19.mlp.experts.47.down_proj", "model.language_model.layers.19.mlp.experts.48.down_proj", "model.language_model.layers.19.mlp.experts.49.down_proj", "model.language_model.layers.19.mlp.experts.50.down_proj", "model.language_model.layers.19.mlp.experts.51.down_proj", "model.language_model.layers.19.mlp.experts.52.down_proj", "model.language_model.layers.19.mlp.experts.53.down_proj", "model.language_model.layers.19.mlp.experts.54.down_proj", "model.language_model.layers.19.mlp.experts.55.down_proj", "model.language_model.layers.19.mlp.experts.56.down_proj", "model.language_model.layers.19.mlp.experts.57.down_proj", "model.language_model.layers.19.mlp.experts.58.down_proj", "model.language_model.layers.19.mlp.experts.59.down_proj", "model.language_model.layers.19.mlp.experts.60.down_proj", "model.language_model.layers.19.mlp.experts.61.down_proj", "model.language_model.layers.19.mlp.experts.62.down_proj", "model.language_model.layers.19.mlp.experts.63.down_proj", "model.language_model.layers.19.mlp.experts.64.down_proj", "model.language_model.layers.19.mlp.experts.65.down_proj", "model.language_model.layers.19.mlp.experts.66.down_proj", "model.language_model.layers.19.mlp.experts.67.down_proj", "model.language_model.layers.19.mlp.experts.68.down_proj", "model.language_model.layers.19.mlp.experts.69.down_proj", "model.language_model.layers.19.mlp.experts.70.down_proj", "model.language_model.layers.19.mlp.experts.71.down_proj", "model.language_model.layers.19.mlp.experts.72.down_proj", "model.language_model.layers.19.mlp.experts.73.down_proj", "model.language_model.layers.19.mlp.experts.74.down_proj", "model.language_model.layers.19.mlp.experts.75.down_proj", "model.language_model.layers.19.mlp.experts.76.down_proj", "model.language_model.layers.19.mlp.experts.77.down_proj", "model.language_model.layers.19.mlp.experts.78.down_proj", "model.language_model.layers.19.mlp.experts.79.down_proj", "model.language_model.layers.19.mlp.experts.80.down_proj", "model.language_model.layers.19.mlp.experts.81.down_proj", "model.language_model.layers.19.mlp.experts.82.down_proj", "model.language_model.layers.19.mlp.experts.83.down_proj", "model.language_model.layers.19.mlp.experts.84.down_proj", "model.language_model.layers.19.mlp.experts.85.down_proj", "model.language_model.layers.19.mlp.experts.86.down_proj", "model.language_model.layers.19.mlp.experts.87.down_proj", "model.language_model.layers.19.mlp.experts.88.down_proj", "model.language_model.layers.19.mlp.experts.89.down_proj", "model.language_model.layers.19.mlp.experts.90.down_proj", "model.language_model.layers.19.mlp.experts.91.down_proj", "model.language_model.layers.19.mlp.experts.92.down_proj", "model.language_model.layers.19.mlp.experts.93.down_proj", "model.language_model.layers.19.mlp.experts.94.down_proj", "model.language_model.layers.19.mlp.experts.95.down_proj", "model.language_model.layers.19.mlp.experts.96.down_proj", "model.language_model.layers.19.mlp.experts.97.down_proj", "model.language_model.layers.19.mlp.experts.98.down_proj", "model.language_model.layers.19.mlp.experts.99.down_proj", "model.language_model.layers.19.mlp.experts.100.down_proj", "model.language_model.layers.19.mlp.experts.101.down_proj", "model.language_model.layers.19.mlp.experts.102.down_proj", "model.language_model.layers.19.mlp.experts.103.down_proj", "model.language_model.layers.19.mlp.experts.104.down_proj", "model.language_model.layers.19.mlp.experts.105.down_proj", "model.language_model.layers.19.mlp.experts.106.down_proj", "model.language_model.layers.19.mlp.experts.107.down_proj", "model.language_model.layers.19.mlp.experts.108.down_proj", "model.language_model.layers.19.mlp.experts.109.down_proj", "model.language_model.layers.19.mlp.experts.110.down_proj", "model.language_model.layers.19.mlp.experts.111.down_proj", "model.language_model.layers.19.mlp.experts.112.down_proj", "model.language_model.layers.19.mlp.experts.113.down_proj", "model.language_model.layers.19.mlp.experts.114.down_proj", "model.language_model.layers.19.mlp.experts.115.down_proj", "model.language_model.layers.19.mlp.experts.116.down_proj", "model.language_model.layers.19.mlp.experts.117.down_proj", "model.language_model.layers.19.mlp.experts.118.down_proj", "model.language_model.layers.19.mlp.experts.119.down_proj", "model.language_model.layers.19.mlp.experts.120.down_proj", "model.language_model.layers.19.mlp.experts.121.down_proj", "model.language_model.layers.19.mlp.experts.122.down_proj", "model.language_model.layers.19.mlp.experts.123.down_proj", "model.language_model.layers.19.mlp.experts.124.down_proj", "model.language_model.layers.19.mlp.experts.125.down_proj", "model.language_model.layers.19.mlp.experts.126.down_proj", "model.language_model.layers.19.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.0020736997947096852, "dbits": 603979776 } ] }, { "idx": 40, "layers": [ "model.language_model.layers.20.self_attn.q_proj", "model.language_model.layers.20.self_attn.k_proj", "model.language_model.layers.20.self_attn.v_proj", "model.language_model.layers.20.self_attn.o_proj" ], "candidates": [ { "dkld": -7.30209983885316e-05, "dbits": 18874368 } ] }, { "idx": 41, "layers": [ "model.language_model.layers.20.mlp.experts.0.gate_proj", "model.language_model.layers.20.mlp.experts.1.gate_proj", "model.language_model.layers.20.mlp.experts.2.gate_proj", "model.language_model.layers.20.mlp.experts.3.gate_proj", "model.language_model.layers.20.mlp.experts.4.gate_proj", "model.language_model.layers.20.mlp.experts.5.gate_proj", "model.language_model.layers.20.mlp.experts.6.gate_proj", "model.language_model.layers.20.mlp.experts.7.gate_proj", "model.language_model.layers.20.mlp.experts.8.gate_proj", "model.language_model.layers.20.mlp.experts.9.gate_proj", "model.language_model.layers.20.mlp.experts.10.gate_proj", "model.language_model.layers.20.mlp.experts.11.gate_proj", "model.language_model.layers.20.mlp.experts.12.gate_proj", "model.language_model.layers.20.mlp.experts.13.gate_proj", "model.language_model.layers.20.mlp.experts.14.gate_proj", "model.language_model.layers.20.mlp.experts.15.gate_proj", "model.language_model.layers.20.mlp.experts.16.gate_proj", "model.language_model.layers.20.mlp.experts.17.gate_proj", "model.language_model.layers.20.mlp.experts.18.gate_proj", "model.language_model.layers.20.mlp.experts.19.gate_proj", "model.language_model.layers.20.mlp.experts.20.gate_proj", "model.language_model.layers.20.mlp.experts.21.gate_proj", "model.language_model.layers.20.mlp.experts.22.gate_proj", "model.language_model.layers.20.mlp.experts.23.gate_proj", "model.language_model.layers.20.mlp.experts.24.gate_proj", "model.language_model.layers.20.mlp.experts.25.gate_proj", "model.language_model.layers.20.mlp.experts.26.gate_proj", "model.language_model.layers.20.mlp.experts.27.gate_proj", "model.language_model.layers.20.mlp.experts.28.gate_proj", "model.language_model.layers.20.mlp.experts.29.gate_proj", "model.language_model.layers.20.mlp.experts.30.gate_proj", "model.language_model.layers.20.mlp.experts.31.gate_proj", "model.language_model.layers.20.mlp.experts.32.gate_proj", "model.language_model.layers.20.mlp.experts.33.gate_proj", "model.language_model.layers.20.mlp.experts.34.gate_proj", "model.language_model.layers.20.mlp.experts.35.gate_proj", "model.language_model.layers.20.mlp.experts.36.gate_proj", "model.language_model.layers.20.mlp.experts.37.gate_proj", "model.language_model.layers.20.mlp.experts.38.gate_proj", "model.language_model.layers.20.mlp.experts.39.gate_proj", "model.language_model.layers.20.mlp.experts.40.gate_proj", "model.language_model.layers.20.mlp.experts.41.gate_proj", "model.language_model.layers.20.mlp.experts.42.gate_proj", "model.language_model.layers.20.mlp.experts.43.gate_proj", "model.language_model.layers.20.mlp.experts.44.gate_proj", "model.language_model.layers.20.mlp.experts.45.gate_proj", "model.language_model.layers.20.mlp.experts.46.gate_proj", "model.language_model.layers.20.mlp.experts.47.gate_proj", "model.language_model.layers.20.mlp.experts.48.gate_proj", "model.language_model.layers.20.mlp.experts.49.gate_proj", "model.language_model.layers.20.mlp.experts.50.gate_proj", "model.language_model.layers.20.mlp.experts.51.gate_proj", "model.language_model.layers.20.mlp.experts.52.gate_proj", "model.language_model.layers.20.mlp.experts.53.gate_proj", "model.language_model.layers.20.mlp.experts.54.gate_proj", "model.language_model.layers.20.mlp.experts.55.gate_proj", "model.language_model.layers.20.mlp.experts.56.gate_proj", "model.language_model.layers.20.mlp.experts.57.gate_proj", "model.language_model.layers.20.mlp.experts.58.gate_proj", "model.language_model.layers.20.mlp.experts.59.gate_proj", "model.language_model.layers.20.mlp.experts.60.gate_proj", "model.language_model.layers.20.mlp.experts.61.gate_proj", "model.language_model.layers.20.mlp.experts.62.gate_proj", "model.language_model.layers.20.mlp.experts.63.gate_proj", "model.language_model.layers.20.mlp.experts.64.gate_proj", "model.language_model.layers.20.mlp.experts.65.gate_proj", "model.language_model.layers.20.mlp.experts.66.gate_proj", "model.language_model.layers.20.mlp.experts.67.gate_proj", "model.language_model.layers.20.mlp.experts.68.gate_proj", "model.language_model.layers.20.mlp.experts.69.gate_proj", "model.language_model.layers.20.mlp.experts.70.gate_proj", "model.language_model.layers.20.mlp.experts.71.gate_proj", "model.language_model.layers.20.mlp.experts.72.gate_proj", "model.language_model.layers.20.mlp.experts.73.gate_proj", "model.language_model.layers.20.mlp.experts.74.gate_proj", "model.language_model.layers.20.mlp.experts.75.gate_proj", "model.language_model.layers.20.mlp.experts.76.gate_proj", "model.language_model.layers.20.mlp.experts.77.gate_proj", "model.language_model.layers.20.mlp.experts.78.gate_proj", "model.language_model.layers.20.mlp.experts.79.gate_proj", "model.language_model.layers.20.mlp.experts.80.gate_proj", "model.language_model.layers.20.mlp.experts.81.gate_proj", "model.language_model.layers.20.mlp.experts.82.gate_proj", "model.language_model.layers.20.mlp.experts.83.gate_proj", "model.language_model.layers.20.mlp.experts.84.gate_proj", "model.language_model.layers.20.mlp.experts.85.gate_proj", "model.language_model.layers.20.mlp.experts.86.gate_proj", "model.language_model.layers.20.mlp.experts.87.gate_proj", "model.language_model.layers.20.mlp.experts.88.gate_proj", "model.language_model.layers.20.mlp.experts.89.gate_proj", "model.language_model.layers.20.mlp.experts.90.gate_proj", "model.language_model.layers.20.mlp.experts.91.gate_proj", "model.language_model.layers.20.mlp.experts.92.gate_proj", "model.language_model.layers.20.mlp.experts.93.gate_proj", "model.language_model.layers.20.mlp.experts.94.gate_proj", "model.language_model.layers.20.mlp.experts.95.gate_proj", "model.language_model.layers.20.mlp.experts.96.gate_proj", "model.language_model.layers.20.mlp.experts.97.gate_proj", "model.language_model.layers.20.mlp.experts.98.gate_proj", "model.language_model.layers.20.mlp.experts.99.gate_proj", "model.language_model.layers.20.mlp.experts.100.gate_proj", "model.language_model.layers.20.mlp.experts.101.gate_proj", "model.language_model.layers.20.mlp.experts.102.gate_proj", "model.language_model.layers.20.mlp.experts.103.gate_proj", "model.language_model.layers.20.mlp.experts.104.gate_proj", "model.language_model.layers.20.mlp.experts.105.gate_proj", "model.language_model.layers.20.mlp.experts.106.gate_proj", "model.language_model.layers.20.mlp.experts.107.gate_proj", "model.language_model.layers.20.mlp.experts.108.gate_proj", "model.language_model.layers.20.mlp.experts.109.gate_proj", "model.language_model.layers.20.mlp.experts.110.gate_proj", "model.language_model.layers.20.mlp.experts.111.gate_proj", "model.language_model.layers.20.mlp.experts.112.gate_proj", "model.language_model.layers.20.mlp.experts.113.gate_proj", "model.language_model.layers.20.mlp.experts.114.gate_proj", "model.language_model.layers.20.mlp.experts.115.gate_proj", "model.language_model.layers.20.mlp.experts.116.gate_proj", "model.language_model.layers.20.mlp.experts.117.gate_proj", "model.language_model.layers.20.mlp.experts.118.gate_proj", "model.language_model.layers.20.mlp.experts.119.gate_proj", "model.language_model.layers.20.mlp.experts.120.gate_proj", "model.language_model.layers.20.mlp.experts.121.gate_proj", "model.language_model.layers.20.mlp.experts.122.gate_proj", "model.language_model.layers.20.mlp.experts.123.gate_proj", "model.language_model.layers.20.mlp.experts.124.gate_proj", "model.language_model.layers.20.mlp.experts.125.gate_proj", "model.language_model.layers.20.mlp.experts.126.gate_proj", "model.language_model.layers.20.mlp.experts.127.gate_proj", "model.language_model.layers.20.mlp.experts.0.up_proj", "model.language_model.layers.20.mlp.experts.1.up_proj", "model.language_model.layers.20.mlp.experts.2.up_proj", "model.language_model.layers.20.mlp.experts.3.up_proj", "model.language_model.layers.20.mlp.experts.4.up_proj", "model.language_model.layers.20.mlp.experts.5.up_proj", "model.language_model.layers.20.mlp.experts.6.up_proj", "model.language_model.layers.20.mlp.experts.7.up_proj", "model.language_model.layers.20.mlp.experts.8.up_proj", "model.language_model.layers.20.mlp.experts.9.up_proj", "model.language_model.layers.20.mlp.experts.10.up_proj", "model.language_model.layers.20.mlp.experts.11.up_proj", "model.language_model.layers.20.mlp.experts.12.up_proj", "model.language_model.layers.20.mlp.experts.13.up_proj", "model.language_model.layers.20.mlp.experts.14.up_proj", "model.language_model.layers.20.mlp.experts.15.up_proj", "model.language_model.layers.20.mlp.experts.16.up_proj", "model.language_model.layers.20.mlp.experts.17.up_proj", "model.language_model.layers.20.mlp.experts.18.up_proj", "model.language_model.layers.20.mlp.experts.19.up_proj", "model.language_model.layers.20.mlp.experts.20.up_proj", "model.language_model.layers.20.mlp.experts.21.up_proj", "model.language_model.layers.20.mlp.experts.22.up_proj", "model.language_model.layers.20.mlp.experts.23.up_proj", "model.language_model.layers.20.mlp.experts.24.up_proj", "model.language_model.layers.20.mlp.experts.25.up_proj", "model.language_model.layers.20.mlp.experts.26.up_proj", "model.language_model.layers.20.mlp.experts.27.up_proj", "model.language_model.layers.20.mlp.experts.28.up_proj", "model.language_model.layers.20.mlp.experts.29.up_proj", "model.language_model.layers.20.mlp.experts.30.up_proj", "model.language_model.layers.20.mlp.experts.31.up_proj", "model.language_model.layers.20.mlp.experts.32.up_proj", "model.language_model.layers.20.mlp.experts.33.up_proj", "model.language_model.layers.20.mlp.experts.34.up_proj", "model.language_model.layers.20.mlp.experts.35.up_proj", "model.language_model.layers.20.mlp.experts.36.up_proj", "model.language_model.layers.20.mlp.experts.37.up_proj", "model.language_model.layers.20.mlp.experts.38.up_proj", "model.language_model.layers.20.mlp.experts.39.up_proj", "model.language_model.layers.20.mlp.experts.40.up_proj", "model.language_model.layers.20.mlp.experts.41.up_proj", "model.language_model.layers.20.mlp.experts.42.up_proj", "model.language_model.layers.20.mlp.experts.43.up_proj", "model.language_model.layers.20.mlp.experts.44.up_proj", "model.language_model.layers.20.mlp.experts.45.up_proj", "model.language_model.layers.20.mlp.experts.46.up_proj", "model.language_model.layers.20.mlp.experts.47.up_proj", "model.language_model.layers.20.mlp.experts.48.up_proj", "model.language_model.layers.20.mlp.experts.49.up_proj", "model.language_model.layers.20.mlp.experts.50.up_proj", "model.language_model.layers.20.mlp.experts.51.up_proj", "model.language_model.layers.20.mlp.experts.52.up_proj", "model.language_model.layers.20.mlp.experts.53.up_proj", "model.language_model.layers.20.mlp.experts.54.up_proj", "model.language_model.layers.20.mlp.experts.55.up_proj", "model.language_model.layers.20.mlp.experts.56.up_proj", "model.language_model.layers.20.mlp.experts.57.up_proj", "model.language_model.layers.20.mlp.experts.58.up_proj", "model.language_model.layers.20.mlp.experts.59.up_proj", "model.language_model.layers.20.mlp.experts.60.up_proj", "model.language_model.layers.20.mlp.experts.61.up_proj", "model.language_model.layers.20.mlp.experts.62.up_proj", "model.language_model.layers.20.mlp.experts.63.up_proj", "model.language_model.layers.20.mlp.experts.64.up_proj", "model.language_model.layers.20.mlp.experts.65.up_proj", "model.language_model.layers.20.mlp.experts.66.up_proj", "model.language_model.layers.20.mlp.experts.67.up_proj", "model.language_model.layers.20.mlp.experts.68.up_proj", "model.language_model.layers.20.mlp.experts.69.up_proj", "model.language_model.layers.20.mlp.experts.70.up_proj", "model.language_model.layers.20.mlp.experts.71.up_proj", "model.language_model.layers.20.mlp.experts.72.up_proj", "model.language_model.layers.20.mlp.experts.73.up_proj", "model.language_model.layers.20.mlp.experts.74.up_proj", "model.language_model.layers.20.mlp.experts.75.up_proj", "model.language_model.layers.20.mlp.experts.76.up_proj", "model.language_model.layers.20.mlp.experts.77.up_proj", "model.language_model.layers.20.mlp.experts.78.up_proj", "model.language_model.layers.20.mlp.experts.79.up_proj", "model.language_model.layers.20.mlp.experts.80.up_proj", "model.language_model.layers.20.mlp.experts.81.up_proj", "model.language_model.layers.20.mlp.experts.82.up_proj", "model.language_model.layers.20.mlp.experts.83.up_proj", "model.language_model.layers.20.mlp.experts.84.up_proj", "model.language_model.layers.20.mlp.experts.85.up_proj", "model.language_model.layers.20.mlp.experts.86.up_proj", "model.language_model.layers.20.mlp.experts.87.up_proj", "model.language_model.layers.20.mlp.experts.88.up_proj", "model.language_model.layers.20.mlp.experts.89.up_proj", "model.language_model.layers.20.mlp.experts.90.up_proj", "model.language_model.layers.20.mlp.experts.91.up_proj", "model.language_model.layers.20.mlp.experts.92.up_proj", "model.language_model.layers.20.mlp.experts.93.up_proj", "model.language_model.layers.20.mlp.experts.94.up_proj", "model.language_model.layers.20.mlp.experts.95.up_proj", "model.language_model.layers.20.mlp.experts.96.up_proj", "model.language_model.layers.20.mlp.experts.97.up_proj", "model.language_model.layers.20.mlp.experts.98.up_proj", "model.language_model.layers.20.mlp.experts.99.up_proj", "model.language_model.layers.20.mlp.experts.100.up_proj", "model.language_model.layers.20.mlp.experts.101.up_proj", "model.language_model.layers.20.mlp.experts.102.up_proj", "model.language_model.layers.20.mlp.experts.103.up_proj", "model.language_model.layers.20.mlp.experts.104.up_proj", "model.language_model.layers.20.mlp.experts.105.up_proj", "model.language_model.layers.20.mlp.experts.106.up_proj", "model.language_model.layers.20.mlp.experts.107.up_proj", "model.language_model.layers.20.mlp.experts.108.up_proj", "model.language_model.layers.20.mlp.experts.109.up_proj", "model.language_model.layers.20.mlp.experts.110.up_proj", "model.language_model.layers.20.mlp.experts.111.up_proj", "model.language_model.layers.20.mlp.experts.112.up_proj", "model.language_model.layers.20.mlp.experts.113.up_proj", "model.language_model.layers.20.mlp.experts.114.up_proj", "model.language_model.layers.20.mlp.experts.115.up_proj", "model.language_model.layers.20.mlp.experts.116.up_proj", "model.language_model.layers.20.mlp.experts.117.up_proj", "model.language_model.layers.20.mlp.experts.118.up_proj", "model.language_model.layers.20.mlp.experts.119.up_proj", "model.language_model.layers.20.mlp.experts.120.up_proj", "model.language_model.layers.20.mlp.experts.121.up_proj", "model.language_model.layers.20.mlp.experts.122.up_proj", "model.language_model.layers.20.mlp.experts.123.up_proj", "model.language_model.layers.20.mlp.experts.124.up_proj", "model.language_model.layers.20.mlp.experts.125.up_proj", "model.language_model.layers.20.mlp.experts.126.up_proj", "model.language_model.layers.20.mlp.experts.127.up_proj", "model.language_model.layers.20.mlp.experts.0.down_proj", "model.language_model.layers.20.mlp.experts.1.down_proj", "model.language_model.layers.20.mlp.experts.2.down_proj", "model.language_model.layers.20.mlp.experts.3.down_proj", "model.language_model.layers.20.mlp.experts.4.down_proj", "model.language_model.layers.20.mlp.experts.5.down_proj", "model.language_model.layers.20.mlp.experts.6.down_proj", "model.language_model.layers.20.mlp.experts.7.down_proj", "model.language_model.layers.20.mlp.experts.8.down_proj", "model.language_model.layers.20.mlp.experts.9.down_proj", "model.language_model.layers.20.mlp.experts.10.down_proj", "model.language_model.layers.20.mlp.experts.11.down_proj", "model.language_model.layers.20.mlp.experts.12.down_proj", "model.language_model.layers.20.mlp.experts.13.down_proj", "model.language_model.layers.20.mlp.experts.14.down_proj", "model.language_model.layers.20.mlp.experts.15.down_proj", "model.language_model.layers.20.mlp.experts.16.down_proj", "model.language_model.layers.20.mlp.experts.17.down_proj", "model.language_model.layers.20.mlp.experts.18.down_proj", "model.language_model.layers.20.mlp.experts.19.down_proj", "model.language_model.layers.20.mlp.experts.20.down_proj", "model.language_model.layers.20.mlp.experts.21.down_proj", "model.language_model.layers.20.mlp.experts.22.down_proj", "model.language_model.layers.20.mlp.experts.23.down_proj", "model.language_model.layers.20.mlp.experts.24.down_proj", "model.language_model.layers.20.mlp.experts.25.down_proj", "model.language_model.layers.20.mlp.experts.26.down_proj", "model.language_model.layers.20.mlp.experts.27.down_proj", "model.language_model.layers.20.mlp.experts.28.down_proj", "model.language_model.layers.20.mlp.experts.29.down_proj", "model.language_model.layers.20.mlp.experts.30.down_proj", "model.language_model.layers.20.mlp.experts.31.down_proj", "model.language_model.layers.20.mlp.experts.32.down_proj", "model.language_model.layers.20.mlp.experts.33.down_proj", "model.language_model.layers.20.mlp.experts.34.down_proj", "model.language_model.layers.20.mlp.experts.35.down_proj", "model.language_model.layers.20.mlp.experts.36.down_proj", "model.language_model.layers.20.mlp.experts.37.down_proj", "model.language_model.layers.20.mlp.experts.38.down_proj", "model.language_model.layers.20.mlp.experts.39.down_proj", "model.language_model.layers.20.mlp.experts.40.down_proj", "model.language_model.layers.20.mlp.experts.41.down_proj", "model.language_model.layers.20.mlp.experts.42.down_proj", "model.language_model.layers.20.mlp.experts.43.down_proj", "model.language_model.layers.20.mlp.experts.44.down_proj", "model.language_model.layers.20.mlp.experts.45.down_proj", "model.language_model.layers.20.mlp.experts.46.down_proj", "model.language_model.layers.20.mlp.experts.47.down_proj", "model.language_model.layers.20.mlp.experts.48.down_proj", "model.language_model.layers.20.mlp.experts.49.down_proj", "model.language_model.layers.20.mlp.experts.50.down_proj", "model.language_model.layers.20.mlp.experts.51.down_proj", "model.language_model.layers.20.mlp.experts.52.down_proj", "model.language_model.layers.20.mlp.experts.53.down_proj", "model.language_model.layers.20.mlp.experts.54.down_proj", "model.language_model.layers.20.mlp.experts.55.down_proj", "model.language_model.layers.20.mlp.experts.56.down_proj", "model.language_model.layers.20.mlp.experts.57.down_proj", "model.language_model.layers.20.mlp.experts.58.down_proj", "model.language_model.layers.20.mlp.experts.59.down_proj", "model.language_model.layers.20.mlp.experts.60.down_proj", "model.language_model.layers.20.mlp.experts.61.down_proj", "model.language_model.layers.20.mlp.experts.62.down_proj", "model.language_model.layers.20.mlp.experts.63.down_proj", "model.language_model.layers.20.mlp.experts.64.down_proj", "model.language_model.layers.20.mlp.experts.65.down_proj", "model.language_model.layers.20.mlp.experts.66.down_proj", "model.language_model.layers.20.mlp.experts.67.down_proj", "model.language_model.layers.20.mlp.experts.68.down_proj", "model.language_model.layers.20.mlp.experts.69.down_proj", "model.language_model.layers.20.mlp.experts.70.down_proj", "model.language_model.layers.20.mlp.experts.71.down_proj", "model.language_model.layers.20.mlp.experts.72.down_proj", "model.language_model.layers.20.mlp.experts.73.down_proj", "model.language_model.layers.20.mlp.experts.74.down_proj", "model.language_model.layers.20.mlp.experts.75.down_proj", "model.language_model.layers.20.mlp.experts.76.down_proj", "model.language_model.layers.20.mlp.experts.77.down_proj", "model.language_model.layers.20.mlp.experts.78.down_proj", "model.language_model.layers.20.mlp.experts.79.down_proj", "model.language_model.layers.20.mlp.experts.80.down_proj", "model.language_model.layers.20.mlp.experts.81.down_proj", "model.language_model.layers.20.mlp.experts.82.down_proj", "model.language_model.layers.20.mlp.experts.83.down_proj", "model.language_model.layers.20.mlp.experts.84.down_proj", "model.language_model.layers.20.mlp.experts.85.down_proj", "model.language_model.layers.20.mlp.experts.86.down_proj", "model.language_model.layers.20.mlp.experts.87.down_proj", "model.language_model.layers.20.mlp.experts.88.down_proj", "model.language_model.layers.20.mlp.experts.89.down_proj", "model.language_model.layers.20.mlp.experts.90.down_proj", "model.language_model.layers.20.mlp.experts.91.down_proj", "model.language_model.layers.20.mlp.experts.92.down_proj", "model.language_model.layers.20.mlp.experts.93.down_proj", "model.language_model.layers.20.mlp.experts.94.down_proj", "model.language_model.layers.20.mlp.experts.95.down_proj", "model.language_model.layers.20.mlp.experts.96.down_proj", "model.language_model.layers.20.mlp.experts.97.down_proj", "model.language_model.layers.20.mlp.experts.98.down_proj", "model.language_model.layers.20.mlp.experts.99.down_proj", "model.language_model.layers.20.mlp.experts.100.down_proj", "model.language_model.layers.20.mlp.experts.101.down_proj", "model.language_model.layers.20.mlp.experts.102.down_proj", "model.language_model.layers.20.mlp.experts.103.down_proj", "model.language_model.layers.20.mlp.experts.104.down_proj", "model.language_model.layers.20.mlp.experts.105.down_proj", "model.language_model.layers.20.mlp.experts.106.down_proj", "model.language_model.layers.20.mlp.experts.107.down_proj", "model.language_model.layers.20.mlp.experts.108.down_proj", "model.language_model.layers.20.mlp.experts.109.down_proj", "model.language_model.layers.20.mlp.experts.110.down_proj", "model.language_model.layers.20.mlp.experts.111.down_proj", "model.language_model.layers.20.mlp.experts.112.down_proj", "model.language_model.layers.20.mlp.experts.113.down_proj", "model.language_model.layers.20.mlp.experts.114.down_proj", "model.language_model.layers.20.mlp.experts.115.down_proj", "model.language_model.layers.20.mlp.experts.116.down_proj", "model.language_model.layers.20.mlp.experts.117.down_proj", "model.language_model.layers.20.mlp.experts.118.down_proj", "model.language_model.layers.20.mlp.experts.119.down_proj", "model.language_model.layers.20.mlp.experts.120.down_proj", "model.language_model.layers.20.mlp.experts.121.down_proj", "model.language_model.layers.20.mlp.experts.122.down_proj", "model.language_model.layers.20.mlp.experts.123.down_proj", "model.language_model.layers.20.mlp.experts.124.down_proj", "model.language_model.layers.20.mlp.experts.125.down_proj", "model.language_model.layers.20.mlp.experts.126.down_proj", "model.language_model.layers.20.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": 0.0005951255559921292, "dbits": 603979776 } ] }, { "idx": 42, "layers": [ "model.language_model.layers.21.self_attn.q_proj", "model.language_model.layers.21.self_attn.k_proj", "model.language_model.layers.21.self_attn.v_proj", "model.language_model.layers.21.self_attn.o_proj" ], "candidates": [ { "dkld": 0.0008615614846348707, "dbits": 18874368 } ] }, { "idx": 43, "layers": [ "model.language_model.layers.21.mlp.experts.0.gate_proj", "model.language_model.layers.21.mlp.experts.1.gate_proj", "model.language_model.layers.21.mlp.experts.2.gate_proj", "model.language_model.layers.21.mlp.experts.3.gate_proj", "model.language_model.layers.21.mlp.experts.4.gate_proj", "model.language_model.layers.21.mlp.experts.5.gate_proj", "model.language_model.layers.21.mlp.experts.6.gate_proj", "model.language_model.layers.21.mlp.experts.7.gate_proj", "model.language_model.layers.21.mlp.experts.8.gate_proj", "model.language_model.layers.21.mlp.experts.9.gate_proj", "model.language_model.layers.21.mlp.experts.10.gate_proj", "model.language_model.layers.21.mlp.experts.11.gate_proj", "model.language_model.layers.21.mlp.experts.12.gate_proj", "model.language_model.layers.21.mlp.experts.13.gate_proj", "model.language_model.layers.21.mlp.experts.14.gate_proj", "model.language_model.layers.21.mlp.experts.15.gate_proj", "model.language_model.layers.21.mlp.experts.16.gate_proj", "model.language_model.layers.21.mlp.experts.17.gate_proj", "model.language_model.layers.21.mlp.experts.18.gate_proj", "model.language_model.layers.21.mlp.experts.19.gate_proj", "model.language_model.layers.21.mlp.experts.20.gate_proj", "model.language_model.layers.21.mlp.experts.21.gate_proj", "model.language_model.layers.21.mlp.experts.22.gate_proj", "model.language_model.layers.21.mlp.experts.23.gate_proj", "model.language_model.layers.21.mlp.experts.24.gate_proj", "model.language_model.layers.21.mlp.experts.25.gate_proj", "model.language_model.layers.21.mlp.experts.26.gate_proj", "model.language_model.layers.21.mlp.experts.27.gate_proj", "model.language_model.layers.21.mlp.experts.28.gate_proj", "model.language_model.layers.21.mlp.experts.29.gate_proj", "model.language_model.layers.21.mlp.experts.30.gate_proj", "model.language_model.layers.21.mlp.experts.31.gate_proj", "model.language_model.layers.21.mlp.experts.32.gate_proj", "model.language_model.layers.21.mlp.experts.33.gate_proj", "model.language_model.layers.21.mlp.experts.34.gate_proj", "model.language_model.layers.21.mlp.experts.35.gate_proj", "model.language_model.layers.21.mlp.experts.36.gate_proj", "model.language_model.layers.21.mlp.experts.37.gate_proj", "model.language_model.layers.21.mlp.experts.38.gate_proj", "model.language_model.layers.21.mlp.experts.39.gate_proj", "model.language_model.layers.21.mlp.experts.40.gate_proj", "model.language_model.layers.21.mlp.experts.41.gate_proj", "model.language_model.layers.21.mlp.experts.42.gate_proj", "model.language_model.layers.21.mlp.experts.43.gate_proj", "model.language_model.layers.21.mlp.experts.44.gate_proj", "model.language_model.layers.21.mlp.experts.45.gate_proj", "model.language_model.layers.21.mlp.experts.46.gate_proj", "model.language_model.layers.21.mlp.experts.47.gate_proj", "model.language_model.layers.21.mlp.experts.48.gate_proj", "model.language_model.layers.21.mlp.experts.49.gate_proj", "model.language_model.layers.21.mlp.experts.50.gate_proj", "model.language_model.layers.21.mlp.experts.51.gate_proj", "model.language_model.layers.21.mlp.experts.52.gate_proj", "model.language_model.layers.21.mlp.experts.53.gate_proj", "model.language_model.layers.21.mlp.experts.54.gate_proj", "model.language_model.layers.21.mlp.experts.55.gate_proj", "model.language_model.layers.21.mlp.experts.56.gate_proj", "model.language_model.layers.21.mlp.experts.57.gate_proj", "model.language_model.layers.21.mlp.experts.58.gate_proj", "model.language_model.layers.21.mlp.experts.59.gate_proj", "model.language_model.layers.21.mlp.experts.60.gate_proj", "model.language_model.layers.21.mlp.experts.61.gate_proj", "model.language_model.layers.21.mlp.experts.62.gate_proj", "model.language_model.layers.21.mlp.experts.63.gate_proj", "model.language_model.layers.21.mlp.experts.64.gate_proj", "model.language_model.layers.21.mlp.experts.65.gate_proj", "model.language_model.layers.21.mlp.experts.66.gate_proj", "model.language_model.layers.21.mlp.experts.67.gate_proj", "model.language_model.layers.21.mlp.experts.68.gate_proj", "model.language_model.layers.21.mlp.experts.69.gate_proj", "model.language_model.layers.21.mlp.experts.70.gate_proj", "model.language_model.layers.21.mlp.experts.71.gate_proj", "model.language_model.layers.21.mlp.experts.72.gate_proj", "model.language_model.layers.21.mlp.experts.73.gate_proj", "model.language_model.layers.21.mlp.experts.74.gate_proj", "model.language_model.layers.21.mlp.experts.75.gate_proj", "model.language_model.layers.21.mlp.experts.76.gate_proj", "model.language_model.layers.21.mlp.experts.77.gate_proj", "model.language_model.layers.21.mlp.experts.78.gate_proj", "model.language_model.layers.21.mlp.experts.79.gate_proj", "model.language_model.layers.21.mlp.experts.80.gate_proj", "model.language_model.layers.21.mlp.experts.81.gate_proj", "model.language_model.layers.21.mlp.experts.82.gate_proj", "model.language_model.layers.21.mlp.experts.83.gate_proj", "model.language_model.layers.21.mlp.experts.84.gate_proj", "model.language_model.layers.21.mlp.experts.85.gate_proj", "model.language_model.layers.21.mlp.experts.86.gate_proj", "model.language_model.layers.21.mlp.experts.87.gate_proj", "model.language_model.layers.21.mlp.experts.88.gate_proj", "model.language_model.layers.21.mlp.experts.89.gate_proj", "model.language_model.layers.21.mlp.experts.90.gate_proj", "model.language_model.layers.21.mlp.experts.91.gate_proj", "model.language_model.layers.21.mlp.experts.92.gate_proj", "model.language_model.layers.21.mlp.experts.93.gate_proj", "model.language_model.layers.21.mlp.experts.94.gate_proj", "model.language_model.layers.21.mlp.experts.95.gate_proj", "model.language_model.layers.21.mlp.experts.96.gate_proj", "model.language_model.layers.21.mlp.experts.97.gate_proj", "model.language_model.layers.21.mlp.experts.98.gate_proj", "model.language_model.layers.21.mlp.experts.99.gate_proj", "model.language_model.layers.21.mlp.experts.100.gate_proj", "model.language_model.layers.21.mlp.experts.101.gate_proj", "model.language_model.layers.21.mlp.experts.102.gate_proj", "model.language_model.layers.21.mlp.experts.103.gate_proj", "model.language_model.layers.21.mlp.experts.104.gate_proj", "model.language_model.layers.21.mlp.experts.105.gate_proj", "model.language_model.layers.21.mlp.experts.106.gate_proj", "model.language_model.layers.21.mlp.experts.107.gate_proj", "model.language_model.layers.21.mlp.experts.108.gate_proj", "model.language_model.layers.21.mlp.experts.109.gate_proj", "model.language_model.layers.21.mlp.experts.110.gate_proj", "model.language_model.layers.21.mlp.experts.111.gate_proj", "model.language_model.layers.21.mlp.experts.112.gate_proj", "model.language_model.layers.21.mlp.experts.113.gate_proj", "model.language_model.layers.21.mlp.experts.114.gate_proj", "model.language_model.layers.21.mlp.experts.115.gate_proj", "model.language_model.layers.21.mlp.experts.116.gate_proj", "model.language_model.layers.21.mlp.experts.117.gate_proj", "model.language_model.layers.21.mlp.experts.118.gate_proj", "model.language_model.layers.21.mlp.experts.119.gate_proj", "model.language_model.layers.21.mlp.experts.120.gate_proj", "model.language_model.layers.21.mlp.experts.121.gate_proj", "model.language_model.layers.21.mlp.experts.122.gate_proj", "model.language_model.layers.21.mlp.experts.123.gate_proj", "model.language_model.layers.21.mlp.experts.124.gate_proj", "model.language_model.layers.21.mlp.experts.125.gate_proj", "model.language_model.layers.21.mlp.experts.126.gate_proj", "model.language_model.layers.21.mlp.experts.127.gate_proj", "model.language_model.layers.21.mlp.experts.0.up_proj", "model.language_model.layers.21.mlp.experts.1.up_proj", "model.language_model.layers.21.mlp.experts.2.up_proj", "model.language_model.layers.21.mlp.experts.3.up_proj", "model.language_model.layers.21.mlp.experts.4.up_proj", "model.language_model.layers.21.mlp.experts.5.up_proj", "model.language_model.layers.21.mlp.experts.6.up_proj", "model.language_model.layers.21.mlp.experts.7.up_proj", "model.language_model.layers.21.mlp.experts.8.up_proj", "model.language_model.layers.21.mlp.experts.9.up_proj", "model.language_model.layers.21.mlp.experts.10.up_proj", "model.language_model.layers.21.mlp.experts.11.up_proj", "model.language_model.layers.21.mlp.experts.12.up_proj", "model.language_model.layers.21.mlp.experts.13.up_proj", "model.language_model.layers.21.mlp.experts.14.up_proj", "model.language_model.layers.21.mlp.experts.15.up_proj", "model.language_model.layers.21.mlp.experts.16.up_proj", "model.language_model.layers.21.mlp.experts.17.up_proj", "model.language_model.layers.21.mlp.experts.18.up_proj", "model.language_model.layers.21.mlp.experts.19.up_proj", "model.language_model.layers.21.mlp.experts.20.up_proj", "model.language_model.layers.21.mlp.experts.21.up_proj", "model.language_model.layers.21.mlp.experts.22.up_proj", "model.language_model.layers.21.mlp.experts.23.up_proj", "model.language_model.layers.21.mlp.experts.24.up_proj", "model.language_model.layers.21.mlp.experts.25.up_proj", "model.language_model.layers.21.mlp.experts.26.up_proj", "model.language_model.layers.21.mlp.experts.27.up_proj", "model.language_model.layers.21.mlp.experts.28.up_proj", "model.language_model.layers.21.mlp.experts.29.up_proj", "model.language_model.layers.21.mlp.experts.30.up_proj", "model.language_model.layers.21.mlp.experts.31.up_proj", "model.language_model.layers.21.mlp.experts.32.up_proj", "model.language_model.layers.21.mlp.experts.33.up_proj", "model.language_model.layers.21.mlp.experts.34.up_proj", "model.language_model.layers.21.mlp.experts.35.up_proj", "model.language_model.layers.21.mlp.experts.36.up_proj", "model.language_model.layers.21.mlp.experts.37.up_proj", "model.language_model.layers.21.mlp.experts.38.up_proj", "model.language_model.layers.21.mlp.experts.39.up_proj", "model.language_model.layers.21.mlp.experts.40.up_proj", "model.language_model.layers.21.mlp.experts.41.up_proj", "model.language_model.layers.21.mlp.experts.42.up_proj", "model.language_model.layers.21.mlp.experts.43.up_proj", "model.language_model.layers.21.mlp.experts.44.up_proj", "model.language_model.layers.21.mlp.experts.45.up_proj", "model.language_model.layers.21.mlp.experts.46.up_proj", "model.language_model.layers.21.mlp.experts.47.up_proj", "model.language_model.layers.21.mlp.experts.48.up_proj", "model.language_model.layers.21.mlp.experts.49.up_proj", "model.language_model.layers.21.mlp.experts.50.up_proj", "model.language_model.layers.21.mlp.experts.51.up_proj", "model.language_model.layers.21.mlp.experts.52.up_proj", "model.language_model.layers.21.mlp.experts.53.up_proj", "model.language_model.layers.21.mlp.experts.54.up_proj", "model.language_model.layers.21.mlp.experts.55.up_proj", "model.language_model.layers.21.mlp.experts.56.up_proj", "model.language_model.layers.21.mlp.experts.57.up_proj", "model.language_model.layers.21.mlp.experts.58.up_proj", "model.language_model.layers.21.mlp.experts.59.up_proj", "model.language_model.layers.21.mlp.experts.60.up_proj", "model.language_model.layers.21.mlp.experts.61.up_proj", "model.language_model.layers.21.mlp.experts.62.up_proj", "model.language_model.layers.21.mlp.experts.63.up_proj", "model.language_model.layers.21.mlp.experts.64.up_proj", "model.language_model.layers.21.mlp.experts.65.up_proj", "model.language_model.layers.21.mlp.experts.66.up_proj", "model.language_model.layers.21.mlp.experts.67.up_proj", "model.language_model.layers.21.mlp.experts.68.up_proj", "model.language_model.layers.21.mlp.experts.69.up_proj", "model.language_model.layers.21.mlp.experts.70.up_proj", "model.language_model.layers.21.mlp.experts.71.up_proj", "model.language_model.layers.21.mlp.experts.72.up_proj", "model.language_model.layers.21.mlp.experts.73.up_proj", "model.language_model.layers.21.mlp.experts.74.up_proj", "model.language_model.layers.21.mlp.experts.75.up_proj", "model.language_model.layers.21.mlp.experts.76.up_proj", "model.language_model.layers.21.mlp.experts.77.up_proj", "model.language_model.layers.21.mlp.experts.78.up_proj", "model.language_model.layers.21.mlp.experts.79.up_proj", "model.language_model.layers.21.mlp.experts.80.up_proj", "model.language_model.layers.21.mlp.experts.81.up_proj", "model.language_model.layers.21.mlp.experts.82.up_proj", "model.language_model.layers.21.mlp.experts.83.up_proj", "model.language_model.layers.21.mlp.experts.84.up_proj", "model.language_model.layers.21.mlp.experts.85.up_proj", "model.language_model.layers.21.mlp.experts.86.up_proj", "model.language_model.layers.21.mlp.experts.87.up_proj", "model.language_model.layers.21.mlp.experts.88.up_proj", "model.language_model.layers.21.mlp.experts.89.up_proj", "model.language_model.layers.21.mlp.experts.90.up_proj", "model.language_model.layers.21.mlp.experts.91.up_proj", "model.language_model.layers.21.mlp.experts.92.up_proj", "model.language_model.layers.21.mlp.experts.93.up_proj", "model.language_model.layers.21.mlp.experts.94.up_proj", "model.language_model.layers.21.mlp.experts.95.up_proj", "model.language_model.layers.21.mlp.experts.96.up_proj", "model.language_model.layers.21.mlp.experts.97.up_proj", "model.language_model.layers.21.mlp.experts.98.up_proj", "model.language_model.layers.21.mlp.experts.99.up_proj", "model.language_model.layers.21.mlp.experts.100.up_proj", "model.language_model.layers.21.mlp.experts.101.up_proj", "model.language_model.layers.21.mlp.experts.102.up_proj", "model.language_model.layers.21.mlp.experts.103.up_proj", "model.language_model.layers.21.mlp.experts.104.up_proj", "model.language_model.layers.21.mlp.experts.105.up_proj", "model.language_model.layers.21.mlp.experts.106.up_proj", "model.language_model.layers.21.mlp.experts.107.up_proj", "model.language_model.layers.21.mlp.experts.108.up_proj", "model.language_model.layers.21.mlp.experts.109.up_proj", "model.language_model.layers.21.mlp.experts.110.up_proj", "model.language_model.layers.21.mlp.experts.111.up_proj", "model.language_model.layers.21.mlp.experts.112.up_proj", "model.language_model.layers.21.mlp.experts.113.up_proj", "model.language_model.layers.21.mlp.experts.114.up_proj", "model.language_model.layers.21.mlp.experts.115.up_proj", "model.language_model.layers.21.mlp.experts.116.up_proj", "model.language_model.layers.21.mlp.experts.117.up_proj", "model.language_model.layers.21.mlp.experts.118.up_proj", "model.language_model.layers.21.mlp.experts.119.up_proj", "model.language_model.layers.21.mlp.experts.120.up_proj", "model.language_model.layers.21.mlp.experts.121.up_proj", "model.language_model.layers.21.mlp.experts.122.up_proj", "model.language_model.layers.21.mlp.experts.123.up_proj", "model.language_model.layers.21.mlp.experts.124.up_proj", "model.language_model.layers.21.mlp.experts.125.up_proj", "model.language_model.layers.21.mlp.experts.126.up_proj", "model.language_model.layers.21.mlp.experts.127.up_proj", "model.language_model.layers.21.mlp.experts.0.down_proj", "model.language_model.layers.21.mlp.experts.1.down_proj", "model.language_model.layers.21.mlp.experts.2.down_proj", "model.language_model.layers.21.mlp.experts.3.down_proj", "model.language_model.layers.21.mlp.experts.4.down_proj", "model.language_model.layers.21.mlp.experts.5.down_proj", "model.language_model.layers.21.mlp.experts.6.down_proj", "model.language_model.layers.21.mlp.experts.7.down_proj", "model.language_model.layers.21.mlp.experts.8.down_proj", "model.language_model.layers.21.mlp.experts.9.down_proj", "model.language_model.layers.21.mlp.experts.10.down_proj", "model.language_model.layers.21.mlp.experts.11.down_proj", "model.language_model.layers.21.mlp.experts.12.down_proj", "model.language_model.layers.21.mlp.experts.13.down_proj", "model.language_model.layers.21.mlp.experts.14.down_proj", "model.language_model.layers.21.mlp.experts.15.down_proj", "model.language_model.layers.21.mlp.experts.16.down_proj", "model.language_model.layers.21.mlp.experts.17.down_proj", "model.language_model.layers.21.mlp.experts.18.down_proj", "model.language_model.layers.21.mlp.experts.19.down_proj", "model.language_model.layers.21.mlp.experts.20.down_proj", "model.language_model.layers.21.mlp.experts.21.down_proj", "model.language_model.layers.21.mlp.experts.22.down_proj", "model.language_model.layers.21.mlp.experts.23.down_proj", "model.language_model.layers.21.mlp.experts.24.down_proj", "model.language_model.layers.21.mlp.experts.25.down_proj", "model.language_model.layers.21.mlp.experts.26.down_proj", "model.language_model.layers.21.mlp.experts.27.down_proj", "model.language_model.layers.21.mlp.experts.28.down_proj", "model.language_model.layers.21.mlp.experts.29.down_proj", "model.language_model.layers.21.mlp.experts.30.down_proj", "model.language_model.layers.21.mlp.experts.31.down_proj", "model.language_model.layers.21.mlp.experts.32.down_proj", "model.language_model.layers.21.mlp.experts.33.down_proj", "model.language_model.layers.21.mlp.experts.34.down_proj", "model.language_model.layers.21.mlp.experts.35.down_proj", "model.language_model.layers.21.mlp.experts.36.down_proj", "model.language_model.layers.21.mlp.experts.37.down_proj", "model.language_model.layers.21.mlp.experts.38.down_proj", "model.language_model.layers.21.mlp.experts.39.down_proj", "model.language_model.layers.21.mlp.experts.40.down_proj", "model.language_model.layers.21.mlp.experts.41.down_proj", "model.language_model.layers.21.mlp.experts.42.down_proj", "model.language_model.layers.21.mlp.experts.43.down_proj", "model.language_model.layers.21.mlp.experts.44.down_proj", "model.language_model.layers.21.mlp.experts.45.down_proj", "model.language_model.layers.21.mlp.experts.46.down_proj", "model.language_model.layers.21.mlp.experts.47.down_proj", "model.language_model.layers.21.mlp.experts.48.down_proj", "model.language_model.layers.21.mlp.experts.49.down_proj", "model.language_model.layers.21.mlp.experts.50.down_proj", "model.language_model.layers.21.mlp.experts.51.down_proj", "model.language_model.layers.21.mlp.experts.52.down_proj", "model.language_model.layers.21.mlp.experts.53.down_proj", "model.language_model.layers.21.mlp.experts.54.down_proj", "model.language_model.layers.21.mlp.experts.55.down_proj", "model.language_model.layers.21.mlp.experts.56.down_proj", "model.language_model.layers.21.mlp.experts.57.down_proj", "model.language_model.layers.21.mlp.experts.58.down_proj", "model.language_model.layers.21.mlp.experts.59.down_proj", "model.language_model.layers.21.mlp.experts.60.down_proj", "model.language_model.layers.21.mlp.experts.61.down_proj", "model.language_model.layers.21.mlp.experts.62.down_proj", "model.language_model.layers.21.mlp.experts.63.down_proj", "model.language_model.layers.21.mlp.experts.64.down_proj", "model.language_model.layers.21.mlp.experts.65.down_proj", "model.language_model.layers.21.mlp.experts.66.down_proj", "model.language_model.layers.21.mlp.experts.67.down_proj", "model.language_model.layers.21.mlp.experts.68.down_proj", "model.language_model.layers.21.mlp.experts.69.down_proj", "model.language_model.layers.21.mlp.experts.70.down_proj", "model.language_model.layers.21.mlp.experts.71.down_proj", "model.language_model.layers.21.mlp.experts.72.down_proj", "model.language_model.layers.21.mlp.experts.73.down_proj", "model.language_model.layers.21.mlp.experts.74.down_proj", "model.language_model.layers.21.mlp.experts.75.down_proj", "model.language_model.layers.21.mlp.experts.76.down_proj", "model.language_model.layers.21.mlp.experts.77.down_proj", "model.language_model.layers.21.mlp.experts.78.down_proj", "model.language_model.layers.21.mlp.experts.79.down_proj", "model.language_model.layers.21.mlp.experts.80.down_proj", "model.language_model.layers.21.mlp.experts.81.down_proj", "model.language_model.layers.21.mlp.experts.82.down_proj", "model.language_model.layers.21.mlp.experts.83.down_proj", "model.language_model.layers.21.mlp.experts.84.down_proj", "model.language_model.layers.21.mlp.experts.85.down_proj", "model.language_model.layers.21.mlp.experts.86.down_proj", "model.language_model.layers.21.mlp.experts.87.down_proj", "model.language_model.layers.21.mlp.experts.88.down_proj", "model.language_model.layers.21.mlp.experts.89.down_proj", "model.language_model.layers.21.mlp.experts.90.down_proj", "model.language_model.layers.21.mlp.experts.91.down_proj", "model.language_model.layers.21.mlp.experts.92.down_proj", "model.language_model.layers.21.mlp.experts.93.down_proj", "model.language_model.layers.21.mlp.experts.94.down_proj", "model.language_model.layers.21.mlp.experts.95.down_proj", "model.language_model.layers.21.mlp.experts.96.down_proj", "model.language_model.layers.21.mlp.experts.97.down_proj", "model.language_model.layers.21.mlp.experts.98.down_proj", "model.language_model.layers.21.mlp.experts.99.down_proj", "model.language_model.layers.21.mlp.experts.100.down_proj", "model.language_model.layers.21.mlp.experts.101.down_proj", "model.language_model.layers.21.mlp.experts.102.down_proj", "model.language_model.layers.21.mlp.experts.103.down_proj", "model.language_model.layers.21.mlp.experts.104.down_proj", "model.language_model.layers.21.mlp.experts.105.down_proj", "model.language_model.layers.21.mlp.experts.106.down_proj", "model.language_model.layers.21.mlp.experts.107.down_proj", "model.language_model.layers.21.mlp.experts.108.down_proj", "model.language_model.layers.21.mlp.experts.109.down_proj", "model.language_model.layers.21.mlp.experts.110.down_proj", "model.language_model.layers.21.mlp.experts.111.down_proj", "model.language_model.layers.21.mlp.experts.112.down_proj", "model.language_model.layers.21.mlp.experts.113.down_proj", "model.language_model.layers.21.mlp.experts.114.down_proj", "model.language_model.layers.21.mlp.experts.115.down_proj", "model.language_model.layers.21.mlp.experts.116.down_proj", "model.language_model.layers.21.mlp.experts.117.down_proj", "model.language_model.layers.21.mlp.experts.118.down_proj", "model.language_model.layers.21.mlp.experts.119.down_proj", "model.language_model.layers.21.mlp.experts.120.down_proj", "model.language_model.layers.21.mlp.experts.121.down_proj", "model.language_model.layers.21.mlp.experts.122.down_proj", "model.language_model.layers.21.mlp.experts.123.down_proj", "model.language_model.layers.21.mlp.experts.124.down_proj", "model.language_model.layers.21.mlp.experts.125.down_proj", "model.language_model.layers.21.mlp.experts.126.down_proj", "model.language_model.layers.21.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.0020672964863479165, "dbits": 603979776 } ] }, { "idx": 44, "layers": [ "model.language_model.layers.22.self_attn.q_proj", "model.language_model.layers.22.self_attn.k_proj", "model.language_model.layers.22.self_attn.v_proj", "model.language_model.layers.22.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0007167781703174142, "dbits": 18874368 } ] }, { "idx": 45, "layers": [ "model.language_model.layers.22.mlp.experts.0.gate_proj", "model.language_model.layers.22.mlp.experts.1.gate_proj", "model.language_model.layers.22.mlp.experts.2.gate_proj", "model.language_model.layers.22.mlp.experts.3.gate_proj", "model.language_model.layers.22.mlp.experts.4.gate_proj", "model.language_model.layers.22.mlp.experts.5.gate_proj", "model.language_model.layers.22.mlp.experts.6.gate_proj", "model.language_model.layers.22.mlp.experts.7.gate_proj", "model.language_model.layers.22.mlp.experts.8.gate_proj", "model.language_model.layers.22.mlp.experts.9.gate_proj", "model.language_model.layers.22.mlp.experts.10.gate_proj", "model.language_model.layers.22.mlp.experts.11.gate_proj", "model.language_model.layers.22.mlp.experts.12.gate_proj", "model.language_model.layers.22.mlp.experts.13.gate_proj", "model.language_model.layers.22.mlp.experts.14.gate_proj", "model.language_model.layers.22.mlp.experts.15.gate_proj", "model.language_model.layers.22.mlp.experts.16.gate_proj", "model.language_model.layers.22.mlp.experts.17.gate_proj", "model.language_model.layers.22.mlp.experts.18.gate_proj", "model.language_model.layers.22.mlp.experts.19.gate_proj", "model.language_model.layers.22.mlp.experts.20.gate_proj", "model.language_model.layers.22.mlp.experts.21.gate_proj", "model.language_model.layers.22.mlp.experts.22.gate_proj", "model.language_model.layers.22.mlp.experts.23.gate_proj", "model.language_model.layers.22.mlp.experts.24.gate_proj", "model.language_model.layers.22.mlp.experts.25.gate_proj", "model.language_model.layers.22.mlp.experts.26.gate_proj", "model.language_model.layers.22.mlp.experts.27.gate_proj", "model.language_model.layers.22.mlp.experts.28.gate_proj", "model.language_model.layers.22.mlp.experts.29.gate_proj", "model.language_model.layers.22.mlp.experts.30.gate_proj", "model.language_model.layers.22.mlp.experts.31.gate_proj", "model.language_model.layers.22.mlp.experts.32.gate_proj", "model.language_model.layers.22.mlp.experts.33.gate_proj", "model.language_model.layers.22.mlp.experts.34.gate_proj", "model.language_model.layers.22.mlp.experts.35.gate_proj", "model.language_model.layers.22.mlp.experts.36.gate_proj", "model.language_model.layers.22.mlp.experts.37.gate_proj", "model.language_model.layers.22.mlp.experts.38.gate_proj", "model.language_model.layers.22.mlp.experts.39.gate_proj", "model.language_model.layers.22.mlp.experts.40.gate_proj", "model.language_model.layers.22.mlp.experts.41.gate_proj", "model.language_model.layers.22.mlp.experts.42.gate_proj", "model.language_model.layers.22.mlp.experts.43.gate_proj", "model.language_model.layers.22.mlp.experts.44.gate_proj", "model.language_model.layers.22.mlp.experts.45.gate_proj", "model.language_model.layers.22.mlp.experts.46.gate_proj", "model.language_model.layers.22.mlp.experts.47.gate_proj", "model.language_model.layers.22.mlp.experts.48.gate_proj", "model.language_model.layers.22.mlp.experts.49.gate_proj", "model.language_model.layers.22.mlp.experts.50.gate_proj", "model.language_model.layers.22.mlp.experts.51.gate_proj", "model.language_model.layers.22.mlp.experts.52.gate_proj", "model.language_model.layers.22.mlp.experts.53.gate_proj", "model.language_model.layers.22.mlp.experts.54.gate_proj", "model.language_model.layers.22.mlp.experts.55.gate_proj", "model.language_model.layers.22.mlp.experts.56.gate_proj", "model.language_model.layers.22.mlp.experts.57.gate_proj", "model.language_model.layers.22.mlp.experts.58.gate_proj", "model.language_model.layers.22.mlp.experts.59.gate_proj", "model.language_model.layers.22.mlp.experts.60.gate_proj", "model.language_model.layers.22.mlp.experts.61.gate_proj", "model.language_model.layers.22.mlp.experts.62.gate_proj", "model.language_model.layers.22.mlp.experts.63.gate_proj", "model.language_model.layers.22.mlp.experts.64.gate_proj", "model.language_model.layers.22.mlp.experts.65.gate_proj", "model.language_model.layers.22.mlp.experts.66.gate_proj", "model.language_model.layers.22.mlp.experts.67.gate_proj", "model.language_model.layers.22.mlp.experts.68.gate_proj", "model.language_model.layers.22.mlp.experts.69.gate_proj", "model.language_model.layers.22.mlp.experts.70.gate_proj", "model.language_model.layers.22.mlp.experts.71.gate_proj", "model.language_model.layers.22.mlp.experts.72.gate_proj", "model.language_model.layers.22.mlp.experts.73.gate_proj", "model.language_model.layers.22.mlp.experts.74.gate_proj", "model.language_model.layers.22.mlp.experts.75.gate_proj", "model.language_model.layers.22.mlp.experts.76.gate_proj", "model.language_model.layers.22.mlp.experts.77.gate_proj", "model.language_model.layers.22.mlp.experts.78.gate_proj", "model.language_model.layers.22.mlp.experts.79.gate_proj", "model.language_model.layers.22.mlp.experts.80.gate_proj", "model.language_model.layers.22.mlp.experts.81.gate_proj", "model.language_model.layers.22.mlp.experts.82.gate_proj", "model.language_model.layers.22.mlp.experts.83.gate_proj", "model.language_model.layers.22.mlp.experts.84.gate_proj", "model.language_model.layers.22.mlp.experts.85.gate_proj", "model.language_model.layers.22.mlp.experts.86.gate_proj", "model.language_model.layers.22.mlp.experts.87.gate_proj", "model.language_model.layers.22.mlp.experts.88.gate_proj", "model.language_model.layers.22.mlp.experts.89.gate_proj", "model.language_model.layers.22.mlp.experts.90.gate_proj", "model.language_model.layers.22.mlp.experts.91.gate_proj", "model.language_model.layers.22.mlp.experts.92.gate_proj", "model.language_model.layers.22.mlp.experts.93.gate_proj", "model.language_model.layers.22.mlp.experts.94.gate_proj", "model.language_model.layers.22.mlp.experts.95.gate_proj", "model.language_model.layers.22.mlp.experts.96.gate_proj", "model.language_model.layers.22.mlp.experts.97.gate_proj", "model.language_model.layers.22.mlp.experts.98.gate_proj", "model.language_model.layers.22.mlp.experts.99.gate_proj", "model.language_model.layers.22.mlp.experts.100.gate_proj", "model.language_model.layers.22.mlp.experts.101.gate_proj", "model.language_model.layers.22.mlp.experts.102.gate_proj", "model.language_model.layers.22.mlp.experts.103.gate_proj", "model.language_model.layers.22.mlp.experts.104.gate_proj", "model.language_model.layers.22.mlp.experts.105.gate_proj", "model.language_model.layers.22.mlp.experts.106.gate_proj", "model.language_model.layers.22.mlp.experts.107.gate_proj", "model.language_model.layers.22.mlp.experts.108.gate_proj", "model.language_model.layers.22.mlp.experts.109.gate_proj", "model.language_model.layers.22.mlp.experts.110.gate_proj", "model.language_model.layers.22.mlp.experts.111.gate_proj", "model.language_model.layers.22.mlp.experts.112.gate_proj", "model.language_model.layers.22.mlp.experts.113.gate_proj", "model.language_model.layers.22.mlp.experts.114.gate_proj", "model.language_model.layers.22.mlp.experts.115.gate_proj", "model.language_model.layers.22.mlp.experts.116.gate_proj", "model.language_model.layers.22.mlp.experts.117.gate_proj", "model.language_model.layers.22.mlp.experts.118.gate_proj", "model.language_model.layers.22.mlp.experts.119.gate_proj", "model.language_model.layers.22.mlp.experts.120.gate_proj", "model.language_model.layers.22.mlp.experts.121.gate_proj", "model.language_model.layers.22.mlp.experts.122.gate_proj", "model.language_model.layers.22.mlp.experts.123.gate_proj", "model.language_model.layers.22.mlp.experts.124.gate_proj", "model.language_model.layers.22.mlp.experts.125.gate_proj", "model.language_model.layers.22.mlp.experts.126.gate_proj", "model.language_model.layers.22.mlp.experts.127.gate_proj", "model.language_model.layers.22.mlp.experts.0.up_proj", "model.language_model.layers.22.mlp.experts.1.up_proj", "model.language_model.layers.22.mlp.experts.2.up_proj", "model.language_model.layers.22.mlp.experts.3.up_proj", "model.language_model.layers.22.mlp.experts.4.up_proj", "model.language_model.layers.22.mlp.experts.5.up_proj", "model.language_model.layers.22.mlp.experts.6.up_proj", "model.language_model.layers.22.mlp.experts.7.up_proj", "model.language_model.layers.22.mlp.experts.8.up_proj", "model.language_model.layers.22.mlp.experts.9.up_proj", "model.language_model.layers.22.mlp.experts.10.up_proj", "model.language_model.layers.22.mlp.experts.11.up_proj", "model.language_model.layers.22.mlp.experts.12.up_proj", "model.language_model.layers.22.mlp.experts.13.up_proj", "model.language_model.layers.22.mlp.experts.14.up_proj", "model.language_model.layers.22.mlp.experts.15.up_proj", "model.language_model.layers.22.mlp.experts.16.up_proj", "model.language_model.layers.22.mlp.experts.17.up_proj", "model.language_model.layers.22.mlp.experts.18.up_proj", "model.language_model.layers.22.mlp.experts.19.up_proj", "model.language_model.layers.22.mlp.experts.20.up_proj", "model.language_model.layers.22.mlp.experts.21.up_proj", "model.language_model.layers.22.mlp.experts.22.up_proj", "model.language_model.layers.22.mlp.experts.23.up_proj", "model.language_model.layers.22.mlp.experts.24.up_proj", "model.language_model.layers.22.mlp.experts.25.up_proj", "model.language_model.layers.22.mlp.experts.26.up_proj", "model.language_model.layers.22.mlp.experts.27.up_proj", "model.language_model.layers.22.mlp.experts.28.up_proj", "model.language_model.layers.22.mlp.experts.29.up_proj", "model.language_model.layers.22.mlp.experts.30.up_proj", "model.language_model.layers.22.mlp.experts.31.up_proj", "model.language_model.layers.22.mlp.experts.32.up_proj", "model.language_model.layers.22.mlp.experts.33.up_proj", "model.language_model.layers.22.mlp.experts.34.up_proj", "model.language_model.layers.22.mlp.experts.35.up_proj", "model.language_model.layers.22.mlp.experts.36.up_proj", "model.language_model.layers.22.mlp.experts.37.up_proj", "model.language_model.layers.22.mlp.experts.38.up_proj", "model.language_model.layers.22.mlp.experts.39.up_proj", "model.language_model.layers.22.mlp.experts.40.up_proj", "model.language_model.layers.22.mlp.experts.41.up_proj", "model.language_model.layers.22.mlp.experts.42.up_proj", "model.language_model.layers.22.mlp.experts.43.up_proj", "model.language_model.layers.22.mlp.experts.44.up_proj", "model.language_model.layers.22.mlp.experts.45.up_proj", "model.language_model.layers.22.mlp.experts.46.up_proj", "model.language_model.layers.22.mlp.experts.47.up_proj", "model.language_model.layers.22.mlp.experts.48.up_proj", "model.language_model.layers.22.mlp.experts.49.up_proj", "model.language_model.layers.22.mlp.experts.50.up_proj", "model.language_model.layers.22.mlp.experts.51.up_proj", "model.language_model.layers.22.mlp.experts.52.up_proj", "model.language_model.layers.22.mlp.experts.53.up_proj", "model.language_model.layers.22.mlp.experts.54.up_proj", "model.language_model.layers.22.mlp.experts.55.up_proj", "model.language_model.layers.22.mlp.experts.56.up_proj", "model.language_model.layers.22.mlp.experts.57.up_proj", "model.language_model.layers.22.mlp.experts.58.up_proj", "model.language_model.layers.22.mlp.experts.59.up_proj", "model.language_model.layers.22.mlp.experts.60.up_proj", "model.language_model.layers.22.mlp.experts.61.up_proj", "model.language_model.layers.22.mlp.experts.62.up_proj", "model.language_model.layers.22.mlp.experts.63.up_proj", "model.language_model.layers.22.mlp.experts.64.up_proj", "model.language_model.layers.22.mlp.experts.65.up_proj", "model.language_model.layers.22.mlp.experts.66.up_proj", "model.language_model.layers.22.mlp.experts.67.up_proj", "model.language_model.layers.22.mlp.experts.68.up_proj", "model.language_model.layers.22.mlp.experts.69.up_proj", "model.language_model.layers.22.mlp.experts.70.up_proj", "model.language_model.layers.22.mlp.experts.71.up_proj", "model.language_model.layers.22.mlp.experts.72.up_proj", "model.language_model.layers.22.mlp.experts.73.up_proj", "model.language_model.layers.22.mlp.experts.74.up_proj", "model.language_model.layers.22.mlp.experts.75.up_proj", "model.language_model.layers.22.mlp.experts.76.up_proj", "model.language_model.layers.22.mlp.experts.77.up_proj", "model.language_model.layers.22.mlp.experts.78.up_proj", "model.language_model.layers.22.mlp.experts.79.up_proj", "model.language_model.layers.22.mlp.experts.80.up_proj", "model.language_model.layers.22.mlp.experts.81.up_proj", "model.language_model.layers.22.mlp.experts.82.up_proj", "model.language_model.layers.22.mlp.experts.83.up_proj", "model.language_model.layers.22.mlp.experts.84.up_proj", "model.language_model.layers.22.mlp.experts.85.up_proj", "model.language_model.layers.22.mlp.experts.86.up_proj", "model.language_model.layers.22.mlp.experts.87.up_proj", "model.language_model.layers.22.mlp.experts.88.up_proj", "model.language_model.layers.22.mlp.experts.89.up_proj", "model.language_model.layers.22.mlp.experts.90.up_proj", "model.language_model.layers.22.mlp.experts.91.up_proj", "model.language_model.layers.22.mlp.experts.92.up_proj", "model.language_model.layers.22.mlp.experts.93.up_proj", "model.language_model.layers.22.mlp.experts.94.up_proj", "model.language_model.layers.22.mlp.experts.95.up_proj", "model.language_model.layers.22.mlp.experts.96.up_proj", "model.language_model.layers.22.mlp.experts.97.up_proj", "model.language_model.layers.22.mlp.experts.98.up_proj", "model.language_model.layers.22.mlp.experts.99.up_proj", "model.language_model.layers.22.mlp.experts.100.up_proj", "model.language_model.layers.22.mlp.experts.101.up_proj", "model.language_model.layers.22.mlp.experts.102.up_proj", "model.language_model.layers.22.mlp.experts.103.up_proj", "model.language_model.layers.22.mlp.experts.104.up_proj", "model.language_model.layers.22.mlp.experts.105.up_proj", "model.language_model.layers.22.mlp.experts.106.up_proj", "model.language_model.layers.22.mlp.experts.107.up_proj", "model.language_model.layers.22.mlp.experts.108.up_proj", "model.language_model.layers.22.mlp.experts.109.up_proj", "model.language_model.layers.22.mlp.experts.110.up_proj", "model.language_model.layers.22.mlp.experts.111.up_proj", "model.language_model.layers.22.mlp.experts.112.up_proj", "model.language_model.layers.22.mlp.experts.113.up_proj", "model.language_model.layers.22.mlp.experts.114.up_proj", "model.language_model.layers.22.mlp.experts.115.up_proj", "model.language_model.layers.22.mlp.experts.116.up_proj", "model.language_model.layers.22.mlp.experts.117.up_proj", "model.language_model.layers.22.mlp.experts.118.up_proj", "model.language_model.layers.22.mlp.experts.119.up_proj", "model.language_model.layers.22.mlp.experts.120.up_proj", "model.language_model.layers.22.mlp.experts.121.up_proj", "model.language_model.layers.22.mlp.experts.122.up_proj", "model.language_model.layers.22.mlp.experts.123.up_proj", "model.language_model.layers.22.mlp.experts.124.up_proj", "model.language_model.layers.22.mlp.experts.125.up_proj", "model.language_model.layers.22.mlp.experts.126.up_proj", "model.language_model.layers.22.mlp.experts.127.up_proj", "model.language_model.layers.22.mlp.experts.0.down_proj", "model.language_model.layers.22.mlp.experts.1.down_proj", "model.language_model.layers.22.mlp.experts.2.down_proj", "model.language_model.layers.22.mlp.experts.3.down_proj", "model.language_model.layers.22.mlp.experts.4.down_proj", "model.language_model.layers.22.mlp.experts.5.down_proj", "model.language_model.layers.22.mlp.experts.6.down_proj", "model.language_model.layers.22.mlp.experts.7.down_proj", "model.language_model.layers.22.mlp.experts.8.down_proj", "model.language_model.layers.22.mlp.experts.9.down_proj", "model.language_model.layers.22.mlp.experts.10.down_proj", "model.language_model.layers.22.mlp.experts.11.down_proj", "model.language_model.layers.22.mlp.experts.12.down_proj", "model.language_model.layers.22.mlp.experts.13.down_proj", "model.language_model.layers.22.mlp.experts.14.down_proj", "model.language_model.layers.22.mlp.experts.15.down_proj", "model.language_model.layers.22.mlp.experts.16.down_proj", "model.language_model.layers.22.mlp.experts.17.down_proj", "model.language_model.layers.22.mlp.experts.18.down_proj", "model.language_model.layers.22.mlp.experts.19.down_proj", "model.language_model.layers.22.mlp.experts.20.down_proj", "model.language_model.layers.22.mlp.experts.21.down_proj", "model.language_model.layers.22.mlp.experts.22.down_proj", "model.language_model.layers.22.mlp.experts.23.down_proj", "model.language_model.layers.22.mlp.experts.24.down_proj", "model.language_model.layers.22.mlp.experts.25.down_proj", "model.language_model.layers.22.mlp.experts.26.down_proj", "model.language_model.layers.22.mlp.experts.27.down_proj", "model.language_model.layers.22.mlp.experts.28.down_proj", "model.language_model.layers.22.mlp.experts.29.down_proj", "model.language_model.layers.22.mlp.experts.30.down_proj", "model.language_model.layers.22.mlp.experts.31.down_proj", "model.language_model.layers.22.mlp.experts.32.down_proj", "model.language_model.layers.22.mlp.experts.33.down_proj", "model.language_model.layers.22.mlp.experts.34.down_proj", "model.language_model.layers.22.mlp.experts.35.down_proj", "model.language_model.layers.22.mlp.experts.36.down_proj", "model.language_model.layers.22.mlp.experts.37.down_proj", "model.language_model.layers.22.mlp.experts.38.down_proj", "model.language_model.layers.22.mlp.experts.39.down_proj", "model.language_model.layers.22.mlp.experts.40.down_proj", "model.language_model.layers.22.mlp.experts.41.down_proj", "model.language_model.layers.22.mlp.experts.42.down_proj", "model.language_model.layers.22.mlp.experts.43.down_proj", "model.language_model.layers.22.mlp.experts.44.down_proj", "model.language_model.layers.22.mlp.experts.45.down_proj", "model.language_model.layers.22.mlp.experts.46.down_proj", "model.language_model.layers.22.mlp.experts.47.down_proj", "model.language_model.layers.22.mlp.experts.48.down_proj", "model.language_model.layers.22.mlp.experts.49.down_proj", "model.language_model.layers.22.mlp.experts.50.down_proj", "model.language_model.layers.22.mlp.experts.51.down_proj", "model.language_model.layers.22.mlp.experts.52.down_proj", "model.language_model.layers.22.mlp.experts.53.down_proj", "model.language_model.layers.22.mlp.experts.54.down_proj", "model.language_model.layers.22.mlp.experts.55.down_proj", "model.language_model.layers.22.mlp.experts.56.down_proj", "model.language_model.layers.22.mlp.experts.57.down_proj", "model.language_model.layers.22.mlp.experts.58.down_proj", "model.language_model.layers.22.mlp.experts.59.down_proj", "model.language_model.layers.22.mlp.experts.60.down_proj", "model.language_model.layers.22.mlp.experts.61.down_proj", "model.language_model.layers.22.mlp.experts.62.down_proj", "model.language_model.layers.22.mlp.experts.63.down_proj", "model.language_model.layers.22.mlp.experts.64.down_proj", "model.language_model.layers.22.mlp.experts.65.down_proj", "model.language_model.layers.22.mlp.experts.66.down_proj", "model.language_model.layers.22.mlp.experts.67.down_proj", "model.language_model.layers.22.mlp.experts.68.down_proj", "model.language_model.layers.22.mlp.experts.69.down_proj", "model.language_model.layers.22.mlp.experts.70.down_proj", "model.language_model.layers.22.mlp.experts.71.down_proj", "model.language_model.layers.22.mlp.experts.72.down_proj", "model.language_model.layers.22.mlp.experts.73.down_proj", "model.language_model.layers.22.mlp.experts.74.down_proj", "model.language_model.layers.22.mlp.experts.75.down_proj", "model.language_model.layers.22.mlp.experts.76.down_proj", "model.language_model.layers.22.mlp.experts.77.down_proj", "model.language_model.layers.22.mlp.experts.78.down_proj", "model.language_model.layers.22.mlp.experts.79.down_proj", "model.language_model.layers.22.mlp.experts.80.down_proj", "model.language_model.layers.22.mlp.experts.81.down_proj", "model.language_model.layers.22.mlp.experts.82.down_proj", "model.language_model.layers.22.mlp.experts.83.down_proj", "model.language_model.layers.22.mlp.experts.84.down_proj", "model.language_model.layers.22.mlp.experts.85.down_proj", "model.language_model.layers.22.mlp.experts.86.down_proj", "model.language_model.layers.22.mlp.experts.87.down_proj", "model.language_model.layers.22.mlp.experts.88.down_proj", "model.language_model.layers.22.mlp.experts.89.down_proj", "model.language_model.layers.22.mlp.experts.90.down_proj", "model.language_model.layers.22.mlp.experts.91.down_proj", "model.language_model.layers.22.mlp.experts.92.down_proj", "model.language_model.layers.22.mlp.experts.93.down_proj", "model.language_model.layers.22.mlp.experts.94.down_proj", "model.language_model.layers.22.mlp.experts.95.down_proj", "model.language_model.layers.22.mlp.experts.96.down_proj", "model.language_model.layers.22.mlp.experts.97.down_proj", "model.language_model.layers.22.mlp.experts.98.down_proj", "model.language_model.layers.22.mlp.experts.99.down_proj", "model.language_model.layers.22.mlp.experts.100.down_proj", "model.language_model.layers.22.mlp.experts.101.down_proj", "model.language_model.layers.22.mlp.experts.102.down_proj", "model.language_model.layers.22.mlp.experts.103.down_proj", "model.language_model.layers.22.mlp.experts.104.down_proj", "model.language_model.layers.22.mlp.experts.105.down_proj", "model.language_model.layers.22.mlp.experts.106.down_proj", "model.language_model.layers.22.mlp.experts.107.down_proj", "model.language_model.layers.22.mlp.experts.108.down_proj", "model.language_model.layers.22.mlp.experts.109.down_proj", "model.language_model.layers.22.mlp.experts.110.down_proj", "model.language_model.layers.22.mlp.experts.111.down_proj", "model.language_model.layers.22.mlp.experts.112.down_proj", "model.language_model.layers.22.mlp.experts.113.down_proj", "model.language_model.layers.22.mlp.experts.114.down_proj", "model.language_model.layers.22.mlp.experts.115.down_proj", "model.language_model.layers.22.mlp.experts.116.down_proj", "model.language_model.layers.22.mlp.experts.117.down_proj", "model.language_model.layers.22.mlp.experts.118.down_proj", "model.language_model.layers.22.mlp.experts.119.down_proj", "model.language_model.layers.22.mlp.experts.120.down_proj", "model.language_model.layers.22.mlp.experts.121.down_proj", "model.language_model.layers.22.mlp.experts.122.down_proj", "model.language_model.layers.22.mlp.experts.123.down_proj", "model.language_model.layers.22.mlp.experts.124.down_proj", "model.language_model.layers.22.mlp.experts.125.down_proj", "model.language_model.layers.22.mlp.experts.126.down_proj", "model.language_model.layers.22.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.0022956071421504076, "dbits": 603979776 } ] }, { "idx": 46, "layers": [ "model.language_model.layers.23.self_attn.q_proj", "model.language_model.layers.23.self_attn.k_proj", "model.language_model.layers.23.self_attn.v_proj", "model.language_model.layers.23.self_attn.o_proj" ], "candidates": [ { "dkld": 0.0003650248050689753, "dbits": 18874368 } ] }, { "idx": 47, "layers": [ "model.language_model.layers.23.mlp.experts.0.gate_proj", "model.language_model.layers.23.mlp.experts.1.gate_proj", "model.language_model.layers.23.mlp.experts.2.gate_proj", "model.language_model.layers.23.mlp.experts.3.gate_proj", "model.language_model.layers.23.mlp.experts.4.gate_proj", "model.language_model.layers.23.mlp.experts.5.gate_proj", "model.language_model.layers.23.mlp.experts.6.gate_proj", "model.language_model.layers.23.mlp.experts.7.gate_proj", "model.language_model.layers.23.mlp.experts.8.gate_proj", "model.language_model.layers.23.mlp.experts.9.gate_proj", "model.language_model.layers.23.mlp.experts.10.gate_proj", "model.language_model.layers.23.mlp.experts.11.gate_proj", "model.language_model.layers.23.mlp.experts.12.gate_proj", "model.language_model.layers.23.mlp.experts.13.gate_proj", "model.language_model.layers.23.mlp.experts.14.gate_proj", "model.language_model.layers.23.mlp.experts.15.gate_proj", "model.language_model.layers.23.mlp.experts.16.gate_proj", "model.language_model.layers.23.mlp.experts.17.gate_proj", "model.language_model.layers.23.mlp.experts.18.gate_proj", "model.language_model.layers.23.mlp.experts.19.gate_proj", "model.language_model.layers.23.mlp.experts.20.gate_proj", "model.language_model.layers.23.mlp.experts.21.gate_proj", "model.language_model.layers.23.mlp.experts.22.gate_proj", "model.language_model.layers.23.mlp.experts.23.gate_proj", "model.language_model.layers.23.mlp.experts.24.gate_proj", "model.language_model.layers.23.mlp.experts.25.gate_proj", "model.language_model.layers.23.mlp.experts.26.gate_proj", "model.language_model.layers.23.mlp.experts.27.gate_proj", "model.language_model.layers.23.mlp.experts.28.gate_proj", "model.language_model.layers.23.mlp.experts.29.gate_proj", "model.language_model.layers.23.mlp.experts.30.gate_proj", "model.language_model.layers.23.mlp.experts.31.gate_proj", "model.language_model.layers.23.mlp.experts.32.gate_proj", "model.language_model.layers.23.mlp.experts.33.gate_proj", "model.language_model.layers.23.mlp.experts.34.gate_proj", "model.language_model.layers.23.mlp.experts.35.gate_proj", "model.language_model.layers.23.mlp.experts.36.gate_proj", "model.language_model.layers.23.mlp.experts.37.gate_proj", "model.language_model.layers.23.mlp.experts.38.gate_proj", "model.language_model.layers.23.mlp.experts.39.gate_proj", "model.language_model.layers.23.mlp.experts.40.gate_proj", "model.language_model.layers.23.mlp.experts.41.gate_proj", "model.language_model.layers.23.mlp.experts.42.gate_proj", "model.language_model.layers.23.mlp.experts.43.gate_proj", "model.language_model.layers.23.mlp.experts.44.gate_proj", "model.language_model.layers.23.mlp.experts.45.gate_proj", "model.language_model.layers.23.mlp.experts.46.gate_proj", "model.language_model.layers.23.mlp.experts.47.gate_proj", "model.language_model.layers.23.mlp.experts.48.gate_proj", "model.language_model.layers.23.mlp.experts.49.gate_proj", "model.language_model.layers.23.mlp.experts.50.gate_proj", "model.language_model.layers.23.mlp.experts.51.gate_proj", "model.language_model.layers.23.mlp.experts.52.gate_proj", "model.language_model.layers.23.mlp.experts.53.gate_proj", "model.language_model.layers.23.mlp.experts.54.gate_proj", "model.language_model.layers.23.mlp.experts.55.gate_proj", "model.language_model.layers.23.mlp.experts.56.gate_proj", "model.language_model.layers.23.mlp.experts.57.gate_proj", "model.language_model.layers.23.mlp.experts.58.gate_proj", "model.language_model.layers.23.mlp.experts.59.gate_proj", "model.language_model.layers.23.mlp.experts.60.gate_proj", "model.language_model.layers.23.mlp.experts.61.gate_proj", "model.language_model.layers.23.mlp.experts.62.gate_proj", "model.language_model.layers.23.mlp.experts.63.gate_proj", "model.language_model.layers.23.mlp.experts.64.gate_proj", "model.language_model.layers.23.mlp.experts.65.gate_proj", "model.language_model.layers.23.mlp.experts.66.gate_proj", "model.language_model.layers.23.mlp.experts.67.gate_proj", "model.language_model.layers.23.mlp.experts.68.gate_proj", "model.language_model.layers.23.mlp.experts.69.gate_proj", "model.language_model.layers.23.mlp.experts.70.gate_proj", "model.language_model.layers.23.mlp.experts.71.gate_proj", "model.language_model.layers.23.mlp.experts.72.gate_proj", "model.language_model.layers.23.mlp.experts.73.gate_proj", "model.language_model.layers.23.mlp.experts.74.gate_proj", "model.language_model.layers.23.mlp.experts.75.gate_proj", "model.language_model.layers.23.mlp.experts.76.gate_proj", "model.language_model.layers.23.mlp.experts.77.gate_proj", "model.language_model.layers.23.mlp.experts.78.gate_proj", "model.language_model.layers.23.mlp.experts.79.gate_proj", "model.language_model.layers.23.mlp.experts.80.gate_proj", "model.language_model.layers.23.mlp.experts.81.gate_proj", "model.language_model.layers.23.mlp.experts.82.gate_proj", "model.language_model.layers.23.mlp.experts.83.gate_proj", "model.language_model.layers.23.mlp.experts.84.gate_proj", "model.language_model.layers.23.mlp.experts.85.gate_proj", "model.language_model.layers.23.mlp.experts.86.gate_proj", "model.language_model.layers.23.mlp.experts.87.gate_proj", "model.language_model.layers.23.mlp.experts.88.gate_proj", "model.language_model.layers.23.mlp.experts.89.gate_proj", "model.language_model.layers.23.mlp.experts.90.gate_proj", "model.language_model.layers.23.mlp.experts.91.gate_proj", "model.language_model.layers.23.mlp.experts.92.gate_proj", "model.language_model.layers.23.mlp.experts.93.gate_proj", "model.language_model.layers.23.mlp.experts.94.gate_proj", "model.language_model.layers.23.mlp.experts.95.gate_proj", "model.language_model.layers.23.mlp.experts.96.gate_proj", "model.language_model.layers.23.mlp.experts.97.gate_proj", "model.language_model.layers.23.mlp.experts.98.gate_proj", "model.language_model.layers.23.mlp.experts.99.gate_proj", "model.language_model.layers.23.mlp.experts.100.gate_proj", "model.language_model.layers.23.mlp.experts.101.gate_proj", "model.language_model.layers.23.mlp.experts.102.gate_proj", "model.language_model.layers.23.mlp.experts.103.gate_proj", "model.language_model.layers.23.mlp.experts.104.gate_proj", "model.language_model.layers.23.mlp.experts.105.gate_proj", "model.language_model.layers.23.mlp.experts.106.gate_proj", "model.language_model.layers.23.mlp.experts.107.gate_proj", "model.language_model.layers.23.mlp.experts.108.gate_proj", "model.language_model.layers.23.mlp.experts.109.gate_proj", "model.language_model.layers.23.mlp.experts.110.gate_proj", "model.language_model.layers.23.mlp.experts.111.gate_proj", "model.language_model.layers.23.mlp.experts.112.gate_proj", "model.language_model.layers.23.mlp.experts.113.gate_proj", "model.language_model.layers.23.mlp.experts.114.gate_proj", "model.language_model.layers.23.mlp.experts.115.gate_proj", "model.language_model.layers.23.mlp.experts.116.gate_proj", "model.language_model.layers.23.mlp.experts.117.gate_proj", "model.language_model.layers.23.mlp.experts.118.gate_proj", "model.language_model.layers.23.mlp.experts.119.gate_proj", "model.language_model.layers.23.mlp.experts.120.gate_proj", "model.language_model.layers.23.mlp.experts.121.gate_proj", "model.language_model.layers.23.mlp.experts.122.gate_proj", "model.language_model.layers.23.mlp.experts.123.gate_proj", "model.language_model.layers.23.mlp.experts.124.gate_proj", "model.language_model.layers.23.mlp.experts.125.gate_proj", "model.language_model.layers.23.mlp.experts.126.gate_proj", "model.language_model.layers.23.mlp.experts.127.gate_proj", "model.language_model.layers.23.mlp.experts.0.up_proj", "model.language_model.layers.23.mlp.experts.1.up_proj", "model.language_model.layers.23.mlp.experts.2.up_proj", "model.language_model.layers.23.mlp.experts.3.up_proj", "model.language_model.layers.23.mlp.experts.4.up_proj", "model.language_model.layers.23.mlp.experts.5.up_proj", "model.language_model.layers.23.mlp.experts.6.up_proj", "model.language_model.layers.23.mlp.experts.7.up_proj", "model.language_model.layers.23.mlp.experts.8.up_proj", "model.language_model.layers.23.mlp.experts.9.up_proj", "model.language_model.layers.23.mlp.experts.10.up_proj", "model.language_model.layers.23.mlp.experts.11.up_proj", "model.language_model.layers.23.mlp.experts.12.up_proj", "model.language_model.layers.23.mlp.experts.13.up_proj", "model.language_model.layers.23.mlp.experts.14.up_proj", "model.language_model.layers.23.mlp.experts.15.up_proj", "model.language_model.layers.23.mlp.experts.16.up_proj", "model.language_model.layers.23.mlp.experts.17.up_proj", "model.language_model.layers.23.mlp.experts.18.up_proj", "model.language_model.layers.23.mlp.experts.19.up_proj", "model.language_model.layers.23.mlp.experts.20.up_proj", "model.language_model.layers.23.mlp.experts.21.up_proj", "model.language_model.layers.23.mlp.experts.22.up_proj", "model.language_model.layers.23.mlp.experts.23.up_proj", "model.language_model.layers.23.mlp.experts.24.up_proj", "model.language_model.layers.23.mlp.experts.25.up_proj", "model.language_model.layers.23.mlp.experts.26.up_proj", "model.language_model.layers.23.mlp.experts.27.up_proj", "model.language_model.layers.23.mlp.experts.28.up_proj", "model.language_model.layers.23.mlp.experts.29.up_proj", "model.language_model.layers.23.mlp.experts.30.up_proj", "model.language_model.layers.23.mlp.experts.31.up_proj", "model.language_model.layers.23.mlp.experts.32.up_proj", "model.language_model.layers.23.mlp.experts.33.up_proj", "model.language_model.layers.23.mlp.experts.34.up_proj", "model.language_model.layers.23.mlp.experts.35.up_proj", "model.language_model.layers.23.mlp.experts.36.up_proj", "model.language_model.layers.23.mlp.experts.37.up_proj", "model.language_model.layers.23.mlp.experts.38.up_proj", "model.language_model.layers.23.mlp.experts.39.up_proj", "model.language_model.layers.23.mlp.experts.40.up_proj", "model.language_model.layers.23.mlp.experts.41.up_proj", "model.language_model.layers.23.mlp.experts.42.up_proj", "model.language_model.layers.23.mlp.experts.43.up_proj", "model.language_model.layers.23.mlp.experts.44.up_proj", "model.language_model.layers.23.mlp.experts.45.up_proj", "model.language_model.layers.23.mlp.experts.46.up_proj", "model.language_model.layers.23.mlp.experts.47.up_proj", "model.language_model.layers.23.mlp.experts.48.up_proj", "model.language_model.layers.23.mlp.experts.49.up_proj", "model.language_model.layers.23.mlp.experts.50.up_proj", "model.language_model.layers.23.mlp.experts.51.up_proj", "model.language_model.layers.23.mlp.experts.52.up_proj", "model.language_model.layers.23.mlp.experts.53.up_proj", "model.language_model.layers.23.mlp.experts.54.up_proj", "model.language_model.layers.23.mlp.experts.55.up_proj", "model.language_model.layers.23.mlp.experts.56.up_proj", "model.language_model.layers.23.mlp.experts.57.up_proj", "model.language_model.layers.23.mlp.experts.58.up_proj", "model.language_model.layers.23.mlp.experts.59.up_proj", "model.language_model.layers.23.mlp.experts.60.up_proj", "model.language_model.layers.23.mlp.experts.61.up_proj", "model.language_model.layers.23.mlp.experts.62.up_proj", "model.language_model.layers.23.mlp.experts.63.up_proj", "model.language_model.layers.23.mlp.experts.64.up_proj", "model.language_model.layers.23.mlp.experts.65.up_proj", "model.language_model.layers.23.mlp.experts.66.up_proj", "model.language_model.layers.23.mlp.experts.67.up_proj", "model.language_model.layers.23.mlp.experts.68.up_proj", "model.language_model.layers.23.mlp.experts.69.up_proj", "model.language_model.layers.23.mlp.experts.70.up_proj", "model.language_model.layers.23.mlp.experts.71.up_proj", "model.language_model.layers.23.mlp.experts.72.up_proj", "model.language_model.layers.23.mlp.experts.73.up_proj", "model.language_model.layers.23.mlp.experts.74.up_proj", "model.language_model.layers.23.mlp.experts.75.up_proj", "model.language_model.layers.23.mlp.experts.76.up_proj", "model.language_model.layers.23.mlp.experts.77.up_proj", "model.language_model.layers.23.mlp.experts.78.up_proj", "model.language_model.layers.23.mlp.experts.79.up_proj", "model.language_model.layers.23.mlp.experts.80.up_proj", "model.language_model.layers.23.mlp.experts.81.up_proj", "model.language_model.layers.23.mlp.experts.82.up_proj", "model.language_model.layers.23.mlp.experts.83.up_proj", "model.language_model.layers.23.mlp.experts.84.up_proj", "model.language_model.layers.23.mlp.experts.85.up_proj", "model.language_model.layers.23.mlp.experts.86.up_proj", "model.language_model.layers.23.mlp.experts.87.up_proj", "model.language_model.layers.23.mlp.experts.88.up_proj", "model.language_model.layers.23.mlp.experts.89.up_proj", "model.language_model.layers.23.mlp.experts.90.up_proj", "model.language_model.layers.23.mlp.experts.91.up_proj", "model.language_model.layers.23.mlp.experts.92.up_proj", "model.language_model.layers.23.mlp.experts.93.up_proj", "model.language_model.layers.23.mlp.experts.94.up_proj", "model.language_model.layers.23.mlp.experts.95.up_proj", "model.language_model.layers.23.mlp.experts.96.up_proj", "model.language_model.layers.23.mlp.experts.97.up_proj", "model.language_model.layers.23.mlp.experts.98.up_proj", "model.language_model.layers.23.mlp.experts.99.up_proj", "model.language_model.layers.23.mlp.experts.100.up_proj", "model.language_model.layers.23.mlp.experts.101.up_proj", "model.language_model.layers.23.mlp.experts.102.up_proj", "model.language_model.layers.23.mlp.experts.103.up_proj", "model.language_model.layers.23.mlp.experts.104.up_proj", "model.language_model.layers.23.mlp.experts.105.up_proj", "model.language_model.layers.23.mlp.experts.106.up_proj", "model.language_model.layers.23.mlp.experts.107.up_proj", "model.language_model.layers.23.mlp.experts.108.up_proj", "model.language_model.layers.23.mlp.experts.109.up_proj", "model.language_model.layers.23.mlp.experts.110.up_proj", "model.language_model.layers.23.mlp.experts.111.up_proj", "model.language_model.layers.23.mlp.experts.112.up_proj", "model.language_model.layers.23.mlp.experts.113.up_proj", "model.language_model.layers.23.mlp.experts.114.up_proj", "model.language_model.layers.23.mlp.experts.115.up_proj", "model.language_model.layers.23.mlp.experts.116.up_proj", "model.language_model.layers.23.mlp.experts.117.up_proj", "model.language_model.layers.23.mlp.experts.118.up_proj", "model.language_model.layers.23.mlp.experts.119.up_proj", "model.language_model.layers.23.mlp.experts.120.up_proj", "model.language_model.layers.23.mlp.experts.121.up_proj", "model.language_model.layers.23.mlp.experts.122.up_proj", "model.language_model.layers.23.mlp.experts.123.up_proj", "model.language_model.layers.23.mlp.experts.124.up_proj", "model.language_model.layers.23.mlp.experts.125.up_proj", "model.language_model.layers.23.mlp.experts.126.up_proj", "model.language_model.layers.23.mlp.experts.127.up_proj", "model.language_model.layers.23.mlp.experts.0.down_proj", "model.language_model.layers.23.mlp.experts.1.down_proj", "model.language_model.layers.23.mlp.experts.2.down_proj", "model.language_model.layers.23.mlp.experts.3.down_proj", "model.language_model.layers.23.mlp.experts.4.down_proj", "model.language_model.layers.23.mlp.experts.5.down_proj", "model.language_model.layers.23.mlp.experts.6.down_proj", "model.language_model.layers.23.mlp.experts.7.down_proj", "model.language_model.layers.23.mlp.experts.8.down_proj", "model.language_model.layers.23.mlp.experts.9.down_proj", "model.language_model.layers.23.mlp.experts.10.down_proj", "model.language_model.layers.23.mlp.experts.11.down_proj", "model.language_model.layers.23.mlp.experts.12.down_proj", "model.language_model.layers.23.mlp.experts.13.down_proj", "model.language_model.layers.23.mlp.experts.14.down_proj", "model.language_model.layers.23.mlp.experts.15.down_proj", "model.language_model.layers.23.mlp.experts.16.down_proj", "model.language_model.layers.23.mlp.experts.17.down_proj", "model.language_model.layers.23.mlp.experts.18.down_proj", "model.language_model.layers.23.mlp.experts.19.down_proj", "model.language_model.layers.23.mlp.experts.20.down_proj", "model.language_model.layers.23.mlp.experts.21.down_proj", "model.language_model.layers.23.mlp.experts.22.down_proj", "model.language_model.layers.23.mlp.experts.23.down_proj", "model.language_model.layers.23.mlp.experts.24.down_proj", "model.language_model.layers.23.mlp.experts.25.down_proj", "model.language_model.layers.23.mlp.experts.26.down_proj", "model.language_model.layers.23.mlp.experts.27.down_proj", "model.language_model.layers.23.mlp.experts.28.down_proj", "model.language_model.layers.23.mlp.experts.29.down_proj", "model.language_model.layers.23.mlp.experts.30.down_proj", "model.language_model.layers.23.mlp.experts.31.down_proj", "model.language_model.layers.23.mlp.experts.32.down_proj", "model.language_model.layers.23.mlp.experts.33.down_proj", "model.language_model.layers.23.mlp.experts.34.down_proj", "model.language_model.layers.23.mlp.experts.35.down_proj", "model.language_model.layers.23.mlp.experts.36.down_proj", "model.language_model.layers.23.mlp.experts.37.down_proj", "model.language_model.layers.23.mlp.experts.38.down_proj", "model.language_model.layers.23.mlp.experts.39.down_proj", "model.language_model.layers.23.mlp.experts.40.down_proj", "model.language_model.layers.23.mlp.experts.41.down_proj", "model.language_model.layers.23.mlp.experts.42.down_proj", "model.language_model.layers.23.mlp.experts.43.down_proj", "model.language_model.layers.23.mlp.experts.44.down_proj", "model.language_model.layers.23.mlp.experts.45.down_proj", "model.language_model.layers.23.mlp.experts.46.down_proj", "model.language_model.layers.23.mlp.experts.47.down_proj", "model.language_model.layers.23.mlp.experts.48.down_proj", "model.language_model.layers.23.mlp.experts.49.down_proj", "model.language_model.layers.23.mlp.experts.50.down_proj", "model.language_model.layers.23.mlp.experts.51.down_proj", "model.language_model.layers.23.mlp.experts.52.down_proj", "model.language_model.layers.23.mlp.experts.53.down_proj", "model.language_model.layers.23.mlp.experts.54.down_proj", "model.language_model.layers.23.mlp.experts.55.down_proj", "model.language_model.layers.23.mlp.experts.56.down_proj", "model.language_model.layers.23.mlp.experts.57.down_proj", "model.language_model.layers.23.mlp.experts.58.down_proj", "model.language_model.layers.23.mlp.experts.59.down_proj", "model.language_model.layers.23.mlp.experts.60.down_proj", "model.language_model.layers.23.mlp.experts.61.down_proj", "model.language_model.layers.23.mlp.experts.62.down_proj", "model.language_model.layers.23.mlp.experts.63.down_proj", "model.language_model.layers.23.mlp.experts.64.down_proj", "model.language_model.layers.23.mlp.experts.65.down_proj", "model.language_model.layers.23.mlp.experts.66.down_proj", "model.language_model.layers.23.mlp.experts.67.down_proj", "model.language_model.layers.23.mlp.experts.68.down_proj", "model.language_model.layers.23.mlp.experts.69.down_proj", "model.language_model.layers.23.mlp.experts.70.down_proj", "model.language_model.layers.23.mlp.experts.71.down_proj", "model.language_model.layers.23.mlp.experts.72.down_proj", "model.language_model.layers.23.mlp.experts.73.down_proj", "model.language_model.layers.23.mlp.experts.74.down_proj", "model.language_model.layers.23.mlp.experts.75.down_proj", "model.language_model.layers.23.mlp.experts.76.down_proj", "model.language_model.layers.23.mlp.experts.77.down_proj", "model.language_model.layers.23.mlp.experts.78.down_proj", "model.language_model.layers.23.mlp.experts.79.down_proj", "model.language_model.layers.23.mlp.experts.80.down_proj", "model.language_model.layers.23.mlp.experts.81.down_proj", "model.language_model.layers.23.mlp.experts.82.down_proj", "model.language_model.layers.23.mlp.experts.83.down_proj", "model.language_model.layers.23.mlp.experts.84.down_proj", "model.language_model.layers.23.mlp.experts.85.down_proj", "model.language_model.layers.23.mlp.experts.86.down_proj", "model.language_model.layers.23.mlp.experts.87.down_proj", "model.language_model.layers.23.mlp.experts.88.down_proj", "model.language_model.layers.23.mlp.experts.89.down_proj", "model.language_model.layers.23.mlp.experts.90.down_proj", "model.language_model.layers.23.mlp.experts.91.down_proj", "model.language_model.layers.23.mlp.experts.92.down_proj", "model.language_model.layers.23.mlp.experts.93.down_proj", "model.language_model.layers.23.mlp.experts.94.down_proj", "model.language_model.layers.23.mlp.experts.95.down_proj", "model.language_model.layers.23.mlp.experts.96.down_proj", "model.language_model.layers.23.mlp.experts.97.down_proj", "model.language_model.layers.23.mlp.experts.98.down_proj", "model.language_model.layers.23.mlp.experts.99.down_proj", "model.language_model.layers.23.mlp.experts.100.down_proj", "model.language_model.layers.23.mlp.experts.101.down_proj", "model.language_model.layers.23.mlp.experts.102.down_proj", "model.language_model.layers.23.mlp.experts.103.down_proj", "model.language_model.layers.23.mlp.experts.104.down_proj", "model.language_model.layers.23.mlp.experts.105.down_proj", "model.language_model.layers.23.mlp.experts.106.down_proj", "model.language_model.layers.23.mlp.experts.107.down_proj", "model.language_model.layers.23.mlp.experts.108.down_proj", "model.language_model.layers.23.mlp.experts.109.down_proj", "model.language_model.layers.23.mlp.experts.110.down_proj", "model.language_model.layers.23.mlp.experts.111.down_proj", "model.language_model.layers.23.mlp.experts.112.down_proj", "model.language_model.layers.23.mlp.experts.113.down_proj", "model.language_model.layers.23.mlp.experts.114.down_proj", "model.language_model.layers.23.mlp.experts.115.down_proj", "model.language_model.layers.23.mlp.experts.116.down_proj", "model.language_model.layers.23.mlp.experts.117.down_proj", "model.language_model.layers.23.mlp.experts.118.down_proj", "model.language_model.layers.23.mlp.experts.119.down_proj", "model.language_model.layers.23.mlp.experts.120.down_proj", "model.language_model.layers.23.mlp.experts.121.down_proj", "model.language_model.layers.23.mlp.experts.122.down_proj", "model.language_model.layers.23.mlp.experts.123.down_proj", "model.language_model.layers.23.mlp.experts.124.down_proj", "model.language_model.layers.23.mlp.experts.125.down_proj", "model.language_model.layers.23.mlp.experts.126.down_proj", "model.language_model.layers.23.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.0016450084745883997, "dbits": 603979776 } ] }, { "idx": 48, "layers": [ "model.language_model.layers.24.self_attn.q_proj", "model.language_model.layers.24.self_attn.k_proj", "model.language_model.layers.24.self_attn.v_proj", "model.language_model.layers.24.self_attn.o_proj" ], "candidates": [ { "dkld": 0.0008779314346611528, "dbits": 18874368 } ] }, { "idx": 49, "layers": [ "model.language_model.layers.24.mlp.experts.0.gate_proj", "model.language_model.layers.24.mlp.experts.1.gate_proj", "model.language_model.layers.24.mlp.experts.2.gate_proj", "model.language_model.layers.24.mlp.experts.3.gate_proj", "model.language_model.layers.24.mlp.experts.4.gate_proj", "model.language_model.layers.24.mlp.experts.5.gate_proj", "model.language_model.layers.24.mlp.experts.6.gate_proj", "model.language_model.layers.24.mlp.experts.7.gate_proj", "model.language_model.layers.24.mlp.experts.8.gate_proj", "model.language_model.layers.24.mlp.experts.9.gate_proj", "model.language_model.layers.24.mlp.experts.10.gate_proj", "model.language_model.layers.24.mlp.experts.11.gate_proj", "model.language_model.layers.24.mlp.experts.12.gate_proj", "model.language_model.layers.24.mlp.experts.13.gate_proj", "model.language_model.layers.24.mlp.experts.14.gate_proj", "model.language_model.layers.24.mlp.experts.15.gate_proj", "model.language_model.layers.24.mlp.experts.16.gate_proj", "model.language_model.layers.24.mlp.experts.17.gate_proj", "model.language_model.layers.24.mlp.experts.18.gate_proj", "model.language_model.layers.24.mlp.experts.19.gate_proj", "model.language_model.layers.24.mlp.experts.20.gate_proj", "model.language_model.layers.24.mlp.experts.21.gate_proj", "model.language_model.layers.24.mlp.experts.22.gate_proj", "model.language_model.layers.24.mlp.experts.23.gate_proj", "model.language_model.layers.24.mlp.experts.24.gate_proj", "model.language_model.layers.24.mlp.experts.25.gate_proj", "model.language_model.layers.24.mlp.experts.26.gate_proj", "model.language_model.layers.24.mlp.experts.27.gate_proj", "model.language_model.layers.24.mlp.experts.28.gate_proj", "model.language_model.layers.24.mlp.experts.29.gate_proj", "model.language_model.layers.24.mlp.experts.30.gate_proj", "model.language_model.layers.24.mlp.experts.31.gate_proj", "model.language_model.layers.24.mlp.experts.32.gate_proj", "model.language_model.layers.24.mlp.experts.33.gate_proj", "model.language_model.layers.24.mlp.experts.34.gate_proj", "model.language_model.layers.24.mlp.experts.35.gate_proj", "model.language_model.layers.24.mlp.experts.36.gate_proj", "model.language_model.layers.24.mlp.experts.37.gate_proj", "model.language_model.layers.24.mlp.experts.38.gate_proj", "model.language_model.layers.24.mlp.experts.39.gate_proj", "model.language_model.layers.24.mlp.experts.40.gate_proj", "model.language_model.layers.24.mlp.experts.41.gate_proj", "model.language_model.layers.24.mlp.experts.42.gate_proj", "model.language_model.layers.24.mlp.experts.43.gate_proj", "model.language_model.layers.24.mlp.experts.44.gate_proj", "model.language_model.layers.24.mlp.experts.45.gate_proj", "model.language_model.layers.24.mlp.experts.46.gate_proj", "model.language_model.layers.24.mlp.experts.47.gate_proj", "model.language_model.layers.24.mlp.experts.48.gate_proj", "model.language_model.layers.24.mlp.experts.49.gate_proj", "model.language_model.layers.24.mlp.experts.50.gate_proj", "model.language_model.layers.24.mlp.experts.51.gate_proj", "model.language_model.layers.24.mlp.experts.52.gate_proj", "model.language_model.layers.24.mlp.experts.53.gate_proj", "model.language_model.layers.24.mlp.experts.54.gate_proj", "model.language_model.layers.24.mlp.experts.55.gate_proj", "model.language_model.layers.24.mlp.experts.56.gate_proj", "model.language_model.layers.24.mlp.experts.57.gate_proj", "model.language_model.layers.24.mlp.experts.58.gate_proj", "model.language_model.layers.24.mlp.experts.59.gate_proj", "model.language_model.layers.24.mlp.experts.60.gate_proj", "model.language_model.layers.24.mlp.experts.61.gate_proj", "model.language_model.layers.24.mlp.experts.62.gate_proj", "model.language_model.layers.24.mlp.experts.63.gate_proj", "model.language_model.layers.24.mlp.experts.64.gate_proj", "model.language_model.layers.24.mlp.experts.65.gate_proj", "model.language_model.layers.24.mlp.experts.66.gate_proj", "model.language_model.layers.24.mlp.experts.67.gate_proj", "model.language_model.layers.24.mlp.experts.68.gate_proj", "model.language_model.layers.24.mlp.experts.69.gate_proj", "model.language_model.layers.24.mlp.experts.70.gate_proj", "model.language_model.layers.24.mlp.experts.71.gate_proj", "model.language_model.layers.24.mlp.experts.72.gate_proj", "model.language_model.layers.24.mlp.experts.73.gate_proj", "model.language_model.layers.24.mlp.experts.74.gate_proj", "model.language_model.layers.24.mlp.experts.75.gate_proj", "model.language_model.layers.24.mlp.experts.76.gate_proj", "model.language_model.layers.24.mlp.experts.77.gate_proj", "model.language_model.layers.24.mlp.experts.78.gate_proj", "model.language_model.layers.24.mlp.experts.79.gate_proj", "model.language_model.layers.24.mlp.experts.80.gate_proj", "model.language_model.layers.24.mlp.experts.81.gate_proj", "model.language_model.layers.24.mlp.experts.82.gate_proj", "model.language_model.layers.24.mlp.experts.83.gate_proj", "model.language_model.layers.24.mlp.experts.84.gate_proj", "model.language_model.layers.24.mlp.experts.85.gate_proj", "model.language_model.layers.24.mlp.experts.86.gate_proj", "model.language_model.layers.24.mlp.experts.87.gate_proj", "model.language_model.layers.24.mlp.experts.88.gate_proj", "model.language_model.layers.24.mlp.experts.89.gate_proj", "model.language_model.layers.24.mlp.experts.90.gate_proj", "model.language_model.layers.24.mlp.experts.91.gate_proj", "model.language_model.layers.24.mlp.experts.92.gate_proj", "model.language_model.layers.24.mlp.experts.93.gate_proj", "model.language_model.layers.24.mlp.experts.94.gate_proj", "model.language_model.layers.24.mlp.experts.95.gate_proj", "model.language_model.layers.24.mlp.experts.96.gate_proj", "model.language_model.layers.24.mlp.experts.97.gate_proj", "model.language_model.layers.24.mlp.experts.98.gate_proj", "model.language_model.layers.24.mlp.experts.99.gate_proj", "model.language_model.layers.24.mlp.experts.100.gate_proj", "model.language_model.layers.24.mlp.experts.101.gate_proj", "model.language_model.layers.24.mlp.experts.102.gate_proj", "model.language_model.layers.24.mlp.experts.103.gate_proj", "model.language_model.layers.24.mlp.experts.104.gate_proj", "model.language_model.layers.24.mlp.experts.105.gate_proj", "model.language_model.layers.24.mlp.experts.106.gate_proj", "model.language_model.layers.24.mlp.experts.107.gate_proj", "model.language_model.layers.24.mlp.experts.108.gate_proj", "model.language_model.layers.24.mlp.experts.109.gate_proj", "model.language_model.layers.24.mlp.experts.110.gate_proj", "model.language_model.layers.24.mlp.experts.111.gate_proj", "model.language_model.layers.24.mlp.experts.112.gate_proj", "model.language_model.layers.24.mlp.experts.113.gate_proj", "model.language_model.layers.24.mlp.experts.114.gate_proj", "model.language_model.layers.24.mlp.experts.115.gate_proj", "model.language_model.layers.24.mlp.experts.116.gate_proj", "model.language_model.layers.24.mlp.experts.117.gate_proj", "model.language_model.layers.24.mlp.experts.118.gate_proj", "model.language_model.layers.24.mlp.experts.119.gate_proj", "model.language_model.layers.24.mlp.experts.120.gate_proj", "model.language_model.layers.24.mlp.experts.121.gate_proj", "model.language_model.layers.24.mlp.experts.122.gate_proj", "model.language_model.layers.24.mlp.experts.123.gate_proj", "model.language_model.layers.24.mlp.experts.124.gate_proj", "model.language_model.layers.24.mlp.experts.125.gate_proj", "model.language_model.layers.24.mlp.experts.126.gate_proj", "model.language_model.layers.24.mlp.experts.127.gate_proj", "model.language_model.layers.24.mlp.experts.0.up_proj", "model.language_model.layers.24.mlp.experts.1.up_proj", "model.language_model.layers.24.mlp.experts.2.up_proj", "model.language_model.layers.24.mlp.experts.3.up_proj", "model.language_model.layers.24.mlp.experts.4.up_proj", "model.language_model.layers.24.mlp.experts.5.up_proj", "model.language_model.layers.24.mlp.experts.6.up_proj", "model.language_model.layers.24.mlp.experts.7.up_proj", "model.language_model.layers.24.mlp.experts.8.up_proj", "model.language_model.layers.24.mlp.experts.9.up_proj", "model.language_model.layers.24.mlp.experts.10.up_proj", "model.language_model.layers.24.mlp.experts.11.up_proj", "model.language_model.layers.24.mlp.experts.12.up_proj", "model.language_model.layers.24.mlp.experts.13.up_proj", "model.language_model.layers.24.mlp.experts.14.up_proj", "model.language_model.layers.24.mlp.experts.15.up_proj", "model.language_model.layers.24.mlp.experts.16.up_proj", "model.language_model.layers.24.mlp.experts.17.up_proj", "model.language_model.layers.24.mlp.experts.18.up_proj", "model.language_model.layers.24.mlp.experts.19.up_proj", "model.language_model.layers.24.mlp.experts.20.up_proj", "model.language_model.layers.24.mlp.experts.21.up_proj", "model.language_model.layers.24.mlp.experts.22.up_proj", "model.language_model.layers.24.mlp.experts.23.up_proj", "model.language_model.layers.24.mlp.experts.24.up_proj", "model.language_model.layers.24.mlp.experts.25.up_proj", "model.language_model.layers.24.mlp.experts.26.up_proj", "model.language_model.layers.24.mlp.experts.27.up_proj", "model.language_model.layers.24.mlp.experts.28.up_proj", "model.language_model.layers.24.mlp.experts.29.up_proj", "model.language_model.layers.24.mlp.experts.30.up_proj", "model.language_model.layers.24.mlp.experts.31.up_proj", "model.language_model.layers.24.mlp.experts.32.up_proj", "model.language_model.layers.24.mlp.experts.33.up_proj", "model.language_model.layers.24.mlp.experts.34.up_proj", "model.language_model.layers.24.mlp.experts.35.up_proj", "model.language_model.layers.24.mlp.experts.36.up_proj", "model.language_model.layers.24.mlp.experts.37.up_proj", "model.language_model.layers.24.mlp.experts.38.up_proj", "model.language_model.layers.24.mlp.experts.39.up_proj", "model.language_model.layers.24.mlp.experts.40.up_proj", "model.language_model.layers.24.mlp.experts.41.up_proj", "model.language_model.layers.24.mlp.experts.42.up_proj", "model.language_model.layers.24.mlp.experts.43.up_proj", "model.language_model.layers.24.mlp.experts.44.up_proj", "model.language_model.layers.24.mlp.experts.45.up_proj", "model.language_model.layers.24.mlp.experts.46.up_proj", "model.language_model.layers.24.mlp.experts.47.up_proj", "model.language_model.layers.24.mlp.experts.48.up_proj", "model.language_model.layers.24.mlp.experts.49.up_proj", "model.language_model.layers.24.mlp.experts.50.up_proj", "model.language_model.layers.24.mlp.experts.51.up_proj", "model.language_model.layers.24.mlp.experts.52.up_proj", "model.language_model.layers.24.mlp.experts.53.up_proj", "model.language_model.layers.24.mlp.experts.54.up_proj", "model.language_model.layers.24.mlp.experts.55.up_proj", "model.language_model.layers.24.mlp.experts.56.up_proj", "model.language_model.layers.24.mlp.experts.57.up_proj", "model.language_model.layers.24.mlp.experts.58.up_proj", "model.language_model.layers.24.mlp.experts.59.up_proj", "model.language_model.layers.24.mlp.experts.60.up_proj", "model.language_model.layers.24.mlp.experts.61.up_proj", "model.language_model.layers.24.mlp.experts.62.up_proj", "model.language_model.layers.24.mlp.experts.63.up_proj", "model.language_model.layers.24.mlp.experts.64.up_proj", "model.language_model.layers.24.mlp.experts.65.up_proj", "model.language_model.layers.24.mlp.experts.66.up_proj", "model.language_model.layers.24.mlp.experts.67.up_proj", "model.language_model.layers.24.mlp.experts.68.up_proj", "model.language_model.layers.24.mlp.experts.69.up_proj", "model.language_model.layers.24.mlp.experts.70.up_proj", "model.language_model.layers.24.mlp.experts.71.up_proj", "model.language_model.layers.24.mlp.experts.72.up_proj", "model.language_model.layers.24.mlp.experts.73.up_proj", "model.language_model.layers.24.mlp.experts.74.up_proj", "model.language_model.layers.24.mlp.experts.75.up_proj", "model.language_model.layers.24.mlp.experts.76.up_proj", "model.language_model.layers.24.mlp.experts.77.up_proj", "model.language_model.layers.24.mlp.experts.78.up_proj", "model.language_model.layers.24.mlp.experts.79.up_proj", "model.language_model.layers.24.mlp.experts.80.up_proj", "model.language_model.layers.24.mlp.experts.81.up_proj", "model.language_model.layers.24.mlp.experts.82.up_proj", "model.language_model.layers.24.mlp.experts.83.up_proj", "model.language_model.layers.24.mlp.experts.84.up_proj", "model.language_model.layers.24.mlp.experts.85.up_proj", "model.language_model.layers.24.mlp.experts.86.up_proj", "model.language_model.layers.24.mlp.experts.87.up_proj", "model.language_model.layers.24.mlp.experts.88.up_proj", "model.language_model.layers.24.mlp.experts.89.up_proj", "model.language_model.layers.24.mlp.experts.90.up_proj", "model.language_model.layers.24.mlp.experts.91.up_proj", "model.language_model.layers.24.mlp.experts.92.up_proj", "model.language_model.layers.24.mlp.experts.93.up_proj", "model.language_model.layers.24.mlp.experts.94.up_proj", "model.language_model.layers.24.mlp.experts.95.up_proj", "model.language_model.layers.24.mlp.experts.96.up_proj", "model.language_model.layers.24.mlp.experts.97.up_proj", "model.language_model.layers.24.mlp.experts.98.up_proj", "model.language_model.layers.24.mlp.experts.99.up_proj", "model.language_model.layers.24.mlp.experts.100.up_proj", "model.language_model.layers.24.mlp.experts.101.up_proj", "model.language_model.layers.24.mlp.experts.102.up_proj", "model.language_model.layers.24.mlp.experts.103.up_proj", "model.language_model.layers.24.mlp.experts.104.up_proj", "model.language_model.layers.24.mlp.experts.105.up_proj", "model.language_model.layers.24.mlp.experts.106.up_proj", "model.language_model.layers.24.mlp.experts.107.up_proj", "model.language_model.layers.24.mlp.experts.108.up_proj", "model.language_model.layers.24.mlp.experts.109.up_proj", "model.language_model.layers.24.mlp.experts.110.up_proj", "model.language_model.layers.24.mlp.experts.111.up_proj", "model.language_model.layers.24.mlp.experts.112.up_proj", "model.language_model.layers.24.mlp.experts.113.up_proj", "model.language_model.layers.24.mlp.experts.114.up_proj", "model.language_model.layers.24.mlp.experts.115.up_proj", "model.language_model.layers.24.mlp.experts.116.up_proj", "model.language_model.layers.24.mlp.experts.117.up_proj", "model.language_model.layers.24.mlp.experts.118.up_proj", "model.language_model.layers.24.mlp.experts.119.up_proj", "model.language_model.layers.24.mlp.experts.120.up_proj", "model.language_model.layers.24.mlp.experts.121.up_proj", "model.language_model.layers.24.mlp.experts.122.up_proj", "model.language_model.layers.24.mlp.experts.123.up_proj", "model.language_model.layers.24.mlp.experts.124.up_proj", "model.language_model.layers.24.mlp.experts.125.up_proj", "model.language_model.layers.24.mlp.experts.126.up_proj", "model.language_model.layers.24.mlp.experts.127.up_proj", "model.language_model.layers.24.mlp.experts.0.down_proj", "model.language_model.layers.24.mlp.experts.1.down_proj", "model.language_model.layers.24.mlp.experts.2.down_proj", "model.language_model.layers.24.mlp.experts.3.down_proj", "model.language_model.layers.24.mlp.experts.4.down_proj", "model.language_model.layers.24.mlp.experts.5.down_proj", "model.language_model.layers.24.mlp.experts.6.down_proj", "model.language_model.layers.24.mlp.experts.7.down_proj", "model.language_model.layers.24.mlp.experts.8.down_proj", "model.language_model.layers.24.mlp.experts.9.down_proj", "model.language_model.layers.24.mlp.experts.10.down_proj", "model.language_model.layers.24.mlp.experts.11.down_proj", "model.language_model.layers.24.mlp.experts.12.down_proj", "model.language_model.layers.24.mlp.experts.13.down_proj", "model.language_model.layers.24.mlp.experts.14.down_proj", "model.language_model.layers.24.mlp.experts.15.down_proj", "model.language_model.layers.24.mlp.experts.16.down_proj", "model.language_model.layers.24.mlp.experts.17.down_proj", "model.language_model.layers.24.mlp.experts.18.down_proj", "model.language_model.layers.24.mlp.experts.19.down_proj", "model.language_model.layers.24.mlp.experts.20.down_proj", "model.language_model.layers.24.mlp.experts.21.down_proj", "model.language_model.layers.24.mlp.experts.22.down_proj", "model.language_model.layers.24.mlp.experts.23.down_proj", "model.language_model.layers.24.mlp.experts.24.down_proj", "model.language_model.layers.24.mlp.experts.25.down_proj", "model.language_model.layers.24.mlp.experts.26.down_proj", "model.language_model.layers.24.mlp.experts.27.down_proj", "model.language_model.layers.24.mlp.experts.28.down_proj", "model.language_model.layers.24.mlp.experts.29.down_proj", "model.language_model.layers.24.mlp.experts.30.down_proj", "model.language_model.layers.24.mlp.experts.31.down_proj", "model.language_model.layers.24.mlp.experts.32.down_proj", "model.language_model.layers.24.mlp.experts.33.down_proj", "model.language_model.layers.24.mlp.experts.34.down_proj", "model.language_model.layers.24.mlp.experts.35.down_proj", "model.language_model.layers.24.mlp.experts.36.down_proj", "model.language_model.layers.24.mlp.experts.37.down_proj", "model.language_model.layers.24.mlp.experts.38.down_proj", "model.language_model.layers.24.mlp.experts.39.down_proj", "model.language_model.layers.24.mlp.experts.40.down_proj", "model.language_model.layers.24.mlp.experts.41.down_proj", "model.language_model.layers.24.mlp.experts.42.down_proj", "model.language_model.layers.24.mlp.experts.43.down_proj", "model.language_model.layers.24.mlp.experts.44.down_proj", "model.language_model.layers.24.mlp.experts.45.down_proj", "model.language_model.layers.24.mlp.experts.46.down_proj", "model.language_model.layers.24.mlp.experts.47.down_proj", "model.language_model.layers.24.mlp.experts.48.down_proj", "model.language_model.layers.24.mlp.experts.49.down_proj", "model.language_model.layers.24.mlp.experts.50.down_proj", "model.language_model.layers.24.mlp.experts.51.down_proj", "model.language_model.layers.24.mlp.experts.52.down_proj", "model.language_model.layers.24.mlp.experts.53.down_proj", "model.language_model.layers.24.mlp.experts.54.down_proj", "model.language_model.layers.24.mlp.experts.55.down_proj", "model.language_model.layers.24.mlp.experts.56.down_proj", "model.language_model.layers.24.mlp.experts.57.down_proj", "model.language_model.layers.24.mlp.experts.58.down_proj", "model.language_model.layers.24.mlp.experts.59.down_proj", "model.language_model.layers.24.mlp.experts.60.down_proj", "model.language_model.layers.24.mlp.experts.61.down_proj", "model.language_model.layers.24.mlp.experts.62.down_proj", "model.language_model.layers.24.mlp.experts.63.down_proj", "model.language_model.layers.24.mlp.experts.64.down_proj", "model.language_model.layers.24.mlp.experts.65.down_proj", "model.language_model.layers.24.mlp.experts.66.down_proj", "model.language_model.layers.24.mlp.experts.67.down_proj", "model.language_model.layers.24.mlp.experts.68.down_proj", "model.language_model.layers.24.mlp.experts.69.down_proj", "model.language_model.layers.24.mlp.experts.70.down_proj", "model.language_model.layers.24.mlp.experts.71.down_proj", "model.language_model.layers.24.mlp.experts.72.down_proj", "model.language_model.layers.24.mlp.experts.73.down_proj", "model.language_model.layers.24.mlp.experts.74.down_proj", "model.language_model.layers.24.mlp.experts.75.down_proj", "model.language_model.layers.24.mlp.experts.76.down_proj", "model.language_model.layers.24.mlp.experts.77.down_proj", "model.language_model.layers.24.mlp.experts.78.down_proj", "model.language_model.layers.24.mlp.experts.79.down_proj", "model.language_model.layers.24.mlp.experts.80.down_proj", "model.language_model.layers.24.mlp.experts.81.down_proj", "model.language_model.layers.24.mlp.experts.82.down_proj", "model.language_model.layers.24.mlp.experts.83.down_proj", "model.language_model.layers.24.mlp.experts.84.down_proj", "model.language_model.layers.24.mlp.experts.85.down_proj", "model.language_model.layers.24.mlp.experts.86.down_proj", "model.language_model.layers.24.mlp.experts.87.down_proj", "model.language_model.layers.24.mlp.experts.88.down_proj", "model.language_model.layers.24.mlp.experts.89.down_proj", "model.language_model.layers.24.mlp.experts.90.down_proj", "model.language_model.layers.24.mlp.experts.91.down_proj", "model.language_model.layers.24.mlp.experts.92.down_proj", "model.language_model.layers.24.mlp.experts.93.down_proj", "model.language_model.layers.24.mlp.experts.94.down_proj", "model.language_model.layers.24.mlp.experts.95.down_proj", "model.language_model.layers.24.mlp.experts.96.down_proj", "model.language_model.layers.24.mlp.experts.97.down_proj", "model.language_model.layers.24.mlp.experts.98.down_proj", "model.language_model.layers.24.mlp.experts.99.down_proj", "model.language_model.layers.24.mlp.experts.100.down_proj", "model.language_model.layers.24.mlp.experts.101.down_proj", "model.language_model.layers.24.mlp.experts.102.down_proj", "model.language_model.layers.24.mlp.experts.103.down_proj", "model.language_model.layers.24.mlp.experts.104.down_proj", "model.language_model.layers.24.mlp.experts.105.down_proj", "model.language_model.layers.24.mlp.experts.106.down_proj", "model.language_model.layers.24.mlp.experts.107.down_proj", "model.language_model.layers.24.mlp.experts.108.down_proj", "model.language_model.layers.24.mlp.experts.109.down_proj", "model.language_model.layers.24.mlp.experts.110.down_proj", "model.language_model.layers.24.mlp.experts.111.down_proj", "model.language_model.layers.24.mlp.experts.112.down_proj", "model.language_model.layers.24.mlp.experts.113.down_proj", "model.language_model.layers.24.mlp.experts.114.down_proj", "model.language_model.layers.24.mlp.experts.115.down_proj", "model.language_model.layers.24.mlp.experts.116.down_proj", "model.language_model.layers.24.mlp.experts.117.down_proj", "model.language_model.layers.24.mlp.experts.118.down_proj", "model.language_model.layers.24.mlp.experts.119.down_proj", "model.language_model.layers.24.mlp.experts.120.down_proj", "model.language_model.layers.24.mlp.experts.121.down_proj", "model.language_model.layers.24.mlp.experts.122.down_proj", "model.language_model.layers.24.mlp.experts.123.down_proj", "model.language_model.layers.24.mlp.experts.124.down_proj", "model.language_model.layers.24.mlp.experts.125.down_proj", "model.language_model.layers.24.mlp.experts.126.down_proj", "model.language_model.layers.24.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.001162506639957428, "dbits": 603979776 } ] }, { "idx": 50, "layers": [ "model.language_model.layers.25.self_attn.q_proj", "model.language_model.layers.25.self_attn.k_proj", "model.language_model.layers.25.self_attn.v_proj", "model.language_model.layers.25.self_attn.o_proj" ], "candidates": [ { "dkld": 0.001035251375287774, "dbits": 18874368 } ] }, { "idx": 51, "layers": [ "model.language_model.layers.25.mlp.experts.0.gate_proj", "model.language_model.layers.25.mlp.experts.1.gate_proj", "model.language_model.layers.25.mlp.experts.2.gate_proj", "model.language_model.layers.25.mlp.experts.3.gate_proj", "model.language_model.layers.25.mlp.experts.4.gate_proj", "model.language_model.layers.25.mlp.experts.5.gate_proj", "model.language_model.layers.25.mlp.experts.6.gate_proj", "model.language_model.layers.25.mlp.experts.7.gate_proj", "model.language_model.layers.25.mlp.experts.8.gate_proj", "model.language_model.layers.25.mlp.experts.9.gate_proj", "model.language_model.layers.25.mlp.experts.10.gate_proj", "model.language_model.layers.25.mlp.experts.11.gate_proj", "model.language_model.layers.25.mlp.experts.12.gate_proj", "model.language_model.layers.25.mlp.experts.13.gate_proj", "model.language_model.layers.25.mlp.experts.14.gate_proj", "model.language_model.layers.25.mlp.experts.15.gate_proj", "model.language_model.layers.25.mlp.experts.16.gate_proj", "model.language_model.layers.25.mlp.experts.17.gate_proj", "model.language_model.layers.25.mlp.experts.18.gate_proj", "model.language_model.layers.25.mlp.experts.19.gate_proj", "model.language_model.layers.25.mlp.experts.20.gate_proj", "model.language_model.layers.25.mlp.experts.21.gate_proj", "model.language_model.layers.25.mlp.experts.22.gate_proj", "model.language_model.layers.25.mlp.experts.23.gate_proj", "model.language_model.layers.25.mlp.experts.24.gate_proj", "model.language_model.layers.25.mlp.experts.25.gate_proj", "model.language_model.layers.25.mlp.experts.26.gate_proj", "model.language_model.layers.25.mlp.experts.27.gate_proj", "model.language_model.layers.25.mlp.experts.28.gate_proj", "model.language_model.layers.25.mlp.experts.29.gate_proj", "model.language_model.layers.25.mlp.experts.30.gate_proj", "model.language_model.layers.25.mlp.experts.31.gate_proj", "model.language_model.layers.25.mlp.experts.32.gate_proj", "model.language_model.layers.25.mlp.experts.33.gate_proj", "model.language_model.layers.25.mlp.experts.34.gate_proj", "model.language_model.layers.25.mlp.experts.35.gate_proj", "model.language_model.layers.25.mlp.experts.36.gate_proj", "model.language_model.layers.25.mlp.experts.37.gate_proj", "model.language_model.layers.25.mlp.experts.38.gate_proj", "model.language_model.layers.25.mlp.experts.39.gate_proj", "model.language_model.layers.25.mlp.experts.40.gate_proj", "model.language_model.layers.25.mlp.experts.41.gate_proj", "model.language_model.layers.25.mlp.experts.42.gate_proj", "model.language_model.layers.25.mlp.experts.43.gate_proj", "model.language_model.layers.25.mlp.experts.44.gate_proj", "model.language_model.layers.25.mlp.experts.45.gate_proj", "model.language_model.layers.25.mlp.experts.46.gate_proj", "model.language_model.layers.25.mlp.experts.47.gate_proj", "model.language_model.layers.25.mlp.experts.48.gate_proj", "model.language_model.layers.25.mlp.experts.49.gate_proj", "model.language_model.layers.25.mlp.experts.50.gate_proj", "model.language_model.layers.25.mlp.experts.51.gate_proj", "model.language_model.layers.25.mlp.experts.52.gate_proj", "model.language_model.layers.25.mlp.experts.53.gate_proj", "model.language_model.layers.25.mlp.experts.54.gate_proj", "model.language_model.layers.25.mlp.experts.55.gate_proj", "model.language_model.layers.25.mlp.experts.56.gate_proj", "model.language_model.layers.25.mlp.experts.57.gate_proj", "model.language_model.layers.25.mlp.experts.58.gate_proj", "model.language_model.layers.25.mlp.experts.59.gate_proj", "model.language_model.layers.25.mlp.experts.60.gate_proj", "model.language_model.layers.25.mlp.experts.61.gate_proj", "model.language_model.layers.25.mlp.experts.62.gate_proj", "model.language_model.layers.25.mlp.experts.63.gate_proj", "model.language_model.layers.25.mlp.experts.64.gate_proj", "model.language_model.layers.25.mlp.experts.65.gate_proj", "model.language_model.layers.25.mlp.experts.66.gate_proj", "model.language_model.layers.25.mlp.experts.67.gate_proj", "model.language_model.layers.25.mlp.experts.68.gate_proj", "model.language_model.layers.25.mlp.experts.69.gate_proj", "model.language_model.layers.25.mlp.experts.70.gate_proj", "model.language_model.layers.25.mlp.experts.71.gate_proj", "model.language_model.layers.25.mlp.experts.72.gate_proj", "model.language_model.layers.25.mlp.experts.73.gate_proj", "model.language_model.layers.25.mlp.experts.74.gate_proj", "model.language_model.layers.25.mlp.experts.75.gate_proj", "model.language_model.layers.25.mlp.experts.76.gate_proj", "model.language_model.layers.25.mlp.experts.77.gate_proj", "model.language_model.layers.25.mlp.experts.78.gate_proj", "model.language_model.layers.25.mlp.experts.79.gate_proj", "model.language_model.layers.25.mlp.experts.80.gate_proj", "model.language_model.layers.25.mlp.experts.81.gate_proj", "model.language_model.layers.25.mlp.experts.82.gate_proj", "model.language_model.layers.25.mlp.experts.83.gate_proj", "model.language_model.layers.25.mlp.experts.84.gate_proj", "model.language_model.layers.25.mlp.experts.85.gate_proj", "model.language_model.layers.25.mlp.experts.86.gate_proj", "model.language_model.layers.25.mlp.experts.87.gate_proj", "model.language_model.layers.25.mlp.experts.88.gate_proj", "model.language_model.layers.25.mlp.experts.89.gate_proj", "model.language_model.layers.25.mlp.experts.90.gate_proj", "model.language_model.layers.25.mlp.experts.91.gate_proj", "model.language_model.layers.25.mlp.experts.92.gate_proj", "model.language_model.layers.25.mlp.experts.93.gate_proj", "model.language_model.layers.25.mlp.experts.94.gate_proj", "model.language_model.layers.25.mlp.experts.95.gate_proj", "model.language_model.layers.25.mlp.experts.96.gate_proj", "model.language_model.layers.25.mlp.experts.97.gate_proj", "model.language_model.layers.25.mlp.experts.98.gate_proj", "model.language_model.layers.25.mlp.experts.99.gate_proj", "model.language_model.layers.25.mlp.experts.100.gate_proj", "model.language_model.layers.25.mlp.experts.101.gate_proj", "model.language_model.layers.25.mlp.experts.102.gate_proj", "model.language_model.layers.25.mlp.experts.103.gate_proj", "model.language_model.layers.25.mlp.experts.104.gate_proj", "model.language_model.layers.25.mlp.experts.105.gate_proj", "model.language_model.layers.25.mlp.experts.106.gate_proj", "model.language_model.layers.25.mlp.experts.107.gate_proj", "model.language_model.layers.25.mlp.experts.108.gate_proj", "model.language_model.layers.25.mlp.experts.109.gate_proj", "model.language_model.layers.25.mlp.experts.110.gate_proj", "model.language_model.layers.25.mlp.experts.111.gate_proj", "model.language_model.layers.25.mlp.experts.112.gate_proj", "model.language_model.layers.25.mlp.experts.113.gate_proj", "model.language_model.layers.25.mlp.experts.114.gate_proj", "model.language_model.layers.25.mlp.experts.115.gate_proj", "model.language_model.layers.25.mlp.experts.116.gate_proj", "model.language_model.layers.25.mlp.experts.117.gate_proj", "model.language_model.layers.25.mlp.experts.118.gate_proj", "model.language_model.layers.25.mlp.experts.119.gate_proj", "model.language_model.layers.25.mlp.experts.120.gate_proj", "model.language_model.layers.25.mlp.experts.121.gate_proj", "model.language_model.layers.25.mlp.experts.122.gate_proj", "model.language_model.layers.25.mlp.experts.123.gate_proj", "model.language_model.layers.25.mlp.experts.124.gate_proj", "model.language_model.layers.25.mlp.experts.125.gate_proj", "model.language_model.layers.25.mlp.experts.126.gate_proj", "model.language_model.layers.25.mlp.experts.127.gate_proj", "model.language_model.layers.25.mlp.experts.0.up_proj", "model.language_model.layers.25.mlp.experts.1.up_proj", "model.language_model.layers.25.mlp.experts.2.up_proj", "model.language_model.layers.25.mlp.experts.3.up_proj", "model.language_model.layers.25.mlp.experts.4.up_proj", "model.language_model.layers.25.mlp.experts.5.up_proj", "model.language_model.layers.25.mlp.experts.6.up_proj", "model.language_model.layers.25.mlp.experts.7.up_proj", "model.language_model.layers.25.mlp.experts.8.up_proj", "model.language_model.layers.25.mlp.experts.9.up_proj", "model.language_model.layers.25.mlp.experts.10.up_proj", "model.language_model.layers.25.mlp.experts.11.up_proj", "model.language_model.layers.25.mlp.experts.12.up_proj", "model.language_model.layers.25.mlp.experts.13.up_proj", "model.language_model.layers.25.mlp.experts.14.up_proj", "model.language_model.layers.25.mlp.experts.15.up_proj", "model.language_model.layers.25.mlp.experts.16.up_proj", "model.language_model.layers.25.mlp.experts.17.up_proj", "model.language_model.layers.25.mlp.experts.18.up_proj", "model.language_model.layers.25.mlp.experts.19.up_proj", "model.language_model.layers.25.mlp.experts.20.up_proj", "model.language_model.layers.25.mlp.experts.21.up_proj", "model.language_model.layers.25.mlp.experts.22.up_proj", "model.language_model.layers.25.mlp.experts.23.up_proj", "model.language_model.layers.25.mlp.experts.24.up_proj", "model.language_model.layers.25.mlp.experts.25.up_proj", "model.language_model.layers.25.mlp.experts.26.up_proj", "model.language_model.layers.25.mlp.experts.27.up_proj", "model.language_model.layers.25.mlp.experts.28.up_proj", "model.language_model.layers.25.mlp.experts.29.up_proj", "model.language_model.layers.25.mlp.experts.30.up_proj", "model.language_model.layers.25.mlp.experts.31.up_proj", "model.language_model.layers.25.mlp.experts.32.up_proj", "model.language_model.layers.25.mlp.experts.33.up_proj", "model.language_model.layers.25.mlp.experts.34.up_proj", "model.language_model.layers.25.mlp.experts.35.up_proj", "model.language_model.layers.25.mlp.experts.36.up_proj", "model.language_model.layers.25.mlp.experts.37.up_proj", "model.language_model.layers.25.mlp.experts.38.up_proj", "model.language_model.layers.25.mlp.experts.39.up_proj", "model.language_model.layers.25.mlp.experts.40.up_proj", "model.language_model.layers.25.mlp.experts.41.up_proj", "model.language_model.layers.25.mlp.experts.42.up_proj", "model.language_model.layers.25.mlp.experts.43.up_proj", "model.language_model.layers.25.mlp.experts.44.up_proj", "model.language_model.layers.25.mlp.experts.45.up_proj", "model.language_model.layers.25.mlp.experts.46.up_proj", "model.language_model.layers.25.mlp.experts.47.up_proj", "model.language_model.layers.25.mlp.experts.48.up_proj", "model.language_model.layers.25.mlp.experts.49.up_proj", "model.language_model.layers.25.mlp.experts.50.up_proj", "model.language_model.layers.25.mlp.experts.51.up_proj", "model.language_model.layers.25.mlp.experts.52.up_proj", "model.language_model.layers.25.mlp.experts.53.up_proj", "model.language_model.layers.25.mlp.experts.54.up_proj", "model.language_model.layers.25.mlp.experts.55.up_proj", "model.language_model.layers.25.mlp.experts.56.up_proj", "model.language_model.layers.25.mlp.experts.57.up_proj", "model.language_model.layers.25.mlp.experts.58.up_proj", "model.language_model.layers.25.mlp.experts.59.up_proj", "model.language_model.layers.25.mlp.experts.60.up_proj", "model.language_model.layers.25.mlp.experts.61.up_proj", "model.language_model.layers.25.mlp.experts.62.up_proj", "model.language_model.layers.25.mlp.experts.63.up_proj", "model.language_model.layers.25.mlp.experts.64.up_proj", "model.language_model.layers.25.mlp.experts.65.up_proj", "model.language_model.layers.25.mlp.experts.66.up_proj", "model.language_model.layers.25.mlp.experts.67.up_proj", "model.language_model.layers.25.mlp.experts.68.up_proj", "model.language_model.layers.25.mlp.experts.69.up_proj", "model.language_model.layers.25.mlp.experts.70.up_proj", "model.language_model.layers.25.mlp.experts.71.up_proj", "model.language_model.layers.25.mlp.experts.72.up_proj", "model.language_model.layers.25.mlp.experts.73.up_proj", "model.language_model.layers.25.mlp.experts.74.up_proj", "model.language_model.layers.25.mlp.experts.75.up_proj", "model.language_model.layers.25.mlp.experts.76.up_proj", "model.language_model.layers.25.mlp.experts.77.up_proj", "model.language_model.layers.25.mlp.experts.78.up_proj", "model.language_model.layers.25.mlp.experts.79.up_proj", "model.language_model.layers.25.mlp.experts.80.up_proj", "model.language_model.layers.25.mlp.experts.81.up_proj", "model.language_model.layers.25.mlp.experts.82.up_proj", "model.language_model.layers.25.mlp.experts.83.up_proj", "model.language_model.layers.25.mlp.experts.84.up_proj", "model.language_model.layers.25.mlp.experts.85.up_proj", "model.language_model.layers.25.mlp.experts.86.up_proj", "model.language_model.layers.25.mlp.experts.87.up_proj", "model.language_model.layers.25.mlp.experts.88.up_proj", "model.language_model.layers.25.mlp.experts.89.up_proj", "model.language_model.layers.25.mlp.experts.90.up_proj", "model.language_model.layers.25.mlp.experts.91.up_proj", "model.language_model.layers.25.mlp.experts.92.up_proj", "model.language_model.layers.25.mlp.experts.93.up_proj", "model.language_model.layers.25.mlp.experts.94.up_proj", "model.language_model.layers.25.mlp.experts.95.up_proj", "model.language_model.layers.25.mlp.experts.96.up_proj", "model.language_model.layers.25.mlp.experts.97.up_proj", "model.language_model.layers.25.mlp.experts.98.up_proj", "model.language_model.layers.25.mlp.experts.99.up_proj", "model.language_model.layers.25.mlp.experts.100.up_proj", "model.language_model.layers.25.mlp.experts.101.up_proj", "model.language_model.layers.25.mlp.experts.102.up_proj", "model.language_model.layers.25.mlp.experts.103.up_proj", "model.language_model.layers.25.mlp.experts.104.up_proj", "model.language_model.layers.25.mlp.experts.105.up_proj", "model.language_model.layers.25.mlp.experts.106.up_proj", "model.language_model.layers.25.mlp.experts.107.up_proj", "model.language_model.layers.25.mlp.experts.108.up_proj", "model.language_model.layers.25.mlp.experts.109.up_proj", "model.language_model.layers.25.mlp.experts.110.up_proj", "model.language_model.layers.25.mlp.experts.111.up_proj", "model.language_model.layers.25.mlp.experts.112.up_proj", "model.language_model.layers.25.mlp.experts.113.up_proj", "model.language_model.layers.25.mlp.experts.114.up_proj", "model.language_model.layers.25.mlp.experts.115.up_proj", "model.language_model.layers.25.mlp.experts.116.up_proj", "model.language_model.layers.25.mlp.experts.117.up_proj", "model.language_model.layers.25.mlp.experts.118.up_proj", "model.language_model.layers.25.mlp.experts.119.up_proj", "model.language_model.layers.25.mlp.experts.120.up_proj", "model.language_model.layers.25.mlp.experts.121.up_proj", "model.language_model.layers.25.mlp.experts.122.up_proj", "model.language_model.layers.25.mlp.experts.123.up_proj", "model.language_model.layers.25.mlp.experts.124.up_proj", "model.language_model.layers.25.mlp.experts.125.up_proj", "model.language_model.layers.25.mlp.experts.126.up_proj", "model.language_model.layers.25.mlp.experts.127.up_proj", "model.language_model.layers.25.mlp.experts.0.down_proj", "model.language_model.layers.25.mlp.experts.1.down_proj", "model.language_model.layers.25.mlp.experts.2.down_proj", "model.language_model.layers.25.mlp.experts.3.down_proj", "model.language_model.layers.25.mlp.experts.4.down_proj", "model.language_model.layers.25.mlp.experts.5.down_proj", "model.language_model.layers.25.mlp.experts.6.down_proj", "model.language_model.layers.25.mlp.experts.7.down_proj", "model.language_model.layers.25.mlp.experts.8.down_proj", "model.language_model.layers.25.mlp.experts.9.down_proj", "model.language_model.layers.25.mlp.experts.10.down_proj", "model.language_model.layers.25.mlp.experts.11.down_proj", "model.language_model.layers.25.mlp.experts.12.down_proj", "model.language_model.layers.25.mlp.experts.13.down_proj", "model.language_model.layers.25.mlp.experts.14.down_proj", "model.language_model.layers.25.mlp.experts.15.down_proj", "model.language_model.layers.25.mlp.experts.16.down_proj", "model.language_model.layers.25.mlp.experts.17.down_proj", "model.language_model.layers.25.mlp.experts.18.down_proj", "model.language_model.layers.25.mlp.experts.19.down_proj", "model.language_model.layers.25.mlp.experts.20.down_proj", "model.language_model.layers.25.mlp.experts.21.down_proj", "model.language_model.layers.25.mlp.experts.22.down_proj", "model.language_model.layers.25.mlp.experts.23.down_proj", "model.language_model.layers.25.mlp.experts.24.down_proj", "model.language_model.layers.25.mlp.experts.25.down_proj", "model.language_model.layers.25.mlp.experts.26.down_proj", "model.language_model.layers.25.mlp.experts.27.down_proj", "model.language_model.layers.25.mlp.experts.28.down_proj", "model.language_model.layers.25.mlp.experts.29.down_proj", "model.language_model.layers.25.mlp.experts.30.down_proj", "model.language_model.layers.25.mlp.experts.31.down_proj", "model.language_model.layers.25.mlp.experts.32.down_proj", "model.language_model.layers.25.mlp.experts.33.down_proj", "model.language_model.layers.25.mlp.experts.34.down_proj", "model.language_model.layers.25.mlp.experts.35.down_proj", "model.language_model.layers.25.mlp.experts.36.down_proj", "model.language_model.layers.25.mlp.experts.37.down_proj", "model.language_model.layers.25.mlp.experts.38.down_proj", "model.language_model.layers.25.mlp.experts.39.down_proj", "model.language_model.layers.25.mlp.experts.40.down_proj", "model.language_model.layers.25.mlp.experts.41.down_proj", "model.language_model.layers.25.mlp.experts.42.down_proj", "model.language_model.layers.25.mlp.experts.43.down_proj", "model.language_model.layers.25.mlp.experts.44.down_proj", "model.language_model.layers.25.mlp.experts.45.down_proj", "model.language_model.layers.25.mlp.experts.46.down_proj", "model.language_model.layers.25.mlp.experts.47.down_proj", "model.language_model.layers.25.mlp.experts.48.down_proj", "model.language_model.layers.25.mlp.experts.49.down_proj", "model.language_model.layers.25.mlp.experts.50.down_proj", "model.language_model.layers.25.mlp.experts.51.down_proj", "model.language_model.layers.25.mlp.experts.52.down_proj", "model.language_model.layers.25.mlp.experts.53.down_proj", "model.language_model.layers.25.mlp.experts.54.down_proj", "model.language_model.layers.25.mlp.experts.55.down_proj", "model.language_model.layers.25.mlp.experts.56.down_proj", "model.language_model.layers.25.mlp.experts.57.down_proj", "model.language_model.layers.25.mlp.experts.58.down_proj", "model.language_model.layers.25.mlp.experts.59.down_proj", "model.language_model.layers.25.mlp.experts.60.down_proj", "model.language_model.layers.25.mlp.experts.61.down_proj", "model.language_model.layers.25.mlp.experts.62.down_proj", "model.language_model.layers.25.mlp.experts.63.down_proj", "model.language_model.layers.25.mlp.experts.64.down_proj", "model.language_model.layers.25.mlp.experts.65.down_proj", "model.language_model.layers.25.mlp.experts.66.down_proj", "model.language_model.layers.25.mlp.experts.67.down_proj", "model.language_model.layers.25.mlp.experts.68.down_proj", "model.language_model.layers.25.mlp.experts.69.down_proj", "model.language_model.layers.25.mlp.experts.70.down_proj", "model.language_model.layers.25.mlp.experts.71.down_proj", "model.language_model.layers.25.mlp.experts.72.down_proj", "model.language_model.layers.25.mlp.experts.73.down_proj", "model.language_model.layers.25.mlp.experts.74.down_proj", "model.language_model.layers.25.mlp.experts.75.down_proj", "model.language_model.layers.25.mlp.experts.76.down_proj", "model.language_model.layers.25.mlp.experts.77.down_proj", "model.language_model.layers.25.mlp.experts.78.down_proj", "model.language_model.layers.25.mlp.experts.79.down_proj", "model.language_model.layers.25.mlp.experts.80.down_proj", "model.language_model.layers.25.mlp.experts.81.down_proj", "model.language_model.layers.25.mlp.experts.82.down_proj", "model.language_model.layers.25.mlp.experts.83.down_proj", "model.language_model.layers.25.mlp.experts.84.down_proj", "model.language_model.layers.25.mlp.experts.85.down_proj", "model.language_model.layers.25.mlp.experts.86.down_proj", "model.language_model.layers.25.mlp.experts.87.down_proj", "model.language_model.layers.25.mlp.experts.88.down_proj", "model.language_model.layers.25.mlp.experts.89.down_proj", "model.language_model.layers.25.mlp.experts.90.down_proj", "model.language_model.layers.25.mlp.experts.91.down_proj", "model.language_model.layers.25.mlp.experts.92.down_proj", "model.language_model.layers.25.mlp.experts.93.down_proj", "model.language_model.layers.25.mlp.experts.94.down_proj", "model.language_model.layers.25.mlp.experts.95.down_proj", "model.language_model.layers.25.mlp.experts.96.down_proj", "model.language_model.layers.25.mlp.experts.97.down_proj", "model.language_model.layers.25.mlp.experts.98.down_proj", "model.language_model.layers.25.mlp.experts.99.down_proj", "model.language_model.layers.25.mlp.experts.100.down_proj", "model.language_model.layers.25.mlp.experts.101.down_proj", "model.language_model.layers.25.mlp.experts.102.down_proj", "model.language_model.layers.25.mlp.experts.103.down_proj", "model.language_model.layers.25.mlp.experts.104.down_proj", "model.language_model.layers.25.mlp.experts.105.down_proj", "model.language_model.layers.25.mlp.experts.106.down_proj", "model.language_model.layers.25.mlp.experts.107.down_proj", "model.language_model.layers.25.mlp.experts.108.down_proj", "model.language_model.layers.25.mlp.experts.109.down_proj", "model.language_model.layers.25.mlp.experts.110.down_proj", "model.language_model.layers.25.mlp.experts.111.down_proj", "model.language_model.layers.25.mlp.experts.112.down_proj", "model.language_model.layers.25.mlp.experts.113.down_proj", "model.language_model.layers.25.mlp.experts.114.down_proj", "model.language_model.layers.25.mlp.experts.115.down_proj", "model.language_model.layers.25.mlp.experts.116.down_proj", "model.language_model.layers.25.mlp.experts.117.down_proj", "model.language_model.layers.25.mlp.experts.118.down_proj", "model.language_model.layers.25.mlp.experts.119.down_proj", "model.language_model.layers.25.mlp.experts.120.down_proj", "model.language_model.layers.25.mlp.experts.121.down_proj", "model.language_model.layers.25.mlp.experts.122.down_proj", "model.language_model.layers.25.mlp.experts.123.down_proj", "model.language_model.layers.25.mlp.experts.124.down_proj", "model.language_model.layers.25.mlp.experts.125.down_proj", "model.language_model.layers.25.mlp.experts.126.down_proj", "model.language_model.layers.25.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.0017370739951729774, "dbits": 603979776 } ] }, { "idx": 52, "layers": [ "model.language_model.layers.26.self_attn.q_proj", "model.language_model.layers.26.self_attn.k_proj", "model.language_model.layers.26.self_attn.v_proj", "model.language_model.layers.26.self_attn.o_proj" ], "candidates": [ { "dkld": 0.0009857217781245736, "dbits": 18874368 } ] }, { "idx": 53, "layers": [ "model.language_model.layers.26.mlp.experts.0.gate_proj", "model.language_model.layers.26.mlp.experts.1.gate_proj", "model.language_model.layers.26.mlp.experts.2.gate_proj", "model.language_model.layers.26.mlp.experts.3.gate_proj", "model.language_model.layers.26.mlp.experts.4.gate_proj", "model.language_model.layers.26.mlp.experts.5.gate_proj", "model.language_model.layers.26.mlp.experts.6.gate_proj", "model.language_model.layers.26.mlp.experts.7.gate_proj", "model.language_model.layers.26.mlp.experts.8.gate_proj", "model.language_model.layers.26.mlp.experts.9.gate_proj", "model.language_model.layers.26.mlp.experts.10.gate_proj", "model.language_model.layers.26.mlp.experts.11.gate_proj", "model.language_model.layers.26.mlp.experts.12.gate_proj", "model.language_model.layers.26.mlp.experts.13.gate_proj", "model.language_model.layers.26.mlp.experts.14.gate_proj", "model.language_model.layers.26.mlp.experts.15.gate_proj", "model.language_model.layers.26.mlp.experts.16.gate_proj", "model.language_model.layers.26.mlp.experts.17.gate_proj", "model.language_model.layers.26.mlp.experts.18.gate_proj", "model.language_model.layers.26.mlp.experts.19.gate_proj", "model.language_model.layers.26.mlp.experts.20.gate_proj", "model.language_model.layers.26.mlp.experts.21.gate_proj", "model.language_model.layers.26.mlp.experts.22.gate_proj", "model.language_model.layers.26.mlp.experts.23.gate_proj", "model.language_model.layers.26.mlp.experts.24.gate_proj", "model.language_model.layers.26.mlp.experts.25.gate_proj", "model.language_model.layers.26.mlp.experts.26.gate_proj", "model.language_model.layers.26.mlp.experts.27.gate_proj", "model.language_model.layers.26.mlp.experts.28.gate_proj", "model.language_model.layers.26.mlp.experts.29.gate_proj", "model.language_model.layers.26.mlp.experts.30.gate_proj", "model.language_model.layers.26.mlp.experts.31.gate_proj", "model.language_model.layers.26.mlp.experts.32.gate_proj", "model.language_model.layers.26.mlp.experts.33.gate_proj", "model.language_model.layers.26.mlp.experts.34.gate_proj", "model.language_model.layers.26.mlp.experts.35.gate_proj", "model.language_model.layers.26.mlp.experts.36.gate_proj", "model.language_model.layers.26.mlp.experts.37.gate_proj", "model.language_model.layers.26.mlp.experts.38.gate_proj", "model.language_model.layers.26.mlp.experts.39.gate_proj", "model.language_model.layers.26.mlp.experts.40.gate_proj", "model.language_model.layers.26.mlp.experts.41.gate_proj", "model.language_model.layers.26.mlp.experts.42.gate_proj", "model.language_model.layers.26.mlp.experts.43.gate_proj", "model.language_model.layers.26.mlp.experts.44.gate_proj", "model.language_model.layers.26.mlp.experts.45.gate_proj", "model.language_model.layers.26.mlp.experts.46.gate_proj", "model.language_model.layers.26.mlp.experts.47.gate_proj", "model.language_model.layers.26.mlp.experts.48.gate_proj", "model.language_model.layers.26.mlp.experts.49.gate_proj", "model.language_model.layers.26.mlp.experts.50.gate_proj", "model.language_model.layers.26.mlp.experts.51.gate_proj", "model.language_model.layers.26.mlp.experts.52.gate_proj", "model.language_model.layers.26.mlp.experts.53.gate_proj", "model.language_model.layers.26.mlp.experts.54.gate_proj", "model.language_model.layers.26.mlp.experts.55.gate_proj", "model.language_model.layers.26.mlp.experts.56.gate_proj", "model.language_model.layers.26.mlp.experts.57.gate_proj", "model.language_model.layers.26.mlp.experts.58.gate_proj", "model.language_model.layers.26.mlp.experts.59.gate_proj", "model.language_model.layers.26.mlp.experts.60.gate_proj", "model.language_model.layers.26.mlp.experts.61.gate_proj", "model.language_model.layers.26.mlp.experts.62.gate_proj", "model.language_model.layers.26.mlp.experts.63.gate_proj", "model.language_model.layers.26.mlp.experts.64.gate_proj", "model.language_model.layers.26.mlp.experts.65.gate_proj", "model.language_model.layers.26.mlp.experts.66.gate_proj", "model.language_model.layers.26.mlp.experts.67.gate_proj", "model.language_model.layers.26.mlp.experts.68.gate_proj", "model.language_model.layers.26.mlp.experts.69.gate_proj", "model.language_model.layers.26.mlp.experts.70.gate_proj", "model.language_model.layers.26.mlp.experts.71.gate_proj", "model.language_model.layers.26.mlp.experts.72.gate_proj", "model.language_model.layers.26.mlp.experts.73.gate_proj", "model.language_model.layers.26.mlp.experts.74.gate_proj", "model.language_model.layers.26.mlp.experts.75.gate_proj", "model.language_model.layers.26.mlp.experts.76.gate_proj", "model.language_model.layers.26.mlp.experts.77.gate_proj", "model.language_model.layers.26.mlp.experts.78.gate_proj", "model.language_model.layers.26.mlp.experts.79.gate_proj", "model.language_model.layers.26.mlp.experts.80.gate_proj", "model.language_model.layers.26.mlp.experts.81.gate_proj", "model.language_model.layers.26.mlp.experts.82.gate_proj", "model.language_model.layers.26.mlp.experts.83.gate_proj", "model.language_model.layers.26.mlp.experts.84.gate_proj", "model.language_model.layers.26.mlp.experts.85.gate_proj", "model.language_model.layers.26.mlp.experts.86.gate_proj", "model.language_model.layers.26.mlp.experts.87.gate_proj", "model.language_model.layers.26.mlp.experts.88.gate_proj", "model.language_model.layers.26.mlp.experts.89.gate_proj", "model.language_model.layers.26.mlp.experts.90.gate_proj", "model.language_model.layers.26.mlp.experts.91.gate_proj", "model.language_model.layers.26.mlp.experts.92.gate_proj", "model.language_model.layers.26.mlp.experts.93.gate_proj", "model.language_model.layers.26.mlp.experts.94.gate_proj", "model.language_model.layers.26.mlp.experts.95.gate_proj", "model.language_model.layers.26.mlp.experts.96.gate_proj", "model.language_model.layers.26.mlp.experts.97.gate_proj", "model.language_model.layers.26.mlp.experts.98.gate_proj", "model.language_model.layers.26.mlp.experts.99.gate_proj", "model.language_model.layers.26.mlp.experts.100.gate_proj", "model.language_model.layers.26.mlp.experts.101.gate_proj", "model.language_model.layers.26.mlp.experts.102.gate_proj", "model.language_model.layers.26.mlp.experts.103.gate_proj", "model.language_model.layers.26.mlp.experts.104.gate_proj", "model.language_model.layers.26.mlp.experts.105.gate_proj", "model.language_model.layers.26.mlp.experts.106.gate_proj", "model.language_model.layers.26.mlp.experts.107.gate_proj", "model.language_model.layers.26.mlp.experts.108.gate_proj", "model.language_model.layers.26.mlp.experts.109.gate_proj", "model.language_model.layers.26.mlp.experts.110.gate_proj", "model.language_model.layers.26.mlp.experts.111.gate_proj", "model.language_model.layers.26.mlp.experts.112.gate_proj", "model.language_model.layers.26.mlp.experts.113.gate_proj", "model.language_model.layers.26.mlp.experts.114.gate_proj", "model.language_model.layers.26.mlp.experts.115.gate_proj", "model.language_model.layers.26.mlp.experts.116.gate_proj", "model.language_model.layers.26.mlp.experts.117.gate_proj", "model.language_model.layers.26.mlp.experts.118.gate_proj", "model.language_model.layers.26.mlp.experts.119.gate_proj", "model.language_model.layers.26.mlp.experts.120.gate_proj", "model.language_model.layers.26.mlp.experts.121.gate_proj", "model.language_model.layers.26.mlp.experts.122.gate_proj", "model.language_model.layers.26.mlp.experts.123.gate_proj", "model.language_model.layers.26.mlp.experts.124.gate_proj", "model.language_model.layers.26.mlp.experts.125.gate_proj", "model.language_model.layers.26.mlp.experts.126.gate_proj", "model.language_model.layers.26.mlp.experts.127.gate_proj", "model.language_model.layers.26.mlp.experts.0.up_proj", "model.language_model.layers.26.mlp.experts.1.up_proj", "model.language_model.layers.26.mlp.experts.2.up_proj", "model.language_model.layers.26.mlp.experts.3.up_proj", "model.language_model.layers.26.mlp.experts.4.up_proj", "model.language_model.layers.26.mlp.experts.5.up_proj", "model.language_model.layers.26.mlp.experts.6.up_proj", "model.language_model.layers.26.mlp.experts.7.up_proj", "model.language_model.layers.26.mlp.experts.8.up_proj", "model.language_model.layers.26.mlp.experts.9.up_proj", "model.language_model.layers.26.mlp.experts.10.up_proj", "model.language_model.layers.26.mlp.experts.11.up_proj", "model.language_model.layers.26.mlp.experts.12.up_proj", "model.language_model.layers.26.mlp.experts.13.up_proj", "model.language_model.layers.26.mlp.experts.14.up_proj", "model.language_model.layers.26.mlp.experts.15.up_proj", "model.language_model.layers.26.mlp.experts.16.up_proj", "model.language_model.layers.26.mlp.experts.17.up_proj", "model.language_model.layers.26.mlp.experts.18.up_proj", "model.language_model.layers.26.mlp.experts.19.up_proj", "model.language_model.layers.26.mlp.experts.20.up_proj", "model.language_model.layers.26.mlp.experts.21.up_proj", "model.language_model.layers.26.mlp.experts.22.up_proj", "model.language_model.layers.26.mlp.experts.23.up_proj", "model.language_model.layers.26.mlp.experts.24.up_proj", "model.language_model.layers.26.mlp.experts.25.up_proj", "model.language_model.layers.26.mlp.experts.26.up_proj", "model.language_model.layers.26.mlp.experts.27.up_proj", "model.language_model.layers.26.mlp.experts.28.up_proj", "model.language_model.layers.26.mlp.experts.29.up_proj", "model.language_model.layers.26.mlp.experts.30.up_proj", "model.language_model.layers.26.mlp.experts.31.up_proj", "model.language_model.layers.26.mlp.experts.32.up_proj", "model.language_model.layers.26.mlp.experts.33.up_proj", "model.language_model.layers.26.mlp.experts.34.up_proj", "model.language_model.layers.26.mlp.experts.35.up_proj", "model.language_model.layers.26.mlp.experts.36.up_proj", "model.language_model.layers.26.mlp.experts.37.up_proj", "model.language_model.layers.26.mlp.experts.38.up_proj", "model.language_model.layers.26.mlp.experts.39.up_proj", "model.language_model.layers.26.mlp.experts.40.up_proj", "model.language_model.layers.26.mlp.experts.41.up_proj", "model.language_model.layers.26.mlp.experts.42.up_proj", "model.language_model.layers.26.mlp.experts.43.up_proj", "model.language_model.layers.26.mlp.experts.44.up_proj", "model.language_model.layers.26.mlp.experts.45.up_proj", "model.language_model.layers.26.mlp.experts.46.up_proj", "model.language_model.layers.26.mlp.experts.47.up_proj", "model.language_model.layers.26.mlp.experts.48.up_proj", "model.language_model.layers.26.mlp.experts.49.up_proj", "model.language_model.layers.26.mlp.experts.50.up_proj", "model.language_model.layers.26.mlp.experts.51.up_proj", "model.language_model.layers.26.mlp.experts.52.up_proj", "model.language_model.layers.26.mlp.experts.53.up_proj", "model.language_model.layers.26.mlp.experts.54.up_proj", "model.language_model.layers.26.mlp.experts.55.up_proj", "model.language_model.layers.26.mlp.experts.56.up_proj", "model.language_model.layers.26.mlp.experts.57.up_proj", "model.language_model.layers.26.mlp.experts.58.up_proj", "model.language_model.layers.26.mlp.experts.59.up_proj", "model.language_model.layers.26.mlp.experts.60.up_proj", "model.language_model.layers.26.mlp.experts.61.up_proj", "model.language_model.layers.26.mlp.experts.62.up_proj", "model.language_model.layers.26.mlp.experts.63.up_proj", "model.language_model.layers.26.mlp.experts.64.up_proj", "model.language_model.layers.26.mlp.experts.65.up_proj", "model.language_model.layers.26.mlp.experts.66.up_proj", "model.language_model.layers.26.mlp.experts.67.up_proj", "model.language_model.layers.26.mlp.experts.68.up_proj", "model.language_model.layers.26.mlp.experts.69.up_proj", "model.language_model.layers.26.mlp.experts.70.up_proj", "model.language_model.layers.26.mlp.experts.71.up_proj", "model.language_model.layers.26.mlp.experts.72.up_proj", "model.language_model.layers.26.mlp.experts.73.up_proj", "model.language_model.layers.26.mlp.experts.74.up_proj", "model.language_model.layers.26.mlp.experts.75.up_proj", "model.language_model.layers.26.mlp.experts.76.up_proj", "model.language_model.layers.26.mlp.experts.77.up_proj", "model.language_model.layers.26.mlp.experts.78.up_proj", "model.language_model.layers.26.mlp.experts.79.up_proj", "model.language_model.layers.26.mlp.experts.80.up_proj", "model.language_model.layers.26.mlp.experts.81.up_proj", "model.language_model.layers.26.mlp.experts.82.up_proj", "model.language_model.layers.26.mlp.experts.83.up_proj", "model.language_model.layers.26.mlp.experts.84.up_proj", "model.language_model.layers.26.mlp.experts.85.up_proj", "model.language_model.layers.26.mlp.experts.86.up_proj", "model.language_model.layers.26.mlp.experts.87.up_proj", "model.language_model.layers.26.mlp.experts.88.up_proj", "model.language_model.layers.26.mlp.experts.89.up_proj", "model.language_model.layers.26.mlp.experts.90.up_proj", "model.language_model.layers.26.mlp.experts.91.up_proj", "model.language_model.layers.26.mlp.experts.92.up_proj", "model.language_model.layers.26.mlp.experts.93.up_proj", "model.language_model.layers.26.mlp.experts.94.up_proj", "model.language_model.layers.26.mlp.experts.95.up_proj", "model.language_model.layers.26.mlp.experts.96.up_proj", "model.language_model.layers.26.mlp.experts.97.up_proj", "model.language_model.layers.26.mlp.experts.98.up_proj", "model.language_model.layers.26.mlp.experts.99.up_proj", "model.language_model.layers.26.mlp.experts.100.up_proj", "model.language_model.layers.26.mlp.experts.101.up_proj", "model.language_model.layers.26.mlp.experts.102.up_proj", "model.language_model.layers.26.mlp.experts.103.up_proj", "model.language_model.layers.26.mlp.experts.104.up_proj", "model.language_model.layers.26.mlp.experts.105.up_proj", "model.language_model.layers.26.mlp.experts.106.up_proj", "model.language_model.layers.26.mlp.experts.107.up_proj", "model.language_model.layers.26.mlp.experts.108.up_proj", "model.language_model.layers.26.mlp.experts.109.up_proj", "model.language_model.layers.26.mlp.experts.110.up_proj", "model.language_model.layers.26.mlp.experts.111.up_proj", "model.language_model.layers.26.mlp.experts.112.up_proj", "model.language_model.layers.26.mlp.experts.113.up_proj", "model.language_model.layers.26.mlp.experts.114.up_proj", "model.language_model.layers.26.mlp.experts.115.up_proj", "model.language_model.layers.26.mlp.experts.116.up_proj", "model.language_model.layers.26.mlp.experts.117.up_proj", "model.language_model.layers.26.mlp.experts.118.up_proj", "model.language_model.layers.26.mlp.experts.119.up_proj", "model.language_model.layers.26.mlp.experts.120.up_proj", "model.language_model.layers.26.mlp.experts.121.up_proj", "model.language_model.layers.26.mlp.experts.122.up_proj", "model.language_model.layers.26.mlp.experts.123.up_proj", "model.language_model.layers.26.mlp.experts.124.up_proj", "model.language_model.layers.26.mlp.experts.125.up_proj", "model.language_model.layers.26.mlp.experts.126.up_proj", "model.language_model.layers.26.mlp.experts.127.up_proj", "model.language_model.layers.26.mlp.experts.0.down_proj", "model.language_model.layers.26.mlp.experts.1.down_proj", "model.language_model.layers.26.mlp.experts.2.down_proj", "model.language_model.layers.26.mlp.experts.3.down_proj", "model.language_model.layers.26.mlp.experts.4.down_proj", "model.language_model.layers.26.mlp.experts.5.down_proj", "model.language_model.layers.26.mlp.experts.6.down_proj", "model.language_model.layers.26.mlp.experts.7.down_proj", "model.language_model.layers.26.mlp.experts.8.down_proj", "model.language_model.layers.26.mlp.experts.9.down_proj", "model.language_model.layers.26.mlp.experts.10.down_proj", "model.language_model.layers.26.mlp.experts.11.down_proj", "model.language_model.layers.26.mlp.experts.12.down_proj", "model.language_model.layers.26.mlp.experts.13.down_proj", "model.language_model.layers.26.mlp.experts.14.down_proj", "model.language_model.layers.26.mlp.experts.15.down_proj", "model.language_model.layers.26.mlp.experts.16.down_proj", "model.language_model.layers.26.mlp.experts.17.down_proj", "model.language_model.layers.26.mlp.experts.18.down_proj", "model.language_model.layers.26.mlp.experts.19.down_proj", "model.language_model.layers.26.mlp.experts.20.down_proj", "model.language_model.layers.26.mlp.experts.21.down_proj", "model.language_model.layers.26.mlp.experts.22.down_proj", "model.language_model.layers.26.mlp.experts.23.down_proj", "model.language_model.layers.26.mlp.experts.24.down_proj", "model.language_model.layers.26.mlp.experts.25.down_proj", "model.language_model.layers.26.mlp.experts.26.down_proj", "model.language_model.layers.26.mlp.experts.27.down_proj", "model.language_model.layers.26.mlp.experts.28.down_proj", "model.language_model.layers.26.mlp.experts.29.down_proj", "model.language_model.layers.26.mlp.experts.30.down_proj", "model.language_model.layers.26.mlp.experts.31.down_proj", "model.language_model.layers.26.mlp.experts.32.down_proj", "model.language_model.layers.26.mlp.experts.33.down_proj", "model.language_model.layers.26.mlp.experts.34.down_proj", "model.language_model.layers.26.mlp.experts.35.down_proj", "model.language_model.layers.26.mlp.experts.36.down_proj", "model.language_model.layers.26.mlp.experts.37.down_proj", "model.language_model.layers.26.mlp.experts.38.down_proj", "model.language_model.layers.26.mlp.experts.39.down_proj", "model.language_model.layers.26.mlp.experts.40.down_proj", "model.language_model.layers.26.mlp.experts.41.down_proj", "model.language_model.layers.26.mlp.experts.42.down_proj", "model.language_model.layers.26.mlp.experts.43.down_proj", "model.language_model.layers.26.mlp.experts.44.down_proj", "model.language_model.layers.26.mlp.experts.45.down_proj", "model.language_model.layers.26.mlp.experts.46.down_proj", "model.language_model.layers.26.mlp.experts.47.down_proj", "model.language_model.layers.26.mlp.experts.48.down_proj", "model.language_model.layers.26.mlp.experts.49.down_proj", "model.language_model.layers.26.mlp.experts.50.down_proj", "model.language_model.layers.26.mlp.experts.51.down_proj", "model.language_model.layers.26.mlp.experts.52.down_proj", "model.language_model.layers.26.mlp.experts.53.down_proj", "model.language_model.layers.26.mlp.experts.54.down_proj", "model.language_model.layers.26.mlp.experts.55.down_proj", "model.language_model.layers.26.mlp.experts.56.down_proj", "model.language_model.layers.26.mlp.experts.57.down_proj", "model.language_model.layers.26.mlp.experts.58.down_proj", "model.language_model.layers.26.mlp.experts.59.down_proj", "model.language_model.layers.26.mlp.experts.60.down_proj", "model.language_model.layers.26.mlp.experts.61.down_proj", "model.language_model.layers.26.mlp.experts.62.down_proj", "model.language_model.layers.26.mlp.experts.63.down_proj", "model.language_model.layers.26.mlp.experts.64.down_proj", "model.language_model.layers.26.mlp.experts.65.down_proj", "model.language_model.layers.26.mlp.experts.66.down_proj", "model.language_model.layers.26.mlp.experts.67.down_proj", "model.language_model.layers.26.mlp.experts.68.down_proj", "model.language_model.layers.26.mlp.experts.69.down_proj", "model.language_model.layers.26.mlp.experts.70.down_proj", "model.language_model.layers.26.mlp.experts.71.down_proj", "model.language_model.layers.26.mlp.experts.72.down_proj", "model.language_model.layers.26.mlp.experts.73.down_proj", "model.language_model.layers.26.mlp.experts.74.down_proj", "model.language_model.layers.26.mlp.experts.75.down_proj", "model.language_model.layers.26.mlp.experts.76.down_proj", "model.language_model.layers.26.mlp.experts.77.down_proj", "model.language_model.layers.26.mlp.experts.78.down_proj", "model.language_model.layers.26.mlp.experts.79.down_proj", "model.language_model.layers.26.mlp.experts.80.down_proj", "model.language_model.layers.26.mlp.experts.81.down_proj", "model.language_model.layers.26.mlp.experts.82.down_proj", "model.language_model.layers.26.mlp.experts.83.down_proj", "model.language_model.layers.26.mlp.experts.84.down_proj", "model.language_model.layers.26.mlp.experts.85.down_proj", "model.language_model.layers.26.mlp.experts.86.down_proj", "model.language_model.layers.26.mlp.experts.87.down_proj", "model.language_model.layers.26.mlp.experts.88.down_proj", "model.language_model.layers.26.mlp.experts.89.down_proj", "model.language_model.layers.26.mlp.experts.90.down_proj", "model.language_model.layers.26.mlp.experts.91.down_proj", "model.language_model.layers.26.mlp.experts.92.down_proj", "model.language_model.layers.26.mlp.experts.93.down_proj", "model.language_model.layers.26.mlp.experts.94.down_proj", "model.language_model.layers.26.mlp.experts.95.down_proj", "model.language_model.layers.26.mlp.experts.96.down_proj", "model.language_model.layers.26.mlp.experts.97.down_proj", "model.language_model.layers.26.mlp.experts.98.down_proj", "model.language_model.layers.26.mlp.experts.99.down_proj", "model.language_model.layers.26.mlp.experts.100.down_proj", "model.language_model.layers.26.mlp.experts.101.down_proj", "model.language_model.layers.26.mlp.experts.102.down_proj", "model.language_model.layers.26.mlp.experts.103.down_proj", "model.language_model.layers.26.mlp.experts.104.down_proj", "model.language_model.layers.26.mlp.experts.105.down_proj", "model.language_model.layers.26.mlp.experts.106.down_proj", "model.language_model.layers.26.mlp.experts.107.down_proj", "model.language_model.layers.26.mlp.experts.108.down_proj", "model.language_model.layers.26.mlp.experts.109.down_proj", "model.language_model.layers.26.mlp.experts.110.down_proj", "model.language_model.layers.26.mlp.experts.111.down_proj", "model.language_model.layers.26.mlp.experts.112.down_proj", "model.language_model.layers.26.mlp.experts.113.down_proj", "model.language_model.layers.26.mlp.experts.114.down_proj", "model.language_model.layers.26.mlp.experts.115.down_proj", "model.language_model.layers.26.mlp.experts.116.down_proj", "model.language_model.layers.26.mlp.experts.117.down_proj", "model.language_model.layers.26.mlp.experts.118.down_proj", "model.language_model.layers.26.mlp.experts.119.down_proj", "model.language_model.layers.26.mlp.experts.120.down_proj", "model.language_model.layers.26.mlp.experts.121.down_proj", "model.language_model.layers.26.mlp.experts.122.down_proj", "model.language_model.layers.26.mlp.experts.123.down_proj", "model.language_model.layers.26.mlp.experts.124.down_proj", "model.language_model.layers.26.mlp.experts.125.down_proj", "model.language_model.layers.26.mlp.experts.126.down_proj", "model.language_model.layers.26.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.0007003538310527829, "dbits": 603979776 } ] }, { "idx": 54, "layers": [ "model.language_model.layers.27.self_attn.q_proj", "model.language_model.layers.27.self_attn.k_proj", "model.language_model.layers.27.self_attn.v_proj", "model.language_model.layers.27.self_attn.o_proj" ], "candidates": [ { "dkld": -0.00022886227816343307, "dbits": 18874368 } ] }, { "idx": 55, "layers": [ "model.language_model.layers.27.mlp.experts.0.gate_proj", "model.language_model.layers.27.mlp.experts.1.gate_proj", "model.language_model.layers.27.mlp.experts.2.gate_proj", "model.language_model.layers.27.mlp.experts.3.gate_proj", "model.language_model.layers.27.mlp.experts.4.gate_proj", "model.language_model.layers.27.mlp.experts.5.gate_proj", "model.language_model.layers.27.mlp.experts.6.gate_proj", "model.language_model.layers.27.mlp.experts.7.gate_proj", "model.language_model.layers.27.mlp.experts.8.gate_proj", "model.language_model.layers.27.mlp.experts.9.gate_proj", "model.language_model.layers.27.mlp.experts.10.gate_proj", "model.language_model.layers.27.mlp.experts.11.gate_proj", "model.language_model.layers.27.mlp.experts.12.gate_proj", "model.language_model.layers.27.mlp.experts.13.gate_proj", "model.language_model.layers.27.mlp.experts.14.gate_proj", "model.language_model.layers.27.mlp.experts.15.gate_proj", "model.language_model.layers.27.mlp.experts.16.gate_proj", "model.language_model.layers.27.mlp.experts.17.gate_proj", "model.language_model.layers.27.mlp.experts.18.gate_proj", "model.language_model.layers.27.mlp.experts.19.gate_proj", "model.language_model.layers.27.mlp.experts.20.gate_proj", "model.language_model.layers.27.mlp.experts.21.gate_proj", "model.language_model.layers.27.mlp.experts.22.gate_proj", "model.language_model.layers.27.mlp.experts.23.gate_proj", "model.language_model.layers.27.mlp.experts.24.gate_proj", "model.language_model.layers.27.mlp.experts.25.gate_proj", "model.language_model.layers.27.mlp.experts.26.gate_proj", "model.language_model.layers.27.mlp.experts.27.gate_proj", "model.language_model.layers.27.mlp.experts.28.gate_proj", "model.language_model.layers.27.mlp.experts.29.gate_proj", "model.language_model.layers.27.mlp.experts.30.gate_proj", "model.language_model.layers.27.mlp.experts.31.gate_proj", "model.language_model.layers.27.mlp.experts.32.gate_proj", "model.language_model.layers.27.mlp.experts.33.gate_proj", "model.language_model.layers.27.mlp.experts.34.gate_proj", "model.language_model.layers.27.mlp.experts.35.gate_proj", "model.language_model.layers.27.mlp.experts.36.gate_proj", "model.language_model.layers.27.mlp.experts.37.gate_proj", "model.language_model.layers.27.mlp.experts.38.gate_proj", "model.language_model.layers.27.mlp.experts.39.gate_proj", "model.language_model.layers.27.mlp.experts.40.gate_proj", "model.language_model.layers.27.mlp.experts.41.gate_proj", "model.language_model.layers.27.mlp.experts.42.gate_proj", "model.language_model.layers.27.mlp.experts.43.gate_proj", "model.language_model.layers.27.mlp.experts.44.gate_proj", "model.language_model.layers.27.mlp.experts.45.gate_proj", "model.language_model.layers.27.mlp.experts.46.gate_proj", "model.language_model.layers.27.mlp.experts.47.gate_proj", "model.language_model.layers.27.mlp.experts.48.gate_proj", "model.language_model.layers.27.mlp.experts.49.gate_proj", "model.language_model.layers.27.mlp.experts.50.gate_proj", "model.language_model.layers.27.mlp.experts.51.gate_proj", "model.language_model.layers.27.mlp.experts.52.gate_proj", "model.language_model.layers.27.mlp.experts.53.gate_proj", "model.language_model.layers.27.mlp.experts.54.gate_proj", "model.language_model.layers.27.mlp.experts.55.gate_proj", "model.language_model.layers.27.mlp.experts.56.gate_proj", "model.language_model.layers.27.mlp.experts.57.gate_proj", "model.language_model.layers.27.mlp.experts.58.gate_proj", "model.language_model.layers.27.mlp.experts.59.gate_proj", "model.language_model.layers.27.mlp.experts.60.gate_proj", "model.language_model.layers.27.mlp.experts.61.gate_proj", "model.language_model.layers.27.mlp.experts.62.gate_proj", "model.language_model.layers.27.mlp.experts.63.gate_proj", "model.language_model.layers.27.mlp.experts.64.gate_proj", "model.language_model.layers.27.mlp.experts.65.gate_proj", "model.language_model.layers.27.mlp.experts.66.gate_proj", "model.language_model.layers.27.mlp.experts.67.gate_proj", "model.language_model.layers.27.mlp.experts.68.gate_proj", "model.language_model.layers.27.mlp.experts.69.gate_proj", "model.language_model.layers.27.mlp.experts.70.gate_proj", "model.language_model.layers.27.mlp.experts.71.gate_proj", "model.language_model.layers.27.mlp.experts.72.gate_proj", "model.language_model.layers.27.mlp.experts.73.gate_proj", "model.language_model.layers.27.mlp.experts.74.gate_proj", "model.language_model.layers.27.mlp.experts.75.gate_proj", "model.language_model.layers.27.mlp.experts.76.gate_proj", "model.language_model.layers.27.mlp.experts.77.gate_proj", "model.language_model.layers.27.mlp.experts.78.gate_proj", "model.language_model.layers.27.mlp.experts.79.gate_proj", "model.language_model.layers.27.mlp.experts.80.gate_proj", "model.language_model.layers.27.mlp.experts.81.gate_proj", "model.language_model.layers.27.mlp.experts.82.gate_proj", "model.language_model.layers.27.mlp.experts.83.gate_proj", "model.language_model.layers.27.mlp.experts.84.gate_proj", "model.language_model.layers.27.mlp.experts.85.gate_proj", "model.language_model.layers.27.mlp.experts.86.gate_proj", "model.language_model.layers.27.mlp.experts.87.gate_proj", "model.language_model.layers.27.mlp.experts.88.gate_proj", "model.language_model.layers.27.mlp.experts.89.gate_proj", "model.language_model.layers.27.mlp.experts.90.gate_proj", "model.language_model.layers.27.mlp.experts.91.gate_proj", "model.language_model.layers.27.mlp.experts.92.gate_proj", "model.language_model.layers.27.mlp.experts.93.gate_proj", "model.language_model.layers.27.mlp.experts.94.gate_proj", "model.language_model.layers.27.mlp.experts.95.gate_proj", "model.language_model.layers.27.mlp.experts.96.gate_proj", "model.language_model.layers.27.mlp.experts.97.gate_proj", "model.language_model.layers.27.mlp.experts.98.gate_proj", "model.language_model.layers.27.mlp.experts.99.gate_proj", "model.language_model.layers.27.mlp.experts.100.gate_proj", "model.language_model.layers.27.mlp.experts.101.gate_proj", "model.language_model.layers.27.mlp.experts.102.gate_proj", "model.language_model.layers.27.mlp.experts.103.gate_proj", "model.language_model.layers.27.mlp.experts.104.gate_proj", "model.language_model.layers.27.mlp.experts.105.gate_proj", "model.language_model.layers.27.mlp.experts.106.gate_proj", "model.language_model.layers.27.mlp.experts.107.gate_proj", "model.language_model.layers.27.mlp.experts.108.gate_proj", "model.language_model.layers.27.mlp.experts.109.gate_proj", "model.language_model.layers.27.mlp.experts.110.gate_proj", "model.language_model.layers.27.mlp.experts.111.gate_proj", "model.language_model.layers.27.mlp.experts.112.gate_proj", "model.language_model.layers.27.mlp.experts.113.gate_proj", "model.language_model.layers.27.mlp.experts.114.gate_proj", "model.language_model.layers.27.mlp.experts.115.gate_proj", "model.language_model.layers.27.mlp.experts.116.gate_proj", "model.language_model.layers.27.mlp.experts.117.gate_proj", "model.language_model.layers.27.mlp.experts.118.gate_proj", "model.language_model.layers.27.mlp.experts.119.gate_proj", "model.language_model.layers.27.mlp.experts.120.gate_proj", "model.language_model.layers.27.mlp.experts.121.gate_proj", "model.language_model.layers.27.mlp.experts.122.gate_proj", "model.language_model.layers.27.mlp.experts.123.gate_proj", "model.language_model.layers.27.mlp.experts.124.gate_proj", "model.language_model.layers.27.mlp.experts.125.gate_proj", "model.language_model.layers.27.mlp.experts.126.gate_proj", "model.language_model.layers.27.mlp.experts.127.gate_proj", "model.language_model.layers.27.mlp.experts.0.up_proj", "model.language_model.layers.27.mlp.experts.1.up_proj", "model.language_model.layers.27.mlp.experts.2.up_proj", "model.language_model.layers.27.mlp.experts.3.up_proj", "model.language_model.layers.27.mlp.experts.4.up_proj", "model.language_model.layers.27.mlp.experts.5.up_proj", "model.language_model.layers.27.mlp.experts.6.up_proj", "model.language_model.layers.27.mlp.experts.7.up_proj", "model.language_model.layers.27.mlp.experts.8.up_proj", "model.language_model.layers.27.mlp.experts.9.up_proj", "model.language_model.layers.27.mlp.experts.10.up_proj", "model.language_model.layers.27.mlp.experts.11.up_proj", "model.language_model.layers.27.mlp.experts.12.up_proj", "model.language_model.layers.27.mlp.experts.13.up_proj", "model.language_model.layers.27.mlp.experts.14.up_proj", "model.language_model.layers.27.mlp.experts.15.up_proj", "model.language_model.layers.27.mlp.experts.16.up_proj", "model.language_model.layers.27.mlp.experts.17.up_proj", "model.language_model.layers.27.mlp.experts.18.up_proj", "model.language_model.layers.27.mlp.experts.19.up_proj", "model.language_model.layers.27.mlp.experts.20.up_proj", "model.language_model.layers.27.mlp.experts.21.up_proj", "model.language_model.layers.27.mlp.experts.22.up_proj", "model.language_model.layers.27.mlp.experts.23.up_proj", "model.language_model.layers.27.mlp.experts.24.up_proj", "model.language_model.layers.27.mlp.experts.25.up_proj", "model.language_model.layers.27.mlp.experts.26.up_proj", "model.language_model.layers.27.mlp.experts.27.up_proj", "model.language_model.layers.27.mlp.experts.28.up_proj", "model.language_model.layers.27.mlp.experts.29.up_proj", "model.language_model.layers.27.mlp.experts.30.up_proj", "model.language_model.layers.27.mlp.experts.31.up_proj", "model.language_model.layers.27.mlp.experts.32.up_proj", "model.language_model.layers.27.mlp.experts.33.up_proj", "model.language_model.layers.27.mlp.experts.34.up_proj", "model.language_model.layers.27.mlp.experts.35.up_proj", "model.language_model.layers.27.mlp.experts.36.up_proj", "model.language_model.layers.27.mlp.experts.37.up_proj", "model.language_model.layers.27.mlp.experts.38.up_proj", "model.language_model.layers.27.mlp.experts.39.up_proj", "model.language_model.layers.27.mlp.experts.40.up_proj", "model.language_model.layers.27.mlp.experts.41.up_proj", "model.language_model.layers.27.mlp.experts.42.up_proj", "model.language_model.layers.27.mlp.experts.43.up_proj", "model.language_model.layers.27.mlp.experts.44.up_proj", "model.language_model.layers.27.mlp.experts.45.up_proj", "model.language_model.layers.27.mlp.experts.46.up_proj", "model.language_model.layers.27.mlp.experts.47.up_proj", "model.language_model.layers.27.mlp.experts.48.up_proj", "model.language_model.layers.27.mlp.experts.49.up_proj", "model.language_model.layers.27.mlp.experts.50.up_proj", "model.language_model.layers.27.mlp.experts.51.up_proj", "model.language_model.layers.27.mlp.experts.52.up_proj", "model.language_model.layers.27.mlp.experts.53.up_proj", "model.language_model.layers.27.mlp.experts.54.up_proj", "model.language_model.layers.27.mlp.experts.55.up_proj", "model.language_model.layers.27.mlp.experts.56.up_proj", "model.language_model.layers.27.mlp.experts.57.up_proj", "model.language_model.layers.27.mlp.experts.58.up_proj", "model.language_model.layers.27.mlp.experts.59.up_proj", "model.language_model.layers.27.mlp.experts.60.up_proj", "model.language_model.layers.27.mlp.experts.61.up_proj", "model.language_model.layers.27.mlp.experts.62.up_proj", "model.language_model.layers.27.mlp.experts.63.up_proj", "model.language_model.layers.27.mlp.experts.64.up_proj", "model.language_model.layers.27.mlp.experts.65.up_proj", "model.language_model.layers.27.mlp.experts.66.up_proj", "model.language_model.layers.27.mlp.experts.67.up_proj", "model.language_model.layers.27.mlp.experts.68.up_proj", "model.language_model.layers.27.mlp.experts.69.up_proj", "model.language_model.layers.27.mlp.experts.70.up_proj", "model.language_model.layers.27.mlp.experts.71.up_proj", "model.language_model.layers.27.mlp.experts.72.up_proj", "model.language_model.layers.27.mlp.experts.73.up_proj", "model.language_model.layers.27.mlp.experts.74.up_proj", "model.language_model.layers.27.mlp.experts.75.up_proj", "model.language_model.layers.27.mlp.experts.76.up_proj", "model.language_model.layers.27.mlp.experts.77.up_proj", "model.language_model.layers.27.mlp.experts.78.up_proj", "model.language_model.layers.27.mlp.experts.79.up_proj", "model.language_model.layers.27.mlp.experts.80.up_proj", "model.language_model.layers.27.mlp.experts.81.up_proj", "model.language_model.layers.27.mlp.experts.82.up_proj", "model.language_model.layers.27.mlp.experts.83.up_proj", "model.language_model.layers.27.mlp.experts.84.up_proj", "model.language_model.layers.27.mlp.experts.85.up_proj", "model.language_model.layers.27.mlp.experts.86.up_proj", "model.language_model.layers.27.mlp.experts.87.up_proj", "model.language_model.layers.27.mlp.experts.88.up_proj", "model.language_model.layers.27.mlp.experts.89.up_proj", "model.language_model.layers.27.mlp.experts.90.up_proj", "model.language_model.layers.27.mlp.experts.91.up_proj", "model.language_model.layers.27.mlp.experts.92.up_proj", "model.language_model.layers.27.mlp.experts.93.up_proj", "model.language_model.layers.27.mlp.experts.94.up_proj", "model.language_model.layers.27.mlp.experts.95.up_proj", "model.language_model.layers.27.mlp.experts.96.up_proj", "model.language_model.layers.27.mlp.experts.97.up_proj", "model.language_model.layers.27.mlp.experts.98.up_proj", "model.language_model.layers.27.mlp.experts.99.up_proj", "model.language_model.layers.27.mlp.experts.100.up_proj", "model.language_model.layers.27.mlp.experts.101.up_proj", "model.language_model.layers.27.mlp.experts.102.up_proj", "model.language_model.layers.27.mlp.experts.103.up_proj", "model.language_model.layers.27.mlp.experts.104.up_proj", "model.language_model.layers.27.mlp.experts.105.up_proj", "model.language_model.layers.27.mlp.experts.106.up_proj", "model.language_model.layers.27.mlp.experts.107.up_proj", "model.language_model.layers.27.mlp.experts.108.up_proj", "model.language_model.layers.27.mlp.experts.109.up_proj", "model.language_model.layers.27.mlp.experts.110.up_proj", "model.language_model.layers.27.mlp.experts.111.up_proj", "model.language_model.layers.27.mlp.experts.112.up_proj", "model.language_model.layers.27.mlp.experts.113.up_proj", "model.language_model.layers.27.mlp.experts.114.up_proj", "model.language_model.layers.27.mlp.experts.115.up_proj", "model.language_model.layers.27.mlp.experts.116.up_proj", "model.language_model.layers.27.mlp.experts.117.up_proj", "model.language_model.layers.27.mlp.experts.118.up_proj", "model.language_model.layers.27.mlp.experts.119.up_proj", "model.language_model.layers.27.mlp.experts.120.up_proj", "model.language_model.layers.27.mlp.experts.121.up_proj", "model.language_model.layers.27.mlp.experts.122.up_proj", "model.language_model.layers.27.mlp.experts.123.up_proj", "model.language_model.layers.27.mlp.experts.124.up_proj", "model.language_model.layers.27.mlp.experts.125.up_proj", "model.language_model.layers.27.mlp.experts.126.up_proj", "model.language_model.layers.27.mlp.experts.127.up_proj", "model.language_model.layers.27.mlp.experts.0.down_proj", "model.language_model.layers.27.mlp.experts.1.down_proj", "model.language_model.layers.27.mlp.experts.2.down_proj", "model.language_model.layers.27.mlp.experts.3.down_proj", "model.language_model.layers.27.mlp.experts.4.down_proj", "model.language_model.layers.27.mlp.experts.5.down_proj", "model.language_model.layers.27.mlp.experts.6.down_proj", "model.language_model.layers.27.mlp.experts.7.down_proj", "model.language_model.layers.27.mlp.experts.8.down_proj", "model.language_model.layers.27.mlp.experts.9.down_proj", "model.language_model.layers.27.mlp.experts.10.down_proj", "model.language_model.layers.27.mlp.experts.11.down_proj", "model.language_model.layers.27.mlp.experts.12.down_proj", "model.language_model.layers.27.mlp.experts.13.down_proj", "model.language_model.layers.27.mlp.experts.14.down_proj", "model.language_model.layers.27.mlp.experts.15.down_proj", "model.language_model.layers.27.mlp.experts.16.down_proj", "model.language_model.layers.27.mlp.experts.17.down_proj", "model.language_model.layers.27.mlp.experts.18.down_proj", "model.language_model.layers.27.mlp.experts.19.down_proj", "model.language_model.layers.27.mlp.experts.20.down_proj", "model.language_model.layers.27.mlp.experts.21.down_proj", "model.language_model.layers.27.mlp.experts.22.down_proj", "model.language_model.layers.27.mlp.experts.23.down_proj", "model.language_model.layers.27.mlp.experts.24.down_proj", "model.language_model.layers.27.mlp.experts.25.down_proj", "model.language_model.layers.27.mlp.experts.26.down_proj", "model.language_model.layers.27.mlp.experts.27.down_proj", "model.language_model.layers.27.mlp.experts.28.down_proj", "model.language_model.layers.27.mlp.experts.29.down_proj", "model.language_model.layers.27.mlp.experts.30.down_proj", "model.language_model.layers.27.mlp.experts.31.down_proj", "model.language_model.layers.27.mlp.experts.32.down_proj", "model.language_model.layers.27.mlp.experts.33.down_proj", "model.language_model.layers.27.mlp.experts.34.down_proj", "model.language_model.layers.27.mlp.experts.35.down_proj", "model.language_model.layers.27.mlp.experts.36.down_proj", "model.language_model.layers.27.mlp.experts.37.down_proj", "model.language_model.layers.27.mlp.experts.38.down_proj", "model.language_model.layers.27.mlp.experts.39.down_proj", "model.language_model.layers.27.mlp.experts.40.down_proj", "model.language_model.layers.27.mlp.experts.41.down_proj", "model.language_model.layers.27.mlp.experts.42.down_proj", "model.language_model.layers.27.mlp.experts.43.down_proj", "model.language_model.layers.27.mlp.experts.44.down_proj", "model.language_model.layers.27.mlp.experts.45.down_proj", "model.language_model.layers.27.mlp.experts.46.down_proj", "model.language_model.layers.27.mlp.experts.47.down_proj", "model.language_model.layers.27.mlp.experts.48.down_proj", "model.language_model.layers.27.mlp.experts.49.down_proj", "model.language_model.layers.27.mlp.experts.50.down_proj", "model.language_model.layers.27.mlp.experts.51.down_proj", "model.language_model.layers.27.mlp.experts.52.down_proj", "model.language_model.layers.27.mlp.experts.53.down_proj", "model.language_model.layers.27.mlp.experts.54.down_proj", "model.language_model.layers.27.mlp.experts.55.down_proj", "model.language_model.layers.27.mlp.experts.56.down_proj", "model.language_model.layers.27.mlp.experts.57.down_proj", "model.language_model.layers.27.mlp.experts.58.down_proj", "model.language_model.layers.27.mlp.experts.59.down_proj", "model.language_model.layers.27.mlp.experts.60.down_proj", "model.language_model.layers.27.mlp.experts.61.down_proj", "model.language_model.layers.27.mlp.experts.62.down_proj", "model.language_model.layers.27.mlp.experts.63.down_proj", "model.language_model.layers.27.mlp.experts.64.down_proj", "model.language_model.layers.27.mlp.experts.65.down_proj", "model.language_model.layers.27.mlp.experts.66.down_proj", "model.language_model.layers.27.mlp.experts.67.down_proj", "model.language_model.layers.27.mlp.experts.68.down_proj", "model.language_model.layers.27.mlp.experts.69.down_proj", "model.language_model.layers.27.mlp.experts.70.down_proj", "model.language_model.layers.27.mlp.experts.71.down_proj", "model.language_model.layers.27.mlp.experts.72.down_proj", "model.language_model.layers.27.mlp.experts.73.down_proj", "model.language_model.layers.27.mlp.experts.74.down_proj", "model.language_model.layers.27.mlp.experts.75.down_proj", "model.language_model.layers.27.mlp.experts.76.down_proj", "model.language_model.layers.27.mlp.experts.77.down_proj", "model.language_model.layers.27.mlp.experts.78.down_proj", "model.language_model.layers.27.mlp.experts.79.down_proj", "model.language_model.layers.27.mlp.experts.80.down_proj", "model.language_model.layers.27.mlp.experts.81.down_proj", "model.language_model.layers.27.mlp.experts.82.down_proj", "model.language_model.layers.27.mlp.experts.83.down_proj", "model.language_model.layers.27.mlp.experts.84.down_proj", "model.language_model.layers.27.mlp.experts.85.down_proj", "model.language_model.layers.27.mlp.experts.86.down_proj", "model.language_model.layers.27.mlp.experts.87.down_proj", "model.language_model.layers.27.mlp.experts.88.down_proj", "model.language_model.layers.27.mlp.experts.89.down_proj", "model.language_model.layers.27.mlp.experts.90.down_proj", "model.language_model.layers.27.mlp.experts.91.down_proj", "model.language_model.layers.27.mlp.experts.92.down_proj", "model.language_model.layers.27.mlp.experts.93.down_proj", "model.language_model.layers.27.mlp.experts.94.down_proj", "model.language_model.layers.27.mlp.experts.95.down_proj", "model.language_model.layers.27.mlp.experts.96.down_proj", "model.language_model.layers.27.mlp.experts.97.down_proj", "model.language_model.layers.27.mlp.experts.98.down_proj", "model.language_model.layers.27.mlp.experts.99.down_proj", "model.language_model.layers.27.mlp.experts.100.down_proj", "model.language_model.layers.27.mlp.experts.101.down_proj", "model.language_model.layers.27.mlp.experts.102.down_proj", "model.language_model.layers.27.mlp.experts.103.down_proj", "model.language_model.layers.27.mlp.experts.104.down_proj", "model.language_model.layers.27.mlp.experts.105.down_proj", "model.language_model.layers.27.mlp.experts.106.down_proj", "model.language_model.layers.27.mlp.experts.107.down_proj", "model.language_model.layers.27.mlp.experts.108.down_proj", "model.language_model.layers.27.mlp.experts.109.down_proj", "model.language_model.layers.27.mlp.experts.110.down_proj", "model.language_model.layers.27.mlp.experts.111.down_proj", "model.language_model.layers.27.mlp.experts.112.down_proj", "model.language_model.layers.27.mlp.experts.113.down_proj", "model.language_model.layers.27.mlp.experts.114.down_proj", "model.language_model.layers.27.mlp.experts.115.down_proj", "model.language_model.layers.27.mlp.experts.116.down_proj", "model.language_model.layers.27.mlp.experts.117.down_proj", "model.language_model.layers.27.mlp.experts.118.down_proj", "model.language_model.layers.27.mlp.experts.119.down_proj", "model.language_model.layers.27.mlp.experts.120.down_proj", "model.language_model.layers.27.mlp.experts.121.down_proj", "model.language_model.layers.27.mlp.experts.122.down_proj", "model.language_model.layers.27.mlp.experts.123.down_proj", "model.language_model.layers.27.mlp.experts.124.down_proj", "model.language_model.layers.27.mlp.experts.125.down_proj", "model.language_model.layers.27.mlp.experts.126.down_proj", "model.language_model.layers.27.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.0011231271550059263, "dbits": 603979776 } ] }, { "idx": 56, "layers": [ "model.language_model.layers.28.self_attn.q_proj", "model.language_model.layers.28.self_attn.k_proj", "model.language_model.layers.28.self_attn.v_proj", "model.language_model.layers.28.self_attn.o_proj" ], "candidates": [ { "dkld": 0.00116003658622503, "dbits": 18874368 } ] }, { "idx": 57, "layers": [ "model.language_model.layers.28.mlp.experts.0.gate_proj", "model.language_model.layers.28.mlp.experts.1.gate_proj", "model.language_model.layers.28.mlp.experts.2.gate_proj", "model.language_model.layers.28.mlp.experts.3.gate_proj", "model.language_model.layers.28.mlp.experts.4.gate_proj", "model.language_model.layers.28.mlp.experts.5.gate_proj", "model.language_model.layers.28.mlp.experts.6.gate_proj", "model.language_model.layers.28.mlp.experts.7.gate_proj", "model.language_model.layers.28.mlp.experts.8.gate_proj", "model.language_model.layers.28.mlp.experts.9.gate_proj", "model.language_model.layers.28.mlp.experts.10.gate_proj", "model.language_model.layers.28.mlp.experts.11.gate_proj", "model.language_model.layers.28.mlp.experts.12.gate_proj", "model.language_model.layers.28.mlp.experts.13.gate_proj", "model.language_model.layers.28.mlp.experts.14.gate_proj", "model.language_model.layers.28.mlp.experts.15.gate_proj", "model.language_model.layers.28.mlp.experts.16.gate_proj", "model.language_model.layers.28.mlp.experts.17.gate_proj", "model.language_model.layers.28.mlp.experts.18.gate_proj", "model.language_model.layers.28.mlp.experts.19.gate_proj", "model.language_model.layers.28.mlp.experts.20.gate_proj", "model.language_model.layers.28.mlp.experts.21.gate_proj", "model.language_model.layers.28.mlp.experts.22.gate_proj", "model.language_model.layers.28.mlp.experts.23.gate_proj", "model.language_model.layers.28.mlp.experts.24.gate_proj", "model.language_model.layers.28.mlp.experts.25.gate_proj", "model.language_model.layers.28.mlp.experts.26.gate_proj", "model.language_model.layers.28.mlp.experts.27.gate_proj", "model.language_model.layers.28.mlp.experts.28.gate_proj", "model.language_model.layers.28.mlp.experts.29.gate_proj", "model.language_model.layers.28.mlp.experts.30.gate_proj", "model.language_model.layers.28.mlp.experts.31.gate_proj", "model.language_model.layers.28.mlp.experts.32.gate_proj", "model.language_model.layers.28.mlp.experts.33.gate_proj", "model.language_model.layers.28.mlp.experts.34.gate_proj", "model.language_model.layers.28.mlp.experts.35.gate_proj", "model.language_model.layers.28.mlp.experts.36.gate_proj", "model.language_model.layers.28.mlp.experts.37.gate_proj", "model.language_model.layers.28.mlp.experts.38.gate_proj", "model.language_model.layers.28.mlp.experts.39.gate_proj", "model.language_model.layers.28.mlp.experts.40.gate_proj", "model.language_model.layers.28.mlp.experts.41.gate_proj", "model.language_model.layers.28.mlp.experts.42.gate_proj", "model.language_model.layers.28.mlp.experts.43.gate_proj", "model.language_model.layers.28.mlp.experts.44.gate_proj", "model.language_model.layers.28.mlp.experts.45.gate_proj", "model.language_model.layers.28.mlp.experts.46.gate_proj", "model.language_model.layers.28.mlp.experts.47.gate_proj", "model.language_model.layers.28.mlp.experts.48.gate_proj", "model.language_model.layers.28.mlp.experts.49.gate_proj", "model.language_model.layers.28.mlp.experts.50.gate_proj", "model.language_model.layers.28.mlp.experts.51.gate_proj", "model.language_model.layers.28.mlp.experts.52.gate_proj", "model.language_model.layers.28.mlp.experts.53.gate_proj", "model.language_model.layers.28.mlp.experts.54.gate_proj", "model.language_model.layers.28.mlp.experts.55.gate_proj", "model.language_model.layers.28.mlp.experts.56.gate_proj", "model.language_model.layers.28.mlp.experts.57.gate_proj", "model.language_model.layers.28.mlp.experts.58.gate_proj", "model.language_model.layers.28.mlp.experts.59.gate_proj", "model.language_model.layers.28.mlp.experts.60.gate_proj", "model.language_model.layers.28.mlp.experts.61.gate_proj", "model.language_model.layers.28.mlp.experts.62.gate_proj", "model.language_model.layers.28.mlp.experts.63.gate_proj", "model.language_model.layers.28.mlp.experts.64.gate_proj", "model.language_model.layers.28.mlp.experts.65.gate_proj", "model.language_model.layers.28.mlp.experts.66.gate_proj", "model.language_model.layers.28.mlp.experts.67.gate_proj", "model.language_model.layers.28.mlp.experts.68.gate_proj", "model.language_model.layers.28.mlp.experts.69.gate_proj", "model.language_model.layers.28.mlp.experts.70.gate_proj", "model.language_model.layers.28.mlp.experts.71.gate_proj", "model.language_model.layers.28.mlp.experts.72.gate_proj", "model.language_model.layers.28.mlp.experts.73.gate_proj", "model.language_model.layers.28.mlp.experts.74.gate_proj", "model.language_model.layers.28.mlp.experts.75.gate_proj", "model.language_model.layers.28.mlp.experts.76.gate_proj", "model.language_model.layers.28.mlp.experts.77.gate_proj", "model.language_model.layers.28.mlp.experts.78.gate_proj", "model.language_model.layers.28.mlp.experts.79.gate_proj", "model.language_model.layers.28.mlp.experts.80.gate_proj", "model.language_model.layers.28.mlp.experts.81.gate_proj", "model.language_model.layers.28.mlp.experts.82.gate_proj", "model.language_model.layers.28.mlp.experts.83.gate_proj", "model.language_model.layers.28.mlp.experts.84.gate_proj", "model.language_model.layers.28.mlp.experts.85.gate_proj", "model.language_model.layers.28.mlp.experts.86.gate_proj", "model.language_model.layers.28.mlp.experts.87.gate_proj", "model.language_model.layers.28.mlp.experts.88.gate_proj", "model.language_model.layers.28.mlp.experts.89.gate_proj", "model.language_model.layers.28.mlp.experts.90.gate_proj", "model.language_model.layers.28.mlp.experts.91.gate_proj", "model.language_model.layers.28.mlp.experts.92.gate_proj", "model.language_model.layers.28.mlp.experts.93.gate_proj", "model.language_model.layers.28.mlp.experts.94.gate_proj", "model.language_model.layers.28.mlp.experts.95.gate_proj", "model.language_model.layers.28.mlp.experts.96.gate_proj", "model.language_model.layers.28.mlp.experts.97.gate_proj", "model.language_model.layers.28.mlp.experts.98.gate_proj", "model.language_model.layers.28.mlp.experts.99.gate_proj", "model.language_model.layers.28.mlp.experts.100.gate_proj", "model.language_model.layers.28.mlp.experts.101.gate_proj", "model.language_model.layers.28.mlp.experts.102.gate_proj", "model.language_model.layers.28.mlp.experts.103.gate_proj", "model.language_model.layers.28.mlp.experts.104.gate_proj", "model.language_model.layers.28.mlp.experts.105.gate_proj", "model.language_model.layers.28.mlp.experts.106.gate_proj", "model.language_model.layers.28.mlp.experts.107.gate_proj", "model.language_model.layers.28.mlp.experts.108.gate_proj", "model.language_model.layers.28.mlp.experts.109.gate_proj", "model.language_model.layers.28.mlp.experts.110.gate_proj", "model.language_model.layers.28.mlp.experts.111.gate_proj", "model.language_model.layers.28.mlp.experts.112.gate_proj", "model.language_model.layers.28.mlp.experts.113.gate_proj", "model.language_model.layers.28.mlp.experts.114.gate_proj", "model.language_model.layers.28.mlp.experts.115.gate_proj", "model.language_model.layers.28.mlp.experts.116.gate_proj", "model.language_model.layers.28.mlp.experts.117.gate_proj", "model.language_model.layers.28.mlp.experts.118.gate_proj", "model.language_model.layers.28.mlp.experts.119.gate_proj", "model.language_model.layers.28.mlp.experts.120.gate_proj", "model.language_model.layers.28.mlp.experts.121.gate_proj", "model.language_model.layers.28.mlp.experts.122.gate_proj", "model.language_model.layers.28.mlp.experts.123.gate_proj", "model.language_model.layers.28.mlp.experts.124.gate_proj", "model.language_model.layers.28.mlp.experts.125.gate_proj", "model.language_model.layers.28.mlp.experts.126.gate_proj", "model.language_model.layers.28.mlp.experts.127.gate_proj", "model.language_model.layers.28.mlp.experts.0.up_proj", "model.language_model.layers.28.mlp.experts.1.up_proj", "model.language_model.layers.28.mlp.experts.2.up_proj", "model.language_model.layers.28.mlp.experts.3.up_proj", "model.language_model.layers.28.mlp.experts.4.up_proj", "model.language_model.layers.28.mlp.experts.5.up_proj", "model.language_model.layers.28.mlp.experts.6.up_proj", "model.language_model.layers.28.mlp.experts.7.up_proj", "model.language_model.layers.28.mlp.experts.8.up_proj", "model.language_model.layers.28.mlp.experts.9.up_proj", "model.language_model.layers.28.mlp.experts.10.up_proj", "model.language_model.layers.28.mlp.experts.11.up_proj", "model.language_model.layers.28.mlp.experts.12.up_proj", "model.language_model.layers.28.mlp.experts.13.up_proj", "model.language_model.layers.28.mlp.experts.14.up_proj", "model.language_model.layers.28.mlp.experts.15.up_proj", "model.language_model.layers.28.mlp.experts.16.up_proj", "model.language_model.layers.28.mlp.experts.17.up_proj", "model.language_model.layers.28.mlp.experts.18.up_proj", "model.language_model.layers.28.mlp.experts.19.up_proj", "model.language_model.layers.28.mlp.experts.20.up_proj", "model.language_model.layers.28.mlp.experts.21.up_proj", "model.language_model.layers.28.mlp.experts.22.up_proj", "model.language_model.layers.28.mlp.experts.23.up_proj", "model.language_model.layers.28.mlp.experts.24.up_proj", "model.language_model.layers.28.mlp.experts.25.up_proj", "model.language_model.layers.28.mlp.experts.26.up_proj", "model.language_model.layers.28.mlp.experts.27.up_proj", "model.language_model.layers.28.mlp.experts.28.up_proj", "model.language_model.layers.28.mlp.experts.29.up_proj", "model.language_model.layers.28.mlp.experts.30.up_proj", "model.language_model.layers.28.mlp.experts.31.up_proj", "model.language_model.layers.28.mlp.experts.32.up_proj", "model.language_model.layers.28.mlp.experts.33.up_proj", "model.language_model.layers.28.mlp.experts.34.up_proj", "model.language_model.layers.28.mlp.experts.35.up_proj", "model.language_model.layers.28.mlp.experts.36.up_proj", "model.language_model.layers.28.mlp.experts.37.up_proj", "model.language_model.layers.28.mlp.experts.38.up_proj", "model.language_model.layers.28.mlp.experts.39.up_proj", "model.language_model.layers.28.mlp.experts.40.up_proj", "model.language_model.layers.28.mlp.experts.41.up_proj", "model.language_model.layers.28.mlp.experts.42.up_proj", "model.language_model.layers.28.mlp.experts.43.up_proj", "model.language_model.layers.28.mlp.experts.44.up_proj", "model.language_model.layers.28.mlp.experts.45.up_proj", "model.language_model.layers.28.mlp.experts.46.up_proj", "model.language_model.layers.28.mlp.experts.47.up_proj", "model.language_model.layers.28.mlp.experts.48.up_proj", "model.language_model.layers.28.mlp.experts.49.up_proj", "model.language_model.layers.28.mlp.experts.50.up_proj", "model.language_model.layers.28.mlp.experts.51.up_proj", "model.language_model.layers.28.mlp.experts.52.up_proj", "model.language_model.layers.28.mlp.experts.53.up_proj", "model.language_model.layers.28.mlp.experts.54.up_proj", "model.language_model.layers.28.mlp.experts.55.up_proj", "model.language_model.layers.28.mlp.experts.56.up_proj", "model.language_model.layers.28.mlp.experts.57.up_proj", "model.language_model.layers.28.mlp.experts.58.up_proj", "model.language_model.layers.28.mlp.experts.59.up_proj", "model.language_model.layers.28.mlp.experts.60.up_proj", "model.language_model.layers.28.mlp.experts.61.up_proj", "model.language_model.layers.28.mlp.experts.62.up_proj", "model.language_model.layers.28.mlp.experts.63.up_proj", "model.language_model.layers.28.mlp.experts.64.up_proj", "model.language_model.layers.28.mlp.experts.65.up_proj", "model.language_model.layers.28.mlp.experts.66.up_proj", "model.language_model.layers.28.mlp.experts.67.up_proj", "model.language_model.layers.28.mlp.experts.68.up_proj", "model.language_model.layers.28.mlp.experts.69.up_proj", "model.language_model.layers.28.mlp.experts.70.up_proj", "model.language_model.layers.28.mlp.experts.71.up_proj", "model.language_model.layers.28.mlp.experts.72.up_proj", "model.language_model.layers.28.mlp.experts.73.up_proj", "model.language_model.layers.28.mlp.experts.74.up_proj", "model.language_model.layers.28.mlp.experts.75.up_proj", "model.language_model.layers.28.mlp.experts.76.up_proj", "model.language_model.layers.28.mlp.experts.77.up_proj", "model.language_model.layers.28.mlp.experts.78.up_proj", "model.language_model.layers.28.mlp.experts.79.up_proj", "model.language_model.layers.28.mlp.experts.80.up_proj", "model.language_model.layers.28.mlp.experts.81.up_proj", "model.language_model.layers.28.mlp.experts.82.up_proj", "model.language_model.layers.28.mlp.experts.83.up_proj", "model.language_model.layers.28.mlp.experts.84.up_proj", "model.language_model.layers.28.mlp.experts.85.up_proj", "model.language_model.layers.28.mlp.experts.86.up_proj", "model.language_model.layers.28.mlp.experts.87.up_proj", "model.language_model.layers.28.mlp.experts.88.up_proj", "model.language_model.layers.28.mlp.experts.89.up_proj", "model.language_model.layers.28.mlp.experts.90.up_proj", "model.language_model.layers.28.mlp.experts.91.up_proj", "model.language_model.layers.28.mlp.experts.92.up_proj", "model.language_model.layers.28.mlp.experts.93.up_proj", "model.language_model.layers.28.mlp.experts.94.up_proj", "model.language_model.layers.28.mlp.experts.95.up_proj", "model.language_model.layers.28.mlp.experts.96.up_proj", "model.language_model.layers.28.mlp.experts.97.up_proj", "model.language_model.layers.28.mlp.experts.98.up_proj", "model.language_model.layers.28.mlp.experts.99.up_proj", "model.language_model.layers.28.mlp.experts.100.up_proj", "model.language_model.layers.28.mlp.experts.101.up_proj", "model.language_model.layers.28.mlp.experts.102.up_proj", "model.language_model.layers.28.mlp.experts.103.up_proj", "model.language_model.layers.28.mlp.experts.104.up_proj", "model.language_model.layers.28.mlp.experts.105.up_proj", "model.language_model.layers.28.mlp.experts.106.up_proj", "model.language_model.layers.28.mlp.experts.107.up_proj", "model.language_model.layers.28.mlp.experts.108.up_proj", "model.language_model.layers.28.mlp.experts.109.up_proj", "model.language_model.layers.28.mlp.experts.110.up_proj", "model.language_model.layers.28.mlp.experts.111.up_proj", "model.language_model.layers.28.mlp.experts.112.up_proj", "model.language_model.layers.28.mlp.experts.113.up_proj", "model.language_model.layers.28.mlp.experts.114.up_proj", "model.language_model.layers.28.mlp.experts.115.up_proj", "model.language_model.layers.28.mlp.experts.116.up_proj", "model.language_model.layers.28.mlp.experts.117.up_proj", "model.language_model.layers.28.mlp.experts.118.up_proj", "model.language_model.layers.28.mlp.experts.119.up_proj", "model.language_model.layers.28.mlp.experts.120.up_proj", "model.language_model.layers.28.mlp.experts.121.up_proj", "model.language_model.layers.28.mlp.experts.122.up_proj", "model.language_model.layers.28.mlp.experts.123.up_proj", "model.language_model.layers.28.mlp.experts.124.up_proj", "model.language_model.layers.28.mlp.experts.125.up_proj", "model.language_model.layers.28.mlp.experts.126.up_proj", "model.language_model.layers.28.mlp.experts.127.up_proj", "model.language_model.layers.28.mlp.experts.0.down_proj", "model.language_model.layers.28.mlp.experts.1.down_proj", "model.language_model.layers.28.mlp.experts.2.down_proj", "model.language_model.layers.28.mlp.experts.3.down_proj", "model.language_model.layers.28.mlp.experts.4.down_proj", "model.language_model.layers.28.mlp.experts.5.down_proj", "model.language_model.layers.28.mlp.experts.6.down_proj", "model.language_model.layers.28.mlp.experts.7.down_proj", "model.language_model.layers.28.mlp.experts.8.down_proj", "model.language_model.layers.28.mlp.experts.9.down_proj", "model.language_model.layers.28.mlp.experts.10.down_proj", "model.language_model.layers.28.mlp.experts.11.down_proj", "model.language_model.layers.28.mlp.experts.12.down_proj", "model.language_model.layers.28.mlp.experts.13.down_proj", "model.language_model.layers.28.mlp.experts.14.down_proj", "model.language_model.layers.28.mlp.experts.15.down_proj", "model.language_model.layers.28.mlp.experts.16.down_proj", "model.language_model.layers.28.mlp.experts.17.down_proj", "model.language_model.layers.28.mlp.experts.18.down_proj", "model.language_model.layers.28.mlp.experts.19.down_proj", "model.language_model.layers.28.mlp.experts.20.down_proj", "model.language_model.layers.28.mlp.experts.21.down_proj", "model.language_model.layers.28.mlp.experts.22.down_proj", "model.language_model.layers.28.mlp.experts.23.down_proj", "model.language_model.layers.28.mlp.experts.24.down_proj", "model.language_model.layers.28.mlp.experts.25.down_proj", "model.language_model.layers.28.mlp.experts.26.down_proj", "model.language_model.layers.28.mlp.experts.27.down_proj", "model.language_model.layers.28.mlp.experts.28.down_proj", "model.language_model.layers.28.mlp.experts.29.down_proj", "model.language_model.layers.28.mlp.experts.30.down_proj", "model.language_model.layers.28.mlp.experts.31.down_proj", "model.language_model.layers.28.mlp.experts.32.down_proj", "model.language_model.layers.28.mlp.experts.33.down_proj", "model.language_model.layers.28.mlp.experts.34.down_proj", "model.language_model.layers.28.mlp.experts.35.down_proj", "model.language_model.layers.28.mlp.experts.36.down_proj", "model.language_model.layers.28.mlp.experts.37.down_proj", "model.language_model.layers.28.mlp.experts.38.down_proj", "model.language_model.layers.28.mlp.experts.39.down_proj", "model.language_model.layers.28.mlp.experts.40.down_proj", "model.language_model.layers.28.mlp.experts.41.down_proj", "model.language_model.layers.28.mlp.experts.42.down_proj", "model.language_model.layers.28.mlp.experts.43.down_proj", "model.language_model.layers.28.mlp.experts.44.down_proj", "model.language_model.layers.28.mlp.experts.45.down_proj", "model.language_model.layers.28.mlp.experts.46.down_proj", "model.language_model.layers.28.mlp.experts.47.down_proj", "model.language_model.layers.28.mlp.experts.48.down_proj", "model.language_model.layers.28.mlp.experts.49.down_proj", "model.language_model.layers.28.mlp.experts.50.down_proj", "model.language_model.layers.28.mlp.experts.51.down_proj", "model.language_model.layers.28.mlp.experts.52.down_proj", "model.language_model.layers.28.mlp.experts.53.down_proj", "model.language_model.layers.28.mlp.experts.54.down_proj", "model.language_model.layers.28.mlp.experts.55.down_proj", "model.language_model.layers.28.mlp.experts.56.down_proj", "model.language_model.layers.28.mlp.experts.57.down_proj", "model.language_model.layers.28.mlp.experts.58.down_proj", "model.language_model.layers.28.mlp.experts.59.down_proj", "model.language_model.layers.28.mlp.experts.60.down_proj", "model.language_model.layers.28.mlp.experts.61.down_proj", "model.language_model.layers.28.mlp.experts.62.down_proj", "model.language_model.layers.28.mlp.experts.63.down_proj", "model.language_model.layers.28.mlp.experts.64.down_proj", "model.language_model.layers.28.mlp.experts.65.down_proj", "model.language_model.layers.28.mlp.experts.66.down_proj", "model.language_model.layers.28.mlp.experts.67.down_proj", "model.language_model.layers.28.mlp.experts.68.down_proj", "model.language_model.layers.28.mlp.experts.69.down_proj", "model.language_model.layers.28.mlp.experts.70.down_proj", "model.language_model.layers.28.mlp.experts.71.down_proj", "model.language_model.layers.28.mlp.experts.72.down_proj", "model.language_model.layers.28.mlp.experts.73.down_proj", "model.language_model.layers.28.mlp.experts.74.down_proj", "model.language_model.layers.28.mlp.experts.75.down_proj", "model.language_model.layers.28.mlp.experts.76.down_proj", "model.language_model.layers.28.mlp.experts.77.down_proj", "model.language_model.layers.28.mlp.experts.78.down_proj", "model.language_model.layers.28.mlp.experts.79.down_proj", "model.language_model.layers.28.mlp.experts.80.down_proj", "model.language_model.layers.28.mlp.experts.81.down_proj", "model.language_model.layers.28.mlp.experts.82.down_proj", "model.language_model.layers.28.mlp.experts.83.down_proj", "model.language_model.layers.28.mlp.experts.84.down_proj", "model.language_model.layers.28.mlp.experts.85.down_proj", "model.language_model.layers.28.mlp.experts.86.down_proj", "model.language_model.layers.28.mlp.experts.87.down_proj", "model.language_model.layers.28.mlp.experts.88.down_proj", "model.language_model.layers.28.mlp.experts.89.down_proj", "model.language_model.layers.28.mlp.experts.90.down_proj", "model.language_model.layers.28.mlp.experts.91.down_proj", "model.language_model.layers.28.mlp.experts.92.down_proj", "model.language_model.layers.28.mlp.experts.93.down_proj", "model.language_model.layers.28.mlp.experts.94.down_proj", "model.language_model.layers.28.mlp.experts.95.down_proj", "model.language_model.layers.28.mlp.experts.96.down_proj", "model.language_model.layers.28.mlp.experts.97.down_proj", "model.language_model.layers.28.mlp.experts.98.down_proj", "model.language_model.layers.28.mlp.experts.99.down_proj", "model.language_model.layers.28.mlp.experts.100.down_proj", "model.language_model.layers.28.mlp.experts.101.down_proj", "model.language_model.layers.28.mlp.experts.102.down_proj", "model.language_model.layers.28.mlp.experts.103.down_proj", "model.language_model.layers.28.mlp.experts.104.down_proj", "model.language_model.layers.28.mlp.experts.105.down_proj", "model.language_model.layers.28.mlp.experts.106.down_proj", "model.language_model.layers.28.mlp.experts.107.down_proj", "model.language_model.layers.28.mlp.experts.108.down_proj", "model.language_model.layers.28.mlp.experts.109.down_proj", "model.language_model.layers.28.mlp.experts.110.down_proj", "model.language_model.layers.28.mlp.experts.111.down_proj", "model.language_model.layers.28.mlp.experts.112.down_proj", "model.language_model.layers.28.mlp.experts.113.down_proj", "model.language_model.layers.28.mlp.experts.114.down_proj", "model.language_model.layers.28.mlp.experts.115.down_proj", "model.language_model.layers.28.mlp.experts.116.down_proj", "model.language_model.layers.28.mlp.experts.117.down_proj", "model.language_model.layers.28.mlp.experts.118.down_proj", "model.language_model.layers.28.mlp.experts.119.down_proj", "model.language_model.layers.28.mlp.experts.120.down_proj", "model.language_model.layers.28.mlp.experts.121.down_proj", "model.language_model.layers.28.mlp.experts.122.down_proj", "model.language_model.layers.28.mlp.experts.123.down_proj", "model.language_model.layers.28.mlp.experts.124.down_proj", "model.language_model.layers.28.mlp.experts.125.down_proj", "model.language_model.layers.28.mlp.experts.126.down_proj", "model.language_model.layers.28.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.0005738901905715521, "dbits": 603979776 } ] }, { "idx": 58, "layers": [ "model.language_model.layers.29.self_attn.q_proj", "model.language_model.layers.29.self_attn.k_proj", "model.language_model.layers.29.self_attn.v_proj", "model.language_model.layers.29.self_attn.o_proj" ], "candidates": [ { "dkld": -0.00038300864398480017, "dbits": 18874368 } ] }, { "idx": 59, "layers": [ "model.language_model.layers.29.mlp.experts.0.gate_proj", "model.language_model.layers.29.mlp.experts.1.gate_proj", "model.language_model.layers.29.mlp.experts.2.gate_proj", "model.language_model.layers.29.mlp.experts.3.gate_proj", "model.language_model.layers.29.mlp.experts.4.gate_proj", "model.language_model.layers.29.mlp.experts.5.gate_proj", "model.language_model.layers.29.mlp.experts.6.gate_proj", "model.language_model.layers.29.mlp.experts.7.gate_proj", "model.language_model.layers.29.mlp.experts.8.gate_proj", "model.language_model.layers.29.mlp.experts.9.gate_proj", "model.language_model.layers.29.mlp.experts.10.gate_proj", "model.language_model.layers.29.mlp.experts.11.gate_proj", "model.language_model.layers.29.mlp.experts.12.gate_proj", "model.language_model.layers.29.mlp.experts.13.gate_proj", "model.language_model.layers.29.mlp.experts.14.gate_proj", "model.language_model.layers.29.mlp.experts.15.gate_proj", "model.language_model.layers.29.mlp.experts.16.gate_proj", "model.language_model.layers.29.mlp.experts.17.gate_proj", "model.language_model.layers.29.mlp.experts.18.gate_proj", "model.language_model.layers.29.mlp.experts.19.gate_proj", "model.language_model.layers.29.mlp.experts.20.gate_proj", "model.language_model.layers.29.mlp.experts.21.gate_proj", "model.language_model.layers.29.mlp.experts.22.gate_proj", "model.language_model.layers.29.mlp.experts.23.gate_proj", "model.language_model.layers.29.mlp.experts.24.gate_proj", "model.language_model.layers.29.mlp.experts.25.gate_proj", "model.language_model.layers.29.mlp.experts.26.gate_proj", "model.language_model.layers.29.mlp.experts.27.gate_proj", "model.language_model.layers.29.mlp.experts.28.gate_proj", "model.language_model.layers.29.mlp.experts.29.gate_proj", "model.language_model.layers.29.mlp.experts.30.gate_proj", "model.language_model.layers.29.mlp.experts.31.gate_proj", "model.language_model.layers.29.mlp.experts.32.gate_proj", "model.language_model.layers.29.mlp.experts.33.gate_proj", "model.language_model.layers.29.mlp.experts.34.gate_proj", "model.language_model.layers.29.mlp.experts.35.gate_proj", "model.language_model.layers.29.mlp.experts.36.gate_proj", "model.language_model.layers.29.mlp.experts.37.gate_proj", "model.language_model.layers.29.mlp.experts.38.gate_proj", "model.language_model.layers.29.mlp.experts.39.gate_proj", "model.language_model.layers.29.mlp.experts.40.gate_proj", "model.language_model.layers.29.mlp.experts.41.gate_proj", "model.language_model.layers.29.mlp.experts.42.gate_proj", "model.language_model.layers.29.mlp.experts.43.gate_proj", "model.language_model.layers.29.mlp.experts.44.gate_proj", "model.language_model.layers.29.mlp.experts.45.gate_proj", "model.language_model.layers.29.mlp.experts.46.gate_proj", "model.language_model.layers.29.mlp.experts.47.gate_proj", "model.language_model.layers.29.mlp.experts.48.gate_proj", "model.language_model.layers.29.mlp.experts.49.gate_proj", "model.language_model.layers.29.mlp.experts.50.gate_proj", "model.language_model.layers.29.mlp.experts.51.gate_proj", "model.language_model.layers.29.mlp.experts.52.gate_proj", "model.language_model.layers.29.mlp.experts.53.gate_proj", "model.language_model.layers.29.mlp.experts.54.gate_proj", "model.language_model.layers.29.mlp.experts.55.gate_proj", "model.language_model.layers.29.mlp.experts.56.gate_proj", "model.language_model.layers.29.mlp.experts.57.gate_proj", "model.language_model.layers.29.mlp.experts.58.gate_proj", "model.language_model.layers.29.mlp.experts.59.gate_proj", "model.language_model.layers.29.mlp.experts.60.gate_proj", "model.language_model.layers.29.mlp.experts.61.gate_proj", "model.language_model.layers.29.mlp.experts.62.gate_proj", "model.language_model.layers.29.mlp.experts.63.gate_proj", "model.language_model.layers.29.mlp.experts.64.gate_proj", "model.language_model.layers.29.mlp.experts.65.gate_proj", "model.language_model.layers.29.mlp.experts.66.gate_proj", "model.language_model.layers.29.mlp.experts.67.gate_proj", "model.language_model.layers.29.mlp.experts.68.gate_proj", "model.language_model.layers.29.mlp.experts.69.gate_proj", "model.language_model.layers.29.mlp.experts.70.gate_proj", "model.language_model.layers.29.mlp.experts.71.gate_proj", "model.language_model.layers.29.mlp.experts.72.gate_proj", "model.language_model.layers.29.mlp.experts.73.gate_proj", "model.language_model.layers.29.mlp.experts.74.gate_proj", "model.language_model.layers.29.mlp.experts.75.gate_proj", "model.language_model.layers.29.mlp.experts.76.gate_proj", "model.language_model.layers.29.mlp.experts.77.gate_proj", "model.language_model.layers.29.mlp.experts.78.gate_proj", "model.language_model.layers.29.mlp.experts.79.gate_proj", "model.language_model.layers.29.mlp.experts.80.gate_proj", "model.language_model.layers.29.mlp.experts.81.gate_proj", "model.language_model.layers.29.mlp.experts.82.gate_proj", "model.language_model.layers.29.mlp.experts.83.gate_proj", "model.language_model.layers.29.mlp.experts.84.gate_proj", "model.language_model.layers.29.mlp.experts.85.gate_proj", "model.language_model.layers.29.mlp.experts.86.gate_proj", "model.language_model.layers.29.mlp.experts.87.gate_proj", "model.language_model.layers.29.mlp.experts.88.gate_proj", "model.language_model.layers.29.mlp.experts.89.gate_proj", "model.language_model.layers.29.mlp.experts.90.gate_proj", "model.language_model.layers.29.mlp.experts.91.gate_proj", "model.language_model.layers.29.mlp.experts.92.gate_proj", "model.language_model.layers.29.mlp.experts.93.gate_proj", "model.language_model.layers.29.mlp.experts.94.gate_proj", "model.language_model.layers.29.mlp.experts.95.gate_proj", "model.language_model.layers.29.mlp.experts.96.gate_proj", "model.language_model.layers.29.mlp.experts.97.gate_proj", "model.language_model.layers.29.mlp.experts.98.gate_proj", "model.language_model.layers.29.mlp.experts.99.gate_proj", "model.language_model.layers.29.mlp.experts.100.gate_proj", "model.language_model.layers.29.mlp.experts.101.gate_proj", "model.language_model.layers.29.mlp.experts.102.gate_proj", "model.language_model.layers.29.mlp.experts.103.gate_proj", "model.language_model.layers.29.mlp.experts.104.gate_proj", "model.language_model.layers.29.mlp.experts.105.gate_proj", "model.language_model.layers.29.mlp.experts.106.gate_proj", "model.language_model.layers.29.mlp.experts.107.gate_proj", "model.language_model.layers.29.mlp.experts.108.gate_proj", "model.language_model.layers.29.mlp.experts.109.gate_proj", "model.language_model.layers.29.mlp.experts.110.gate_proj", "model.language_model.layers.29.mlp.experts.111.gate_proj", "model.language_model.layers.29.mlp.experts.112.gate_proj", "model.language_model.layers.29.mlp.experts.113.gate_proj", "model.language_model.layers.29.mlp.experts.114.gate_proj", "model.language_model.layers.29.mlp.experts.115.gate_proj", "model.language_model.layers.29.mlp.experts.116.gate_proj", "model.language_model.layers.29.mlp.experts.117.gate_proj", "model.language_model.layers.29.mlp.experts.118.gate_proj", "model.language_model.layers.29.mlp.experts.119.gate_proj", "model.language_model.layers.29.mlp.experts.120.gate_proj", "model.language_model.layers.29.mlp.experts.121.gate_proj", "model.language_model.layers.29.mlp.experts.122.gate_proj", "model.language_model.layers.29.mlp.experts.123.gate_proj", "model.language_model.layers.29.mlp.experts.124.gate_proj", "model.language_model.layers.29.mlp.experts.125.gate_proj", "model.language_model.layers.29.mlp.experts.126.gate_proj", "model.language_model.layers.29.mlp.experts.127.gate_proj", "model.language_model.layers.29.mlp.experts.0.up_proj", "model.language_model.layers.29.mlp.experts.1.up_proj", "model.language_model.layers.29.mlp.experts.2.up_proj", "model.language_model.layers.29.mlp.experts.3.up_proj", "model.language_model.layers.29.mlp.experts.4.up_proj", "model.language_model.layers.29.mlp.experts.5.up_proj", "model.language_model.layers.29.mlp.experts.6.up_proj", "model.language_model.layers.29.mlp.experts.7.up_proj", "model.language_model.layers.29.mlp.experts.8.up_proj", "model.language_model.layers.29.mlp.experts.9.up_proj", "model.language_model.layers.29.mlp.experts.10.up_proj", "model.language_model.layers.29.mlp.experts.11.up_proj", "model.language_model.layers.29.mlp.experts.12.up_proj", "model.language_model.layers.29.mlp.experts.13.up_proj", "model.language_model.layers.29.mlp.experts.14.up_proj", "model.language_model.layers.29.mlp.experts.15.up_proj", "model.language_model.layers.29.mlp.experts.16.up_proj", "model.language_model.layers.29.mlp.experts.17.up_proj", "model.language_model.layers.29.mlp.experts.18.up_proj", "model.language_model.layers.29.mlp.experts.19.up_proj", "model.language_model.layers.29.mlp.experts.20.up_proj", "model.language_model.layers.29.mlp.experts.21.up_proj", "model.language_model.layers.29.mlp.experts.22.up_proj", "model.language_model.layers.29.mlp.experts.23.up_proj", "model.language_model.layers.29.mlp.experts.24.up_proj", "model.language_model.layers.29.mlp.experts.25.up_proj", "model.language_model.layers.29.mlp.experts.26.up_proj", "model.language_model.layers.29.mlp.experts.27.up_proj", "model.language_model.layers.29.mlp.experts.28.up_proj", "model.language_model.layers.29.mlp.experts.29.up_proj", "model.language_model.layers.29.mlp.experts.30.up_proj", "model.language_model.layers.29.mlp.experts.31.up_proj", "model.language_model.layers.29.mlp.experts.32.up_proj", "model.language_model.layers.29.mlp.experts.33.up_proj", "model.language_model.layers.29.mlp.experts.34.up_proj", "model.language_model.layers.29.mlp.experts.35.up_proj", "model.language_model.layers.29.mlp.experts.36.up_proj", "model.language_model.layers.29.mlp.experts.37.up_proj", "model.language_model.layers.29.mlp.experts.38.up_proj", "model.language_model.layers.29.mlp.experts.39.up_proj", "model.language_model.layers.29.mlp.experts.40.up_proj", "model.language_model.layers.29.mlp.experts.41.up_proj", "model.language_model.layers.29.mlp.experts.42.up_proj", "model.language_model.layers.29.mlp.experts.43.up_proj", "model.language_model.layers.29.mlp.experts.44.up_proj", "model.language_model.layers.29.mlp.experts.45.up_proj", "model.language_model.layers.29.mlp.experts.46.up_proj", "model.language_model.layers.29.mlp.experts.47.up_proj", "model.language_model.layers.29.mlp.experts.48.up_proj", "model.language_model.layers.29.mlp.experts.49.up_proj", "model.language_model.layers.29.mlp.experts.50.up_proj", "model.language_model.layers.29.mlp.experts.51.up_proj", "model.language_model.layers.29.mlp.experts.52.up_proj", "model.language_model.layers.29.mlp.experts.53.up_proj", "model.language_model.layers.29.mlp.experts.54.up_proj", "model.language_model.layers.29.mlp.experts.55.up_proj", "model.language_model.layers.29.mlp.experts.56.up_proj", "model.language_model.layers.29.mlp.experts.57.up_proj", "model.language_model.layers.29.mlp.experts.58.up_proj", "model.language_model.layers.29.mlp.experts.59.up_proj", "model.language_model.layers.29.mlp.experts.60.up_proj", "model.language_model.layers.29.mlp.experts.61.up_proj", "model.language_model.layers.29.mlp.experts.62.up_proj", "model.language_model.layers.29.mlp.experts.63.up_proj", "model.language_model.layers.29.mlp.experts.64.up_proj", "model.language_model.layers.29.mlp.experts.65.up_proj", "model.language_model.layers.29.mlp.experts.66.up_proj", "model.language_model.layers.29.mlp.experts.67.up_proj", "model.language_model.layers.29.mlp.experts.68.up_proj", "model.language_model.layers.29.mlp.experts.69.up_proj", "model.language_model.layers.29.mlp.experts.70.up_proj", "model.language_model.layers.29.mlp.experts.71.up_proj", "model.language_model.layers.29.mlp.experts.72.up_proj", "model.language_model.layers.29.mlp.experts.73.up_proj", "model.language_model.layers.29.mlp.experts.74.up_proj", "model.language_model.layers.29.mlp.experts.75.up_proj", "model.language_model.layers.29.mlp.experts.76.up_proj", "model.language_model.layers.29.mlp.experts.77.up_proj", "model.language_model.layers.29.mlp.experts.78.up_proj", "model.language_model.layers.29.mlp.experts.79.up_proj", "model.language_model.layers.29.mlp.experts.80.up_proj", "model.language_model.layers.29.mlp.experts.81.up_proj", "model.language_model.layers.29.mlp.experts.82.up_proj", "model.language_model.layers.29.mlp.experts.83.up_proj", "model.language_model.layers.29.mlp.experts.84.up_proj", "model.language_model.layers.29.mlp.experts.85.up_proj", "model.language_model.layers.29.mlp.experts.86.up_proj", "model.language_model.layers.29.mlp.experts.87.up_proj", "model.language_model.layers.29.mlp.experts.88.up_proj", "model.language_model.layers.29.mlp.experts.89.up_proj", "model.language_model.layers.29.mlp.experts.90.up_proj", "model.language_model.layers.29.mlp.experts.91.up_proj", "model.language_model.layers.29.mlp.experts.92.up_proj", "model.language_model.layers.29.mlp.experts.93.up_proj", "model.language_model.layers.29.mlp.experts.94.up_proj", "model.language_model.layers.29.mlp.experts.95.up_proj", "model.language_model.layers.29.mlp.experts.96.up_proj", "model.language_model.layers.29.mlp.experts.97.up_proj", "model.language_model.layers.29.mlp.experts.98.up_proj", "model.language_model.layers.29.mlp.experts.99.up_proj", "model.language_model.layers.29.mlp.experts.100.up_proj", "model.language_model.layers.29.mlp.experts.101.up_proj", "model.language_model.layers.29.mlp.experts.102.up_proj", "model.language_model.layers.29.mlp.experts.103.up_proj", "model.language_model.layers.29.mlp.experts.104.up_proj", "model.language_model.layers.29.mlp.experts.105.up_proj", "model.language_model.layers.29.mlp.experts.106.up_proj", "model.language_model.layers.29.mlp.experts.107.up_proj", "model.language_model.layers.29.mlp.experts.108.up_proj", "model.language_model.layers.29.mlp.experts.109.up_proj", "model.language_model.layers.29.mlp.experts.110.up_proj", "model.language_model.layers.29.mlp.experts.111.up_proj", "model.language_model.layers.29.mlp.experts.112.up_proj", "model.language_model.layers.29.mlp.experts.113.up_proj", "model.language_model.layers.29.mlp.experts.114.up_proj", "model.language_model.layers.29.mlp.experts.115.up_proj", "model.language_model.layers.29.mlp.experts.116.up_proj", "model.language_model.layers.29.mlp.experts.117.up_proj", "model.language_model.layers.29.mlp.experts.118.up_proj", "model.language_model.layers.29.mlp.experts.119.up_proj", "model.language_model.layers.29.mlp.experts.120.up_proj", "model.language_model.layers.29.mlp.experts.121.up_proj", "model.language_model.layers.29.mlp.experts.122.up_proj", "model.language_model.layers.29.mlp.experts.123.up_proj", "model.language_model.layers.29.mlp.experts.124.up_proj", "model.language_model.layers.29.mlp.experts.125.up_proj", "model.language_model.layers.29.mlp.experts.126.up_proj", "model.language_model.layers.29.mlp.experts.127.up_proj", "model.language_model.layers.29.mlp.experts.0.down_proj", "model.language_model.layers.29.mlp.experts.1.down_proj", "model.language_model.layers.29.mlp.experts.2.down_proj", "model.language_model.layers.29.mlp.experts.3.down_proj", "model.language_model.layers.29.mlp.experts.4.down_proj", "model.language_model.layers.29.mlp.experts.5.down_proj", "model.language_model.layers.29.mlp.experts.6.down_proj", "model.language_model.layers.29.mlp.experts.7.down_proj", "model.language_model.layers.29.mlp.experts.8.down_proj", "model.language_model.layers.29.mlp.experts.9.down_proj", "model.language_model.layers.29.mlp.experts.10.down_proj", "model.language_model.layers.29.mlp.experts.11.down_proj", "model.language_model.layers.29.mlp.experts.12.down_proj", "model.language_model.layers.29.mlp.experts.13.down_proj", "model.language_model.layers.29.mlp.experts.14.down_proj", "model.language_model.layers.29.mlp.experts.15.down_proj", "model.language_model.layers.29.mlp.experts.16.down_proj", "model.language_model.layers.29.mlp.experts.17.down_proj", "model.language_model.layers.29.mlp.experts.18.down_proj", "model.language_model.layers.29.mlp.experts.19.down_proj", "model.language_model.layers.29.mlp.experts.20.down_proj", "model.language_model.layers.29.mlp.experts.21.down_proj", "model.language_model.layers.29.mlp.experts.22.down_proj", "model.language_model.layers.29.mlp.experts.23.down_proj", "model.language_model.layers.29.mlp.experts.24.down_proj", "model.language_model.layers.29.mlp.experts.25.down_proj", "model.language_model.layers.29.mlp.experts.26.down_proj", "model.language_model.layers.29.mlp.experts.27.down_proj", "model.language_model.layers.29.mlp.experts.28.down_proj", "model.language_model.layers.29.mlp.experts.29.down_proj", "model.language_model.layers.29.mlp.experts.30.down_proj", "model.language_model.layers.29.mlp.experts.31.down_proj", "model.language_model.layers.29.mlp.experts.32.down_proj", "model.language_model.layers.29.mlp.experts.33.down_proj", "model.language_model.layers.29.mlp.experts.34.down_proj", "model.language_model.layers.29.mlp.experts.35.down_proj", "model.language_model.layers.29.mlp.experts.36.down_proj", "model.language_model.layers.29.mlp.experts.37.down_proj", "model.language_model.layers.29.mlp.experts.38.down_proj", "model.language_model.layers.29.mlp.experts.39.down_proj", "model.language_model.layers.29.mlp.experts.40.down_proj", "model.language_model.layers.29.mlp.experts.41.down_proj", "model.language_model.layers.29.mlp.experts.42.down_proj", "model.language_model.layers.29.mlp.experts.43.down_proj", "model.language_model.layers.29.mlp.experts.44.down_proj", "model.language_model.layers.29.mlp.experts.45.down_proj", "model.language_model.layers.29.mlp.experts.46.down_proj", "model.language_model.layers.29.mlp.experts.47.down_proj", "model.language_model.layers.29.mlp.experts.48.down_proj", "model.language_model.layers.29.mlp.experts.49.down_proj", "model.language_model.layers.29.mlp.experts.50.down_proj", "model.language_model.layers.29.mlp.experts.51.down_proj", "model.language_model.layers.29.mlp.experts.52.down_proj", "model.language_model.layers.29.mlp.experts.53.down_proj", "model.language_model.layers.29.mlp.experts.54.down_proj", "model.language_model.layers.29.mlp.experts.55.down_proj", "model.language_model.layers.29.mlp.experts.56.down_proj", "model.language_model.layers.29.mlp.experts.57.down_proj", "model.language_model.layers.29.mlp.experts.58.down_proj", "model.language_model.layers.29.mlp.experts.59.down_proj", "model.language_model.layers.29.mlp.experts.60.down_proj", "model.language_model.layers.29.mlp.experts.61.down_proj", "model.language_model.layers.29.mlp.experts.62.down_proj", "model.language_model.layers.29.mlp.experts.63.down_proj", "model.language_model.layers.29.mlp.experts.64.down_proj", "model.language_model.layers.29.mlp.experts.65.down_proj", "model.language_model.layers.29.mlp.experts.66.down_proj", "model.language_model.layers.29.mlp.experts.67.down_proj", "model.language_model.layers.29.mlp.experts.68.down_proj", "model.language_model.layers.29.mlp.experts.69.down_proj", "model.language_model.layers.29.mlp.experts.70.down_proj", "model.language_model.layers.29.mlp.experts.71.down_proj", "model.language_model.layers.29.mlp.experts.72.down_proj", "model.language_model.layers.29.mlp.experts.73.down_proj", "model.language_model.layers.29.mlp.experts.74.down_proj", "model.language_model.layers.29.mlp.experts.75.down_proj", "model.language_model.layers.29.mlp.experts.76.down_proj", "model.language_model.layers.29.mlp.experts.77.down_proj", "model.language_model.layers.29.mlp.experts.78.down_proj", "model.language_model.layers.29.mlp.experts.79.down_proj", "model.language_model.layers.29.mlp.experts.80.down_proj", "model.language_model.layers.29.mlp.experts.81.down_proj", "model.language_model.layers.29.mlp.experts.82.down_proj", "model.language_model.layers.29.mlp.experts.83.down_proj", "model.language_model.layers.29.mlp.experts.84.down_proj", "model.language_model.layers.29.mlp.experts.85.down_proj", "model.language_model.layers.29.mlp.experts.86.down_proj", "model.language_model.layers.29.mlp.experts.87.down_proj", "model.language_model.layers.29.mlp.experts.88.down_proj", "model.language_model.layers.29.mlp.experts.89.down_proj", "model.language_model.layers.29.mlp.experts.90.down_proj", "model.language_model.layers.29.mlp.experts.91.down_proj", "model.language_model.layers.29.mlp.experts.92.down_proj", "model.language_model.layers.29.mlp.experts.93.down_proj", "model.language_model.layers.29.mlp.experts.94.down_proj", "model.language_model.layers.29.mlp.experts.95.down_proj", "model.language_model.layers.29.mlp.experts.96.down_proj", "model.language_model.layers.29.mlp.experts.97.down_proj", "model.language_model.layers.29.mlp.experts.98.down_proj", "model.language_model.layers.29.mlp.experts.99.down_proj", "model.language_model.layers.29.mlp.experts.100.down_proj", "model.language_model.layers.29.mlp.experts.101.down_proj", "model.language_model.layers.29.mlp.experts.102.down_proj", "model.language_model.layers.29.mlp.experts.103.down_proj", "model.language_model.layers.29.mlp.experts.104.down_proj", "model.language_model.layers.29.mlp.experts.105.down_proj", "model.language_model.layers.29.mlp.experts.106.down_proj", "model.language_model.layers.29.mlp.experts.107.down_proj", "model.language_model.layers.29.mlp.experts.108.down_proj", "model.language_model.layers.29.mlp.experts.109.down_proj", "model.language_model.layers.29.mlp.experts.110.down_proj", "model.language_model.layers.29.mlp.experts.111.down_proj", "model.language_model.layers.29.mlp.experts.112.down_proj", "model.language_model.layers.29.mlp.experts.113.down_proj", "model.language_model.layers.29.mlp.experts.114.down_proj", "model.language_model.layers.29.mlp.experts.115.down_proj", "model.language_model.layers.29.mlp.experts.116.down_proj", "model.language_model.layers.29.mlp.experts.117.down_proj", "model.language_model.layers.29.mlp.experts.118.down_proj", "model.language_model.layers.29.mlp.experts.119.down_proj", "model.language_model.layers.29.mlp.experts.120.down_proj", "model.language_model.layers.29.mlp.experts.121.down_proj", "model.language_model.layers.29.mlp.experts.122.down_proj", "model.language_model.layers.29.mlp.experts.123.down_proj", "model.language_model.layers.29.mlp.experts.124.down_proj", "model.language_model.layers.29.mlp.experts.125.down_proj", "model.language_model.layers.29.mlp.experts.126.down_proj", "model.language_model.layers.29.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.00037848316133022586, "dbits": 603979776 } ] }, { "idx": 60, "layers": [ "model.language_model.layers.30.self_attn.q_proj", "model.language_model.layers.30.self_attn.k_proj", "model.language_model.layers.30.self_attn.v_proj", "model.language_model.layers.30.self_attn.o_proj" ], "candidates": [ { "dkld": 0.00027989009395242015, "dbits": 18874368 } ] }, { "idx": 61, "layers": [ "model.language_model.layers.30.mlp.experts.0.gate_proj", "model.language_model.layers.30.mlp.experts.1.gate_proj", "model.language_model.layers.30.mlp.experts.2.gate_proj", "model.language_model.layers.30.mlp.experts.3.gate_proj", "model.language_model.layers.30.mlp.experts.4.gate_proj", "model.language_model.layers.30.mlp.experts.5.gate_proj", "model.language_model.layers.30.mlp.experts.6.gate_proj", "model.language_model.layers.30.mlp.experts.7.gate_proj", "model.language_model.layers.30.mlp.experts.8.gate_proj", "model.language_model.layers.30.mlp.experts.9.gate_proj", "model.language_model.layers.30.mlp.experts.10.gate_proj", "model.language_model.layers.30.mlp.experts.11.gate_proj", "model.language_model.layers.30.mlp.experts.12.gate_proj", "model.language_model.layers.30.mlp.experts.13.gate_proj", "model.language_model.layers.30.mlp.experts.14.gate_proj", "model.language_model.layers.30.mlp.experts.15.gate_proj", "model.language_model.layers.30.mlp.experts.16.gate_proj", "model.language_model.layers.30.mlp.experts.17.gate_proj", "model.language_model.layers.30.mlp.experts.18.gate_proj", "model.language_model.layers.30.mlp.experts.19.gate_proj", "model.language_model.layers.30.mlp.experts.20.gate_proj", "model.language_model.layers.30.mlp.experts.21.gate_proj", "model.language_model.layers.30.mlp.experts.22.gate_proj", "model.language_model.layers.30.mlp.experts.23.gate_proj", "model.language_model.layers.30.mlp.experts.24.gate_proj", "model.language_model.layers.30.mlp.experts.25.gate_proj", "model.language_model.layers.30.mlp.experts.26.gate_proj", "model.language_model.layers.30.mlp.experts.27.gate_proj", "model.language_model.layers.30.mlp.experts.28.gate_proj", "model.language_model.layers.30.mlp.experts.29.gate_proj", "model.language_model.layers.30.mlp.experts.30.gate_proj", "model.language_model.layers.30.mlp.experts.31.gate_proj", "model.language_model.layers.30.mlp.experts.32.gate_proj", "model.language_model.layers.30.mlp.experts.33.gate_proj", "model.language_model.layers.30.mlp.experts.34.gate_proj", "model.language_model.layers.30.mlp.experts.35.gate_proj", "model.language_model.layers.30.mlp.experts.36.gate_proj", "model.language_model.layers.30.mlp.experts.37.gate_proj", "model.language_model.layers.30.mlp.experts.38.gate_proj", "model.language_model.layers.30.mlp.experts.39.gate_proj", "model.language_model.layers.30.mlp.experts.40.gate_proj", "model.language_model.layers.30.mlp.experts.41.gate_proj", "model.language_model.layers.30.mlp.experts.42.gate_proj", "model.language_model.layers.30.mlp.experts.43.gate_proj", "model.language_model.layers.30.mlp.experts.44.gate_proj", "model.language_model.layers.30.mlp.experts.45.gate_proj", "model.language_model.layers.30.mlp.experts.46.gate_proj", "model.language_model.layers.30.mlp.experts.47.gate_proj", "model.language_model.layers.30.mlp.experts.48.gate_proj", "model.language_model.layers.30.mlp.experts.49.gate_proj", "model.language_model.layers.30.mlp.experts.50.gate_proj", "model.language_model.layers.30.mlp.experts.51.gate_proj", "model.language_model.layers.30.mlp.experts.52.gate_proj", "model.language_model.layers.30.mlp.experts.53.gate_proj", "model.language_model.layers.30.mlp.experts.54.gate_proj", "model.language_model.layers.30.mlp.experts.55.gate_proj", "model.language_model.layers.30.mlp.experts.56.gate_proj", "model.language_model.layers.30.mlp.experts.57.gate_proj", "model.language_model.layers.30.mlp.experts.58.gate_proj", "model.language_model.layers.30.mlp.experts.59.gate_proj", "model.language_model.layers.30.mlp.experts.60.gate_proj", "model.language_model.layers.30.mlp.experts.61.gate_proj", "model.language_model.layers.30.mlp.experts.62.gate_proj", "model.language_model.layers.30.mlp.experts.63.gate_proj", "model.language_model.layers.30.mlp.experts.64.gate_proj", "model.language_model.layers.30.mlp.experts.65.gate_proj", "model.language_model.layers.30.mlp.experts.66.gate_proj", "model.language_model.layers.30.mlp.experts.67.gate_proj", "model.language_model.layers.30.mlp.experts.68.gate_proj", "model.language_model.layers.30.mlp.experts.69.gate_proj", "model.language_model.layers.30.mlp.experts.70.gate_proj", "model.language_model.layers.30.mlp.experts.71.gate_proj", "model.language_model.layers.30.mlp.experts.72.gate_proj", "model.language_model.layers.30.mlp.experts.73.gate_proj", "model.language_model.layers.30.mlp.experts.74.gate_proj", "model.language_model.layers.30.mlp.experts.75.gate_proj", "model.language_model.layers.30.mlp.experts.76.gate_proj", "model.language_model.layers.30.mlp.experts.77.gate_proj", "model.language_model.layers.30.mlp.experts.78.gate_proj", "model.language_model.layers.30.mlp.experts.79.gate_proj", "model.language_model.layers.30.mlp.experts.80.gate_proj", "model.language_model.layers.30.mlp.experts.81.gate_proj", "model.language_model.layers.30.mlp.experts.82.gate_proj", "model.language_model.layers.30.mlp.experts.83.gate_proj", "model.language_model.layers.30.mlp.experts.84.gate_proj", "model.language_model.layers.30.mlp.experts.85.gate_proj", "model.language_model.layers.30.mlp.experts.86.gate_proj", "model.language_model.layers.30.mlp.experts.87.gate_proj", "model.language_model.layers.30.mlp.experts.88.gate_proj", "model.language_model.layers.30.mlp.experts.89.gate_proj", "model.language_model.layers.30.mlp.experts.90.gate_proj", "model.language_model.layers.30.mlp.experts.91.gate_proj", "model.language_model.layers.30.mlp.experts.92.gate_proj", "model.language_model.layers.30.mlp.experts.93.gate_proj", "model.language_model.layers.30.mlp.experts.94.gate_proj", "model.language_model.layers.30.mlp.experts.95.gate_proj", "model.language_model.layers.30.mlp.experts.96.gate_proj", "model.language_model.layers.30.mlp.experts.97.gate_proj", "model.language_model.layers.30.mlp.experts.98.gate_proj", "model.language_model.layers.30.mlp.experts.99.gate_proj", "model.language_model.layers.30.mlp.experts.100.gate_proj", "model.language_model.layers.30.mlp.experts.101.gate_proj", "model.language_model.layers.30.mlp.experts.102.gate_proj", "model.language_model.layers.30.mlp.experts.103.gate_proj", "model.language_model.layers.30.mlp.experts.104.gate_proj", "model.language_model.layers.30.mlp.experts.105.gate_proj", "model.language_model.layers.30.mlp.experts.106.gate_proj", "model.language_model.layers.30.mlp.experts.107.gate_proj", "model.language_model.layers.30.mlp.experts.108.gate_proj", "model.language_model.layers.30.mlp.experts.109.gate_proj", "model.language_model.layers.30.mlp.experts.110.gate_proj", "model.language_model.layers.30.mlp.experts.111.gate_proj", "model.language_model.layers.30.mlp.experts.112.gate_proj", "model.language_model.layers.30.mlp.experts.113.gate_proj", "model.language_model.layers.30.mlp.experts.114.gate_proj", "model.language_model.layers.30.mlp.experts.115.gate_proj", "model.language_model.layers.30.mlp.experts.116.gate_proj", "model.language_model.layers.30.mlp.experts.117.gate_proj", "model.language_model.layers.30.mlp.experts.118.gate_proj", "model.language_model.layers.30.mlp.experts.119.gate_proj", "model.language_model.layers.30.mlp.experts.120.gate_proj", "model.language_model.layers.30.mlp.experts.121.gate_proj", "model.language_model.layers.30.mlp.experts.122.gate_proj", "model.language_model.layers.30.mlp.experts.123.gate_proj", "model.language_model.layers.30.mlp.experts.124.gate_proj", "model.language_model.layers.30.mlp.experts.125.gate_proj", "model.language_model.layers.30.mlp.experts.126.gate_proj", "model.language_model.layers.30.mlp.experts.127.gate_proj", "model.language_model.layers.30.mlp.experts.0.up_proj", "model.language_model.layers.30.mlp.experts.1.up_proj", "model.language_model.layers.30.mlp.experts.2.up_proj", "model.language_model.layers.30.mlp.experts.3.up_proj", "model.language_model.layers.30.mlp.experts.4.up_proj", "model.language_model.layers.30.mlp.experts.5.up_proj", "model.language_model.layers.30.mlp.experts.6.up_proj", "model.language_model.layers.30.mlp.experts.7.up_proj", "model.language_model.layers.30.mlp.experts.8.up_proj", "model.language_model.layers.30.mlp.experts.9.up_proj", "model.language_model.layers.30.mlp.experts.10.up_proj", "model.language_model.layers.30.mlp.experts.11.up_proj", "model.language_model.layers.30.mlp.experts.12.up_proj", "model.language_model.layers.30.mlp.experts.13.up_proj", "model.language_model.layers.30.mlp.experts.14.up_proj", "model.language_model.layers.30.mlp.experts.15.up_proj", "model.language_model.layers.30.mlp.experts.16.up_proj", "model.language_model.layers.30.mlp.experts.17.up_proj", "model.language_model.layers.30.mlp.experts.18.up_proj", "model.language_model.layers.30.mlp.experts.19.up_proj", "model.language_model.layers.30.mlp.experts.20.up_proj", "model.language_model.layers.30.mlp.experts.21.up_proj", "model.language_model.layers.30.mlp.experts.22.up_proj", "model.language_model.layers.30.mlp.experts.23.up_proj", "model.language_model.layers.30.mlp.experts.24.up_proj", "model.language_model.layers.30.mlp.experts.25.up_proj", "model.language_model.layers.30.mlp.experts.26.up_proj", "model.language_model.layers.30.mlp.experts.27.up_proj", "model.language_model.layers.30.mlp.experts.28.up_proj", "model.language_model.layers.30.mlp.experts.29.up_proj", "model.language_model.layers.30.mlp.experts.30.up_proj", "model.language_model.layers.30.mlp.experts.31.up_proj", "model.language_model.layers.30.mlp.experts.32.up_proj", "model.language_model.layers.30.mlp.experts.33.up_proj", "model.language_model.layers.30.mlp.experts.34.up_proj", "model.language_model.layers.30.mlp.experts.35.up_proj", "model.language_model.layers.30.mlp.experts.36.up_proj", "model.language_model.layers.30.mlp.experts.37.up_proj", "model.language_model.layers.30.mlp.experts.38.up_proj", "model.language_model.layers.30.mlp.experts.39.up_proj", "model.language_model.layers.30.mlp.experts.40.up_proj", "model.language_model.layers.30.mlp.experts.41.up_proj", "model.language_model.layers.30.mlp.experts.42.up_proj", "model.language_model.layers.30.mlp.experts.43.up_proj", "model.language_model.layers.30.mlp.experts.44.up_proj", "model.language_model.layers.30.mlp.experts.45.up_proj", "model.language_model.layers.30.mlp.experts.46.up_proj", "model.language_model.layers.30.mlp.experts.47.up_proj", "model.language_model.layers.30.mlp.experts.48.up_proj", "model.language_model.layers.30.mlp.experts.49.up_proj", "model.language_model.layers.30.mlp.experts.50.up_proj", "model.language_model.layers.30.mlp.experts.51.up_proj", "model.language_model.layers.30.mlp.experts.52.up_proj", "model.language_model.layers.30.mlp.experts.53.up_proj", "model.language_model.layers.30.mlp.experts.54.up_proj", "model.language_model.layers.30.mlp.experts.55.up_proj", "model.language_model.layers.30.mlp.experts.56.up_proj", "model.language_model.layers.30.mlp.experts.57.up_proj", "model.language_model.layers.30.mlp.experts.58.up_proj", "model.language_model.layers.30.mlp.experts.59.up_proj", "model.language_model.layers.30.mlp.experts.60.up_proj", "model.language_model.layers.30.mlp.experts.61.up_proj", "model.language_model.layers.30.mlp.experts.62.up_proj", "model.language_model.layers.30.mlp.experts.63.up_proj", "model.language_model.layers.30.mlp.experts.64.up_proj", "model.language_model.layers.30.mlp.experts.65.up_proj", "model.language_model.layers.30.mlp.experts.66.up_proj", "model.language_model.layers.30.mlp.experts.67.up_proj", "model.language_model.layers.30.mlp.experts.68.up_proj", "model.language_model.layers.30.mlp.experts.69.up_proj", "model.language_model.layers.30.mlp.experts.70.up_proj", "model.language_model.layers.30.mlp.experts.71.up_proj", "model.language_model.layers.30.mlp.experts.72.up_proj", "model.language_model.layers.30.mlp.experts.73.up_proj", "model.language_model.layers.30.mlp.experts.74.up_proj", "model.language_model.layers.30.mlp.experts.75.up_proj", "model.language_model.layers.30.mlp.experts.76.up_proj", "model.language_model.layers.30.mlp.experts.77.up_proj", "model.language_model.layers.30.mlp.experts.78.up_proj", "model.language_model.layers.30.mlp.experts.79.up_proj", "model.language_model.layers.30.mlp.experts.80.up_proj", "model.language_model.layers.30.mlp.experts.81.up_proj", "model.language_model.layers.30.mlp.experts.82.up_proj", "model.language_model.layers.30.mlp.experts.83.up_proj", "model.language_model.layers.30.mlp.experts.84.up_proj", "model.language_model.layers.30.mlp.experts.85.up_proj", "model.language_model.layers.30.mlp.experts.86.up_proj", "model.language_model.layers.30.mlp.experts.87.up_proj", "model.language_model.layers.30.mlp.experts.88.up_proj", "model.language_model.layers.30.mlp.experts.89.up_proj", "model.language_model.layers.30.mlp.experts.90.up_proj", "model.language_model.layers.30.mlp.experts.91.up_proj", "model.language_model.layers.30.mlp.experts.92.up_proj", "model.language_model.layers.30.mlp.experts.93.up_proj", "model.language_model.layers.30.mlp.experts.94.up_proj", "model.language_model.layers.30.mlp.experts.95.up_proj", "model.language_model.layers.30.mlp.experts.96.up_proj", "model.language_model.layers.30.mlp.experts.97.up_proj", "model.language_model.layers.30.mlp.experts.98.up_proj", "model.language_model.layers.30.mlp.experts.99.up_proj", "model.language_model.layers.30.mlp.experts.100.up_proj", "model.language_model.layers.30.mlp.experts.101.up_proj", "model.language_model.layers.30.mlp.experts.102.up_proj", "model.language_model.layers.30.mlp.experts.103.up_proj", "model.language_model.layers.30.mlp.experts.104.up_proj", "model.language_model.layers.30.mlp.experts.105.up_proj", "model.language_model.layers.30.mlp.experts.106.up_proj", "model.language_model.layers.30.mlp.experts.107.up_proj", "model.language_model.layers.30.mlp.experts.108.up_proj", "model.language_model.layers.30.mlp.experts.109.up_proj", "model.language_model.layers.30.mlp.experts.110.up_proj", "model.language_model.layers.30.mlp.experts.111.up_proj", "model.language_model.layers.30.mlp.experts.112.up_proj", "model.language_model.layers.30.mlp.experts.113.up_proj", "model.language_model.layers.30.mlp.experts.114.up_proj", "model.language_model.layers.30.mlp.experts.115.up_proj", "model.language_model.layers.30.mlp.experts.116.up_proj", "model.language_model.layers.30.mlp.experts.117.up_proj", "model.language_model.layers.30.mlp.experts.118.up_proj", "model.language_model.layers.30.mlp.experts.119.up_proj", "model.language_model.layers.30.mlp.experts.120.up_proj", "model.language_model.layers.30.mlp.experts.121.up_proj", "model.language_model.layers.30.mlp.experts.122.up_proj", "model.language_model.layers.30.mlp.experts.123.up_proj", "model.language_model.layers.30.mlp.experts.124.up_proj", "model.language_model.layers.30.mlp.experts.125.up_proj", "model.language_model.layers.30.mlp.experts.126.up_proj", "model.language_model.layers.30.mlp.experts.127.up_proj", "model.language_model.layers.30.mlp.experts.0.down_proj", "model.language_model.layers.30.mlp.experts.1.down_proj", "model.language_model.layers.30.mlp.experts.2.down_proj", "model.language_model.layers.30.mlp.experts.3.down_proj", "model.language_model.layers.30.mlp.experts.4.down_proj", "model.language_model.layers.30.mlp.experts.5.down_proj", "model.language_model.layers.30.mlp.experts.6.down_proj", "model.language_model.layers.30.mlp.experts.7.down_proj", "model.language_model.layers.30.mlp.experts.8.down_proj", "model.language_model.layers.30.mlp.experts.9.down_proj", "model.language_model.layers.30.mlp.experts.10.down_proj", "model.language_model.layers.30.mlp.experts.11.down_proj", "model.language_model.layers.30.mlp.experts.12.down_proj", "model.language_model.layers.30.mlp.experts.13.down_proj", "model.language_model.layers.30.mlp.experts.14.down_proj", "model.language_model.layers.30.mlp.experts.15.down_proj", "model.language_model.layers.30.mlp.experts.16.down_proj", "model.language_model.layers.30.mlp.experts.17.down_proj", "model.language_model.layers.30.mlp.experts.18.down_proj", "model.language_model.layers.30.mlp.experts.19.down_proj", "model.language_model.layers.30.mlp.experts.20.down_proj", "model.language_model.layers.30.mlp.experts.21.down_proj", "model.language_model.layers.30.mlp.experts.22.down_proj", "model.language_model.layers.30.mlp.experts.23.down_proj", "model.language_model.layers.30.mlp.experts.24.down_proj", "model.language_model.layers.30.mlp.experts.25.down_proj", "model.language_model.layers.30.mlp.experts.26.down_proj", "model.language_model.layers.30.mlp.experts.27.down_proj", "model.language_model.layers.30.mlp.experts.28.down_proj", "model.language_model.layers.30.mlp.experts.29.down_proj", "model.language_model.layers.30.mlp.experts.30.down_proj", "model.language_model.layers.30.mlp.experts.31.down_proj", "model.language_model.layers.30.mlp.experts.32.down_proj", "model.language_model.layers.30.mlp.experts.33.down_proj", "model.language_model.layers.30.mlp.experts.34.down_proj", "model.language_model.layers.30.mlp.experts.35.down_proj", "model.language_model.layers.30.mlp.experts.36.down_proj", "model.language_model.layers.30.mlp.experts.37.down_proj", "model.language_model.layers.30.mlp.experts.38.down_proj", "model.language_model.layers.30.mlp.experts.39.down_proj", "model.language_model.layers.30.mlp.experts.40.down_proj", "model.language_model.layers.30.mlp.experts.41.down_proj", "model.language_model.layers.30.mlp.experts.42.down_proj", "model.language_model.layers.30.mlp.experts.43.down_proj", "model.language_model.layers.30.mlp.experts.44.down_proj", "model.language_model.layers.30.mlp.experts.45.down_proj", "model.language_model.layers.30.mlp.experts.46.down_proj", "model.language_model.layers.30.mlp.experts.47.down_proj", "model.language_model.layers.30.mlp.experts.48.down_proj", "model.language_model.layers.30.mlp.experts.49.down_proj", "model.language_model.layers.30.mlp.experts.50.down_proj", "model.language_model.layers.30.mlp.experts.51.down_proj", "model.language_model.layers.30.mlp.experts.52.down_proj", "model.language_model.layers.30.mlp.experts.53.down_proj", "model.language_model.layers.30.mlp.experts.54.down_proj", "model.language_model.layers.30.mlp.experts.55.down_proj", "model.language_model.layers.30.mlp.experts.56.down_proj", "model.language_model.layers.30.mlp.experts.57.down_proj", "model.language_model.layers.30.mlp.experts.58.down_proj", "model.language_model.layers.30.mlp.experts.59.down_proj", "model.language_model.layers.30.mlp.experts.60.down_proj", "model.language_model.layers.30.mlp.experts.61.down_proj", "model.language_model.layers.30.mlp.experts.62.down_proj", "model.language_model.layers.30.mlp.experts.63.down_proj", "model.language_model.layers.30.mlp.experts.64.down_proj", "model.language_model.layers.30.mlp.experts.65.down_proj", "model.language_model.layers.30.mlp.experts.66.down_proj", "model.language_model.layers.30.mlp.experts.67.down_proj", "model.language_model.layers.30.mlp.experts.68.down_proj", "model.language_model.layers.30.mlp.experts.69.down_proj", "model.language_model.layers.30.mlp.experts.70.down_proj", "model.language_model.layers.30.mlp.experts.71.down_proj", "model.language_model.layers.30.mlp.experts.72.down_proj", "model.language_model.layers.30.mlp.experts.73.down_proj", "model.language_model.layers.30.mlp.experts.74.down_proj", "model.language_model.layers.30.mlp.experts.75.down_proj", "model.language_model.layers.30.mlp.experts.76.down_proj", "model.language_model.layers.30.mlp.experts.77.down_proj", "model.language_model.layers.30.mlp.experts.78.down_proj", "model.language_model.layers.30.mlp.experts.79.down_proj", "model.language_model.layers.30.mlp.experts.80.down_proj", "model.language_model.layers.30.mlp.experts.81.down_proj", "model.language_model.layers.30.mlp.experts.82.down_proj", "model.language_model.layers.30.mlp.experts.83.down_proj", "model.language_model.layers.30.mlp.experts.84.down_proj", "model.language_model.layers.30.mlp.experts.85.down_proj", "model.language_model.layers.30.mlp.experts.86.down_proj", "model.language_model.layers.30.mlp.experts.87.down_proj", "model.language_model.layers.30.mlp.experts.88.down_proj", "model.language_model.layers.30.mlp.experts.89.down_proj", "model.language_model.layers.30.mlp.experts.90.down_proj", "model.language_model.layers.30.mlp.experts.91.down_proj", "model.language_model.layers.30.mlp.experts.92.down_proj", "model.language_model.layers.30.mlp.experts.93.down_proj", "model.language_model.layers.30.mlp.experts.94.down_proj", "model.language_model.layers.30.mlp.experts.95.down_proj", "model.language_model.layers.30.mlp.experts.96.down_proj", "model.language_model.layers.30.mlp.experts.97.down_proj", "model.language_model.layers.30.mlp.experts.98.down_proj", "model.language_model.layers.30.mlp.experts.99.down_proj", "model.language_model.layers.30.mlp.experts.100.down_proj", "model.language_model.layers.30.mlp.experts.101.down_proj", "model.language_model.layers.30.mlp.experts.102.down_proj", "model.language_model.layers.30.mlp.experts.103.down_proj", "model.language_model.layers.30.mlp.experts.104.down_proj", "model.language_model.layers.30.mlp.experts.105.down_proj", "model.language_model.layers.30.mlp.experts.106.down_proj", "model.language_model.layers.30.mlp.experts.107.down_proj", "model.language_model.layers.30.mlp.experts.108.down_proj", "model.language_model.layers.30.mlp.experts.109.down_proj", "model.language_model.layers.30.mlp.experts.110.down_proj", "model.language_model.layers.30.mlp.experts.111.down_proj", "model.language_model.layers.30.mlp.experts.112.down_proj", "model.language_model.layers.30.mlp.experts.113.down_proj", "model.language_model.layers.30.mlp.experts.114.down_proj", "model.language_model.layers.30.mlp.experts.115.down_proj", "model.language_model.layers.30.mlp.experts.116.down_proj", "model.language_model.layers.30.mlp.experts.117.down_proj", "model.language_model.layers.30.mlp.experts.118.down_proj", "model.language_model.layers.30.mlp.experts.119.down_proj", "model.language_model.layers.30.mlp.experts.120.down_proj", "model.language_model.layers.30.mlp.experts.121.down_proj", "model.language_model.layers.30.mlp.experts.122.down_proj", "model.language_model.layers.30.mlp.experts.123.down_proj", "model.language_model.layers.30.mlp.experts.124.down_proj", "model.language_model.layers.30.mlp.experts.125.down_proj", "model.language_model.layers.30.mlp.experts.126.down_proj", "model.language_model.layers.30.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.001199736073613164, "dbits": 603979776 } ] }, { "idx": 62, "layers": [ "model.language_model.layers.31.self_attn.q_proj", "model.language_model.layers.31.self_attn.k_proj", "model.language_model.layers.31.self_attn.v_proj", "model.language_model.layers.31.self_attn.o_proj" ], "candidates": [ { "dkld": 0.000431752949953082, "dbits": 18874368 } ] }, { "idx": 63, "layers": [ "model.language_model.layers.31.mlp.experts.0.gate_proj", "model.language_model.layers.31.mlp.experts.1.gate_proj", "model.language_model.layers.31.mlp.experts.2.gate_proj", "model.language_model.layers.31.mlp.experts.3.gate_proj", "model.language_model.layers.31.mlp.experts.4.gate_proj", "model.language_model.layers.31.mlp.experts.5.gate_proj", "model.language_model.layers.31.mlp.experts.6.gate_proj", "model.language_model.layers.31.mlp.experts.7.gate_proj", "model.language_model.layers.31.mlp.experts.8.gate_proj", "model.language_model.layers.31.mlp.experts.9.gate_proj", "model.language_model.layers.31.mlp.experts.10.gate_proj", "model.language_model.layers.31.mlp.experts.11.gate_proj", "model.language_model.layers.31.mlp.experts.12.gate_proj", "model.language_model.layers.31.mlp.experts.13.gate_proj", "model.language_model.layers.31.mlp.experts.14.gate_proj", "model.language_model.layers.31.mlp.experts.15.gate_proj", "model.language_model.layers.31.mlp.experts.16.gate_proj", "model.language_model.layers.31.mlp.experts.17.gate_proj", "model.language_model.layers.31.mlp.experts.18.gate_proj", "model.language_model.layers.31.mlp.experts.19.gate_proj", "model.language_model.layers.31.mlp.experts.20.gate_proj", "model.language_model.layers.31.mlp.experts.21.gate_proj", "model.language_model.layers.31.mlp.experts.22.gate_proj", "model.language_model.layers.31.mlp.experts.23.gate_proj", "model.language_model.layers.31.mlp.experts.24.gate_proj", "model.language_model.layers.31.mlp.experts.25.gate_proj", "model.language_model.layers.31.mlp.experts.26.gate_proj", "model.language_model.layers.31.mlp.experts.27.gate_proj", "model.language_model.layers.31.mlp.experts.28.gate_proj", "model.language_model.layers.31.mlp.experts.29.gate_proj", "model.language_model.layers.31.mlp.experts.30.gate_proj", "model.language_model.layers.31.mlp.experts.31.gate_proj", "model.language_model.layers.31.mlp.experts.32.gate_proj", "model.language_model.layers.31.mlp.experts.33.gate_proj", "model.language_model.layers.31.mlp.experts.34.gate_proj", "model.language_model.layers.31.mlp.experts.35.gate_proj", "model.language_model.layers.31.mlp.experts.36.gate_proj", "model.language_model.layers.31.mlp.experts.37.gate_proj", "model.language_model.layers.31.mlp.experts.38.gate_proj", "model.language_model.layers.31.mlp.experts.39.gate_proj", "model.language_model.layers.31.mlp.experts.40.gate_proj", "model.language_model.layers.31.mlp.experts.41.gate_proj", "model.language_model.layers.31.mlp.experts.42.gate_proj", "model.language_model.layers.31.mlp.experts.43.gate_proj", "model.language_model.layers.31.mlp.experts.44.gate_proj", "model.language_model.layers.31.mlp.experts.45.gate_proj", "model.language_model.layers.31.mlp.experts.46.gate_proj", "model.language_model.layers.31.mlp.experts.47.gate_proj", "model.language_model.layers.31.mlp.experts.48.gate_proj", "model.language_model.layers.31.mlp.experts.49.gate_proj", "model.language_model.layers.31.mlp.experts.50.gate_proj", "model.language_model.layers.31.mlp.experts.51.gate_proj", "model.language_model.layers.31.mlp.experts.52.gate_proj", "model.language_model.layers.31.mlp.experts.53.gate_proj", "model.language_model.layers.31.mlp.experts.54.gate_proj", "model.language_model.layers.31.mlp.experts.55.gate_proj", "model.language_model.layers.31.mlp.experts.56.gate_proj", "model.language_model.layers.31.mlp.experts.57.gate_proj", "model.language_model.layers.31.mlp.experts.58.gate_proj", "model.language_model.layers.31.mlp.experts.59.gate_proj", "model.language_model.layers.31.mlp.experts.60.gate_proj", "model.language_model.layers.31.mlp.experts.61.gate_proj", "model.language_model.layers.31.mlp.experts.62.gate_proj", "model.language_model.layers.31.mlp.experts.63.gate_proj", "model.language_model.layers.31.mlp.experts.64.gate_proj", "model.language_model.layers.31.mlp.experts.65.gate_proj", "model.language_model.layers.31.mlp.experts.66.gate_proj", "model.language_model.layers.31.mlp.experts.67.gate_proj", "model.language_model.layers.31.mlp.experts.68.gate_proj", "model.language_model.layers.31.mlp.experts.69.gate_proj", "model.language_model.layers.31.mlp.experts.70.gate_proj", "model.language_model.layers.31.mlp.experts.71.gate_proj", "model.language_model.layers.31.mlp.experts.72.gate_proj", "model.language_model.layers.31.mlp.experts.73.gate_proj", "model.language_model.layers.31.mlp.experts.74.gate_proj", "model.language_model.layers.31.mlp.experts.75.gate_proj", "model.language_model.layers.31.mlp.experts.76.gate_proj", "model.language_model.layers.31.mlp.experts.77.gate_proj", "model.language_model.layers.31.mlp.experts.78.gate_proj", "model.language_model.layers.31.mlp.experts.79.gate_proj", "model.language_model.layers.31.mlp.experts.80.gate_proj", "model.language_model.layers.31.mlp.experts.81.gate_proj", "model.language_model.layers.31.mlp.experts.82.gate_proj", "model.language_model.layers.31.mlp.experts.83.gate_proj", "model.language_model.layers.31.mlp.experts.84.gate_proj", "model.language_model.layers.31.mlp.experts.85.gate_proj", "model.language_model.layers.31.mlp.experts.86.gate_proj", "model.language_model.layers.31.mlp.experts.87.gate_proj", "model.language_model.layers.31.mlp.experts.88.gate_proj", "model.language_model.layers.31.mlp.experts.89.gate_proj", "model.language_model.layers.31.mlp.experts.90.gate_proj", "model.language_model.layers.31.mlp.experts.91.gate_proj", "model.language_model.layers.31.mlp.experts.92.gate_proj", "model.language_model.layers.31.mlp.experts.93.gate_proj", "model.language_model.layers.31.mlp.experts.94.gate_proj", "model.language_model.layers.31.mlp.experts.95.gate_proj", "model.language_model.layers.31.mlp.experts.96.gate_proj", "model.language_model.layers.31.mlp.experts.97.gate_proj", "model.language_model.layers.31.mlp.experts.98.gate_proj", "model.language_model.layers.31.mlp.experts.99.gate_proj", "model.language_model.layers.31.mlp.experts.100.gate_proj", "model.language_model.layers.31.mlp.experts.101.gate_proj", "model.language_model.layers.31.mlp.experts.102.gate_proj", "model.language_model.layers.31.mlp.experts.103.gate_proj", "model.language_model.layers.31.mlp.experts.104.gate_proj", "model.language_model.layers.31.mlp.experts.105.gate_proj", "model.language_model.layers.31.mlp.experts.106.gate_proj", "model.language_model.layers.31.mlp.experts.107.gate_proj", "model.language_model.layers.31.mlp.experts.108.gate_proj", "model.language_model.layers.31.mlp.experts.109.gate_proj", "model.language_model.layers.31.mlp.experts.110.gate_proj", "model.language_model.layers.31.mlp.experts.111.gate_proj", "model.language_model.layers.31.mlp.experts.112.gate_proj", "model.language_model.layers.31.mlp.experts.113.gate_proj", "model.language_model.layers.31.mlp.experts.114.gate_proj", "model.language_model.layers.31.mlp.experts.115.gate_proj", "model.language_model.layers.31.mlp.experts.116.gate_proj", "model.language_model.layers.31.mlp.experts.117.gate_proj", "model.language_model.layers.31.mlp.experts.118.gate_proj", "model.language_model.layers.31.mlp.experts.119.gate_proj", "model.language_model.layers.31.mlp.experts.120.gate_proj", "model.language_model.layers.31.mlp.experts.121.gate_proj", "model.language_model.layers.31.mlp.experts.122.gate_proj", "model.language_model.layers.31.mlp.experts.123.gate_proj", "model.language_model.layers.31.mlp.experts.124.gate_proj", "model.language_model.layers.31.mlp.experts.125.gate_proj", "model.language_model.layers.31.mlp.experts.126.gate_proj", "model.language_model.layers.31.mlp.experts.127.gate_proj", "model.language_model.layers.31.mlp.experts.0.up_proj", "model.language_model.layers.31.mlp.experts.1.up_proj", "model.language_model.layers.31.mlp.experts.2.up_proj", "model.language_model.layers.31.mlp.experts.3.up_proj", "model.language_model.layers.31.mlp.experts.4.up_proj", "model.language_model.layers.31.mlp.experts.5.up_proj", "model.language_model.layers.31.mlp.experts.6.up_proj", "model.language_model.layers.31.mlp.experts.7.up_proj", "model.language_model.layers.31.mlp.experts.8.up_proj", "model.language_model.layers.31.mlp.experts.9.up_proj", "model.language_model.layers.31.mlp.experts.10.up_proj", "model.language_model.layers.31.mlp.experts.11.up_proj", "model.language_model.layers.31.mlp.experts.12.up_proj", "model.language_model.layers.31.mlp.experts.13.up_proj", "model.language_model.layers.31.mlp.experts.14.up_proj", "model.language_model.layers.31.mlp.experts.15.up_proj", "model.language_model.layers.31.mlp.experts.16.up_proj", "model.language_model.layers.31.mlp.experts.17.up_proj", "model.language_model.layers.31.mlp.experts.18.up_proj", "model.language_model.layers.31.mlp.experts.19.up_proj", "model.language_model.layers.31.mlp.experts.20.up_proj", "model.language_model.layers.31.mlp.experts.21.up_proj", "model.language_model.layers.31.mlp.experts.22.up_proj", "model.language_model.layers.31.mlp.experts.23.up_proj", "model.language_model.layers.31.mlp.experts.24.up_proj", "model.language_model.layers.31.mlp.experts.25.up_proj", "model.language_model.layers.31.mlp.experts.26.up_proj", "model.language_model.layers.31.mlp.experts.27.up_proj", "model.language_model.layers.31.mlp.experts.28.up_proj", "model.language_model.layers.31.mlp.experts.29.up_proj", "model.language_model.layers.31.mlp.experts.30.up_proj", "model.language_model.layers.31.mlp.experts.31.up_proj", "model.language_model.layers.31.mlp.experts.32.up_proj", "model.language_model.layers.31.mlp.experts.33.up_proj", "model.language_model.layers.31.mlp.experts.34.up_proj", "model.language_model.layers.31.mlp.experts.35.up_proj", "model.language_model.layers.31.mlp.experts.36.up_proj", "model.language_model.layers.31.mlp.experts.37.up_proj", "model.language_model.layers.31.mlp.experts.38.up_proj", "model.language_model.layers.31.mlp.experts.39.up_proj", "model.language_model.layers.31.mlp.experts.40.up_proj", "model.language_model.layers.31.mlp.experts.41.up_proj", "model.language_model.layers.31.mlp.experts.42.up_proj", "model.language_model.layers.31.mlp.experts.43.up_proj", "model.language_model.layers.31.mlp.experts.44.up_proj", "model.language_model.layers.31.mlp.experts.45.up_proj", "model.language_model.layers.31.mlp.experts.46.up_proj", "model.language_model.layers.31.mlp.experts.47.up_proj", "model.language_model.layers.31.mlp.experts.48.up_proj", "model.language_model.layers.31.mlp.experts.49.up_proj", "model.language_model.layers.31.mlp.experts.50.up_proj", "model.language_model.layers.31.mlp.experts.51.up_proj", "model.language_model.layers.31.mlp.experts.52.up_proj", "model.language_model.layers.31.mlp.experts.53.up_proj", "model.language_model.layers.31.mlp.experts.54.up_proj", "model.language_model.layers.31.mlp.experts.55.up_proj", "model.language_model.layers.31.mlp.experts.56.up_proj", "model.language_model.layers.31.mlp.experts.57.up_proj", "model.language_model.layers.31.mlp.experts.58.up_proj", "model.language_model.layers.31.mlp.experts.59.up_proj", "model.language_model.layers.31.mlp.experts.60.up_proj", "model.language_model.layers.31.mlp.experts.61.up_proj", "model.language_model.layers.31.mlp.experts.62.up_proj", "model.language_model.layers.31.mlp.experts.63.up_proj", "model.language_model.layers.31.mlp.experts.64.up_proj", "model.language_model.layers.31.mlp.experts.65.up_proj", "model.language_model.layers.31.mlp.experts.66.up_proj", "model.language_model.layers.31.mlp.experts.67.up_proj", "model.language_model.layers.31.mlp.experts.68.up_proj", "model.language_model.layers.31.mlp.experts.69.up_proj", "model.language_model.layers.31.mlp.experts.70.up_proj", "model.language_model.layers.31.mlp.experts.71.up_proj", "model.language_model.layers.31.mlp.experts.72.up_proj", "model.language_model.layers.31.mlp.experts.73.up_proj", "model.language_model.layers.31.mlp.experts.74.up_proj", "model.language_model.layers.31.mlp.experts.75.up_proj", "model.language_model.layers.31.mlp.experts.76.up_proj", "model.language_model.layers.31.mlp.experts.77.up_proj", "model.language_model.layers.31.mlp.experts.78.up_proj", "model.language_model.layers.31.mlp.experts.79.up_proj", "model.language_model.layers.31.mlp.experts.80.up_proj", "model.language_model.layers.31.mlp.experts.81.up_proj", "model.language_model.layers.31.mlp.experts.82.up_proj", "model.language_model.layers.31.mlp.experts.83.up_proj", "model.language_model.layers.31.mlp.experts.84.up_proj", "model.language_model.layers.31.mlp.experts.85.up_proj", "model.language_model.layers.31.mlp.experts.86.up_proj", "model.language_model.layers.31.mlp.experts.87.up_proj", "model.language_model.layers.31.mlp.experts.88.up_proj", "model.language_model.layers.31.mlp.experts.89.up_proj", "model.language_model.layers.31.mlp.experts.90.up_proj", "model.language_model.layers.31.mlp.experts.91.up_proj", "model.language_model.layers.31.mlp.experts.92.up_proj", "model.language_model.layers.31.mlp.experts.93.up_proj", "model.language_model.layers.31.mlp.experts.94.up_proj", "model.language_model.layers.31.mlp.experts.95.up_proj", "model.language_model.layers.31.mlp.experts.96.up_proj", "model.language_model.layers.31.mlp.experts.97.up_proj", "model.language_model.layers.31.mlp.experts.98.up_proj", "model.language_model.layers.31.mlp.experts.99.up_proj", "model.language_model.layers.31.mlp.experts.100.up_proj", "model.language_model.layers.31.mlp.experts.101.up_proj", "model.language_model.layers.31.mlp.experts.102.up_proj", "model.language_model.layers.31.mlp.experts.103.up_proj", "model.language_model.layers.31.mlp.experts.104.up_proj", "model.language_model.layers.31.mlp.experts.105.up_proj", "model.language_model.layers.31.mlp.experts.106.up_proj", "model.language_model.layers.31.mlp.experts.107.up_proj", "model.language_model.layers.31.mlp.experts.108.up_proj", "model.language_model.layers.31.mlp.experts.109.up_proj", "model.language_model.layers.31.mlp.experts.110.up_proj", "model.language_model.layers.31.mlp.experts.111.up_proj", "model.language_model.layers.31.mlp.experts.112.up_proj", "model.language_model.layers.31.mlp.experts.113.up_proj", "model.language_model.layers.31.mlp.experts.114.up_proj", "model.language_model.layers.31.mlp.experts.115.up_proj", "model.language_model.layers.31.mlp.experts.116.up_proj", "model.language_model.layers.31.mlp.experts.117.up_proj", "model.language_model.layers.31.mlp.experts.118.up_proj", "model.language_model.layers.31.mlp.experts.119.up_proj", "model.language_model.layers.31.mlp.experts.120.up_proj", "model.language_model.layers.31.mlp.experts.121.up_proj", "model.language_model.layers.31.mlp.experts.122.up_proj", "model.language_model.layers.31.mlp.experts.123.up_proj", "model.language_model.layers.31.mlp.experts.124.up_proj", "model.language_model.layers.31.mlp.experts.125.up_proj", "model.language_model.layers.31.mlp.experts.126.up_proj", "model.language_model.layers.31.mlp.experts.127.up_proj", "model.language_model.layers.31.mlp.experts.0.down_proj", "model.language_model.layers.31.mlp.experts.1.down_proj", "model.language_model.layers.31.mlp.experts.2.down_proj", "model.language_model.layers.31.mlp.experts.3.down_proj", "model.language_model.layers.31.mlp.experts.4.down_proj", "model.language_model.layers.31.mlp.experts.5.down_proj", "model.language_model.layers.31.mlp.experts.6.down_proj", "model.language_model.layers.31.mlp.experts.7.down_proj", "model.language_model.layers.31.mlp.experts.8.down_proj", "model.language_model.layers.31.mlp.experts.9.down_proj", "model.language_model.layers.31.mlp.experts.10.down_proj", "model.language_model.layers.31.mlp.experts.11.down_proj", "model.language_model.layers.31.mlp.experts.12.down_proj", "model.language_model.layers.31.mlp.experts.13.down_proj", "model.language_model.layers.31.mlp.experts.14.down_proj", "model.language_model.layers.31.mlp.experts.15.down_proj", "model.language_model.layers.31.mlp.experts.16.down_proj", "model.language_model.layers.31.mlp.experts.17.down_proj", "model.language_model.layers.31.mlp.experts.18.down_proj", "model.language_model.layers.31.mlp.experts.19.down_proj", "model.language_model.layers.31.mlp.experts.20.down_proj", "model.language_model.layers.31.mlp.experts.21.down_proj", "model.language_model.layers.31.mlp.experts.22.down_proj", "model.language_model.layers.31.mlp.experts.23.down_proj", "model.language_model.layers.31.mlp.experts.24.down_proj", "model.language_model.layers.31.mlp.experts.25.down_proj", "model.language_model.layers.31.mlp.experts.26.down_proj", "model.language_model.layers.31.mlp.experts.27.down_proj", "model.language_model.layers.31.mlp.experts.28.down_proj", "model.language_model.layers.31.mlp.experts.29.down_proj", "model.language_model.layers.31.mlp.experts.30.down_proj", "model.language_model.layers.31.mlp.experts.31.down_proj", "model.language_model.layers.31.mlp.experts.32.down_proj", "model.language_model.layers.31.mlp.experts.33.down_proj", "model.language_model.layers.31.mlp.experts.34.down_proj", "model.language_model.layers.31.mlp.experts.35.down_proj", "model.language_model.layers.31.mlp.experts.36.down_proj", "model.language_model.layers.31.mlp.experts.37.down_proj", "model.language_model.layers.31.mlp.experts.38.down_proj", "model.language_model.layers.31.mlp.experts.39.down_proj", "model.language_model.layers.31.mlp.experts.40.down_proj", "model.language_model.layers.31.mlp.experts.41.down_proj", "model.language_model.layers.31.mlp.experts.42.down_proj", "model.language_model.layers.31.mlp.experts.43.down_proj", "model.language_model.layers.31.mlp.experts.44.down_proj", "model.language_model.layers.31.mlp.experts.45.down_proj", "model.language_model.layers.31.mlp.experts.46.down_proj", "model.language_model.layers.31.mlp.experts.47.down_proj", "model.language_model.layers.31.mlp.experts.48.down_proj", "model.language_model.layers.31.mlp.experts.49.down_proj", "model.language_model.layers.31.mlp.experts.50.down_proj", "model.language_model.layers.31.mlp.experts.51.down_proj", "model.language_model.layers.31.mlp.experts.52.down_proj", "model.language_model.layers.31.mlp.experts.53.down_proj", "model.language_model.layers.31.mlp.experts.54.down_proj", "model.language_model.layers.31.mlp.experts.55.down_proj", "model.language_model.layers.31.mlp.experts.56.down_proj", "model.language_model.layers.31.mlp.experts.57.down_proj", "model.language_model.layers.31.mlp.experts.58.down_proj", "model.language_model.layers.31.mlp.experts.59.down_proj", "model.language_model.layers.31.mlp.experts.60.down_proj", "model.language_model.layers.31.mlp.experts.61.down_proj", "model.language_model.layers.31.mlp.experts.62.down_proj", "model.language_model.layers.31.mlp.experts.63.down_proj", "model.language_model.layers.31.mlp.experts.64.down_proj", "model.language_model.layers.31.mlp.experts.65.down_proj", "model.language_model.layers.31.mlp.experts.66.down_proj", "model.language_model.layers.31.mlp.experts.67.down_proj", "model.language_model.layers.31.mlp.experts.68.down_proj", "model.language_model.layers.31.mlp.experts.69.down_proj", "model.language_model.layers.31.mlp.experts.70.down_proj", "model.language_model.layers.31.mlp.experts.71.down_proj", "model.language_model.layers.31.mlp.experts.72.down_proj", "model.language_model.layers.31.mlp.experts.73.down_proj", "model.language_model.layers.31.mlp.experts.74.down_proj", "model.language_model.layers.31.mlp.experts.75.down_proj", "model.language_model.layers.31.mlp.experts.76.down_proj", "model.language_model.layers.31.mlp.experts.77.down_proj", "model.language_model.layers.31.mlp.experts.78.down_proj", "model.language_model.layers.31.mlp.experts.79.down_proj", "model.language_model.layers.31.mlp.experts.80.down_proj", "model.language_model.layers.31.mlp.experts.81.down_proj", "model.language_model.layers.31.mlp.experts.82.down_proj", "model.language_model.layers.31.mlp.experts.83.down_proj", "model.language_model.layers.31.mlp.experts.84.down_proj", "model.language_model.layers.31.mlp.experts.85.down_proj", "model.language_model.layers.31.mlp.experts.86.down_proj", "model.language_model.layers.31.mlp.experts.87.down_proj", "model.language_model.layers.31.mlp.experts.88.down_proj", "model.language_model.layers.31.mlp.experts.89.down_proj", "model.language_model.layers.31.mlp.experts.90.down_proj", "model.language_model.layers.31.mlp.experts.91.down_proj", "model.language_model.layers.31.mlp.experts.92.down_proj", "model.language_model.layers.31.mlp.experts.93.down_proj", "model.language_model.layers.31.mlp.experts.94.down_proj", "model.language_model.layers.31.mlp.experts.95.down_proj", "model.language_model.layers.31.mlp.experts.96.down_proj", "model.language_model.layers.31.mlp.experts.97.down_proj", "model.language_model.layers.31.mlp.experts.98.down_proj", "model.language_model.layers.31.mlp.experts.99.down_proj", "model.language_model.layers.31.mlp.experts.100.down_proj", "model.language_model.layers.31.mlp.experts.101.down_proj", "model.language_model.layers.31.mlp.experts.102.down_proj", "model.language_model.layers.31.mlp.experts.103.down_proj", "model.language_model.layers.31.mlp.experts.104.down_proj", "model.language_model.layers.31.mlp.experts.105.down_proj", "model.language_model.layers.31.mlp.experts.106.down_proj", "model.language_model.layers.31.mlp.experts.107.down_proj", "model.language_model.layers.31.mlp.experts.108.down_proj", "model.language_model.layers.31.mlp.experts.109.down_proj", "model.language_model.layers.31.mlp.experts.110.down_proj", "model.language_model.layers.31.mlp.experts.111.down_proj", "model.language_model.layers.31.mlp.experts.112.down_proj", "model.language_model.layers.31.mlp.experts.113.down_proj", "model.language_model.layers.31.mlp.experts.114.down_proj", "model.language_model.layers.31.mlp.experts.115.down_proj", "model.language_model.layers.31.mlp.experts.116.down_proj", "model.language_model.layers.31.mlp.experts.117.down_proj", "model.language_model.layers.31.mlp.experts.118.down_proj", "model.language_model.layers.31.mlp.experts.119.down_proj", "model.language_model.layers.31.mlp.experts.120.down_proj", "model.language_model.layers.31.mlp.experts.121.down_proj", "model.language_model.layers.31.mlp.experts.122.down_proj", "model.language_model.layers.31.mlp.experts.123.down_proj", "model.language_model.layers.31.mlp.experts.124.down_proj", "model.language_model.layers.31.mlp.experts.125.down_proj", "model.language_model.layers.31.mlp.experts.126.down_proj", "model.language_model.layers.31.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.0011562000960111674, "dbits": 603979776 } ] }, { "idx": 64, "layers": [ "model.language_model.layers.32.self_attn.q_proj", "model.language_model.layers.32.self_attn.k_proj", "model.language_model.layers.32.self_attn.v_proj", "model.language_model.layers.32.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0010029899887740584, "dbits": 18874368 } ] }, { "idx": 65, "layers": [ "model.language_model.layers.32.mlp.experts.0.gate_proj", "model.language_model.layers.32.mlp.experts.1.gate_proj", "model.language_model.layers.32.mlp.experts.2.gate_proj", "model.language_model.layers.32.mlp.experts.3.gate_proj", "model.language_model.layers.32.mlp.experts.4.gate_proj", "model.language_model.layers.32.mlp.experts.5.gate_proj", "model.language_model.layers.32.mlp.experts.6.gate_proj", "model.language_model.layers.32.mlp.experts.7.gate_proj", "model.language_model.layers.32.mlp.experts.8.gate_proj", "model.language_model.layers.32.mlp.experts.9.gate_proj", "model.language_model.layers.32.mlp.experts.10.gate_proj", "model.language_model.layers.32.mlp.experts.11.gate_proj", "model.language_model.layers.32.mlp.experts.12.gate_proj", "model.language_model.layers.32.mlp.experts.13.gate_proj", "model.language_model.layers.32.mlp.experts.14.gate_proj", "model.language_model.layers.32.mlp.experts.15.gate_proj", "model.language_model.layers.32.mlp.experts.16.gate_proj", "model.language_model.layers.32.mlp.experts.17.gate_proj", "model.language_model.layers.32.mlp.experts.18.gate_proj", "model.language_model.layers.32.mlp.experts.19.gate_proj", "model.language_model.layers.32.mlp.experts.20.gate_proj", "model.language_model.layers.32.mlp.experts.21.gate_proj", "model.language_model.layers.32.mlp.experts.22.gate_proj", "model.language_model.layers.32.mlp.experts.23.gate_proj", "model.language_model.layers.32.mlp.experts.24.gate_proj", "model.language_model.layers.32.mlp.experts.25.gate_proj", "model.language_model.layers.32.mlp.experts.26.gate_proj", "model.language_model.layers.32.mlp.experts.27.gate_proj", "model.language_model.layers.32.mlp.experts.28.gate_proj", "model.language_model.layers.32.mlp.experts.29.gate_proj", "model.language_model.layers.32.mlp.experts.30.gate_proj", "model.language_model.layers.32.mlp.experts.31.gate_proj", "model.language_model.layers.32.mlp.experts.32.gate_proj", "model.language_model.layers.32.mlp.experts.33.gate_proj", "model.language_model.layers.32.mlp.experts.34.gate_proj", "model.language_model.layers.32.mlp.experts.35.gate_proj", "model.language_model.layers.32.mlp.experts.36.gate_proj", "model.language_model.layers.32.mlp.experts.37.gate_proj", "model.language_model.layers.32.mlp.experts.38.gate_proj", "model.language_model.layers.32.mlp.experts.39.gate_proj", "model.language_model.layers.32.mlp.experts.40.gate_proj", "model.language_model.layers.32.mlp.experts.41.gate_proj", "model.language_model.layers.32.mlp.experts.42.gate_proj", "model.language_model.layers.32.mlp.experts.43.gate_proj", "model.language_model.layers.32.mlp.experts.44.gate_proj", "model.language_model.layers.32.mlp.experts.45.gate_proj", "model.language_model.layers.32.mlp.experts.46.gate_proj", "model.language_model.layers.32.mlp.experts.47.gate_proj", "model.language_model.layers.32.mlp.experts.48.gate_proj", "model.language_model.layers.32.mlp.experts.49.gate_proj", "model.language_model.layers.32.mlp.experts.50.gate_proj", "model.language_model.layers.32.mlp.experts.51.gate_proj", "model.language_model.layers.32.mlp.experts.52.gate_proj", "model.language_model.layers.32.mlp.experts.53.gate_proj", "model.language_model.layers.32.mlp.experts.54.gate_proj", "model.language_model.layers.32.mlp.experts.55.gate_proj", "model.language_model.layers.32.mlp.experts.56.gate_proj", "model.language_model.layers.32.mlp.experts.57.gate_proj", "model.language_model.layers.32.mlp.experts.58.gate_proj", "model.language_model.layers.32.mlp.experts.59.gate_proj", "model.language_model.layers.32.mlp.experts.60.gate_proj", "model.language_model.layers.32.mlp.experts.61.gate_proj", "model.language_model.layers.32.mlp.experts.62.gate_proj", "model.language_model.layers.32.mlp.experts.63.gate_proj", "model.language_model.layers.32.mlp.experts.64.gate_proj", "model.language_model.layers.32.mlp.experts.65.gate_proj", "model.language_model.layers.32.mlp.experts.66.gate_proj", "model.language_model.layers.32.mlp.experts.67.gate_proj", "model.language_model.layers.32.mlp.experts.68.gate_proj", "model.language_model.layers.32.mlp.experts.69.gate_proj", "model.language_model.layers.32.mlp.experts.70.gate_proj", "model.language_model.layers.32.mlp.experts.71.gate_proj", "model.language_model.layers.32.mlp.experts.72.gate_proj", "model.language_model.layers.32.mlp.experts.73.gate_proj", "model.language_model.layers.32.mlp.experts.74.gate_proj", "model.language_model.layers.32.mlp.experts.75.gate_proj", "model.language_model.layers.32.mlp.experts.76.gate_proj", "model.language_model.layers.32.mlp.experts.77.gate_proj", "model.language_model.layers.32.mlp.experts.78.gate_proj", "model.language_model.layers.32.mlp.experts.79.gate_proj", "model.language_model.layers.32.mlp.experts.80.gate_proj", "model.language_model.layers.32.mlp.experts.81.gate_proj", "model.language_model.layers.32.mlp.experts.82.gate_proj", "model.language_model.layers.32.mlp.experts.83.gate_proj", "model.language_model.layers.32.mlp.experts.84.gate_proj", "model.language_model.layers.32.mlp.experts.85.gate_proj", "model.language_model.layers.32.mlp.experts.86.gate_proj", "model.language_model.layers.32.mlp.experts.87.gate_proj", "model.language_model.layers.32.mlp.experts.88.gate_proj", "model.language_model.layers.32.mlp.experts.89.gate_proj", "model.language_model.layers.32.mlp.experts.90.gate_proj", "model.language_model.layers.32.mlp.experts.91.gate_proj", "model.language_model.layers.32.mlp.experts.92.gate_proj", "model.language_model.layers.32.mlp.experts.93.gate_proj", "model.language_model.layers.32.mlp.experts.94.gate_proj", "model.language_model.layers.32.mlp.experts.95.gate_proj", "model.language_model.layers.32.mlp.experts.96.gate_proj", "model.language_model.layers.32.mlp.experts.97.gate_proj", "model.language_model.layers.32.mlp.experts.98.gate_proj", "model.language_model.layers.32.mlp.experts.99.gate_proj", "model.language_model.layers.32.mlp.experts.100.gate_proj", "model.language_model.layers.32.mlp.experts.101.gate_proj", "model.language_model.layers.32.mlp.experts.102.gate_proj", "model.language_model.layers.32.mlp.experts.103.gate_proj", "model.language_model.layers.32.mlp.experts.104.gate_proj", "model.language_model.layers.32.mlp.experts.105.gate_proj", "model.language_model.layers.32.mlp.experts.106.gate_proj", "model.language_model.layers.32.mlp.experts.107.gate_proj", "model.language_model.layers.32.mlp.experts.108.gate_proj", "model.language_model.layers.32.mlp.experts.109.gate_proj", "model.language_model.layers.32.mlp.experts.110.gate_proj", "model.language_model.layers.32.mlp.experts.111.gate_proj", "model.language_model.layers.32.mlp.experts.112.gate_proj", "model.language_model.layers.32.mlp.experts.113.gate_proj", "model.language_model.layers.32.mlp.experts.114.gate_proj", "model.language_model.layers.32.mlp.experts.115.gate_proj", "model.language_model.layers.32.mlp.experts.116.gate_proj", "model.language_model.layers.32.mlp.experts.117.gate_proj", "model.language_model.layers.32.mlp.experts.118.gate_proj", "model.language_model.layers.32.mlp.experts.119.gate_proj", "model.language_model.layers.32.mlp.experts.120.gate_proj", "model.language_model.layers.32.mlp.experts.121.gate_proj", "model.language_model.layers.32.mlp.experts.122.gate_proj", "model.language_model.layers.32.mlp.experts.123.gate_proj", "model.language_model.layers.32.mlp.experts.124.gate_proj", "model.language_model.layers.32.mlp.experts.125.gate_proj", "model.language_model.layers.32.mlp.experts.126.gate_proj", "model.language_model.layers.32.mlp.experts.127.gate_proj", "model.language_model.layers.32.mlp.experts.0.up_proj", "model.language_model.layers.32.mlp.experts.1.up_proj", "model.language_model.layers.32.mlp.experts.2.up_proj", "model.language_model.layers.32.mlp.experts.3.up_proj", "model.language_model.layers.32.mlp.experts.4.up_proj", "model.language_model.layers.32.mlp.experts.5.up_proj", "model.language_model.layers.32.mlp.experts.6.up_proj", "model.language_model.layers.32.mlp.experts.7.up_proj", "model.language_model.layers.32.mlp.experts.8.up_proj", "model.language_model.layers.32.mlp.experts.9.up_proj", "model.language_model.layers.32.mlp.experts.10.up_proj", "model.language_model.layers.32.mlp.experts.11.up_proj", "model.language_model.layers.32.mlp.experts.12.up_proj", "model.language_model.layers.32.mlp.experts.13.up_proj", "model.language_model.layers.32.mlp.experts.14.up_proj", "model.language_model.layers.32.mlp.experts.15.up_proj", "model.language_model.layers.32.mlp.experts.16.up_proj", "model.language_model.layers.32.mlp.experts.17.up_proj", "model.language_model.layers.32.mlp.experts.18.up_proj", "model.language_model.layers.32.mlp.experts.19.up_proj", "model.language_model.layers.32.mlp.experts.20.up_proj", "model.language_model.layers.32.mlp.experts.21.up_proj", "model.language_model.layers.32.mlp.experts.22.up_proj", "model.language_model.layers.32.mlp.experts.23.up_proj", "model.language_model.layers.32.mlp.experts.24.up_proj", "model.language_model.layers.32.mlp.experts.25.up_proj", "model.language_model.layers.32.mlp.experts.26.up_proj", "model.language_model.layers.32.mlp.experts.27.up_proj", "model.language_model.layers.32.mlp.experts.28.up_proj", "model.language_model.layers.32.mlp.experts.29.up_proj", "model.language_model.layers.32.mlp.experts.30.up_proj", "model.language_model.layers.32.mlp.experts.31.up_proj", "model.language_model.layers.32.mlp.experts.32.up_proj", "model.language_model.layers.32.mlp.experts.33.up_proj", "model.language_model.layers.32.mlp.experts.34.up_proj", "model.language_model.layers.32.mlp.experts.35.up_proj", "model.language_model.layers.32.mlp.experts.36.up_proj", "model.language_model.layers.32.mlp.experts.37.up_proj", "model.language_model.layers.32.mlp.experts.38.up_proj", "model.language_model.layers.32.mlp.experts.39.up_proj", "model.language_model.layers.32.mlp.experts.40.up_proj", "model.language_model.layers.32.mlp.experts.41.up_proj", "model.language_model.layers.32.mlp.experts.42.up_proj", "model.language_model.layers.32.mlp.experts.43.up_proj", "model.language_model.layers.32.mlp.experts.44.up_proj", "model.language_model.layers.32.mlp.experts.45.up_proj", "model.language_model.layers.32.mlp.experts.46.up_proj", "model.language_model.layers.32.mlp.experts.47.up_proj", "model.language_model.layers.32.mlp.experts.48.up_proj", "model.language_model.layers.32.mlp.experts.49.up_proj", "model.language_model.layers.32.mlp.experts.50.up_proj", "model.language_model.layers.32.mlp.experts.51.up_proj", "model.language_model.layers.32.mlp.experts.52.up_proj", "model.language_model.layers.32.mlp.experts.53.up_proj", "model.language_model.layers.32.mlp.experts.54.up_proj", "model.language_model.layers.32.mlp.experts.55.up_proj", "model.language_model.layers.32.mlp.experts.56.up_proj", "model.language_model.layers.32.mlp.experts.57.up_proj", "model.language_model.layers.32.mlp.experts.58.up_proj", "model.language_model.layers.32.mlp.experts.59.up_proj", "model.language_model.layers.32.mlp.experts.60.up_proj", "model.language_model.layers.32.mlp.experts.61.up_proj", "model.language_model.layers.32.mlp.experts.62.up_proj", "model.language_model.layers.32.mlp.experts.63.up_proj", "model.language_model.layers.32.mlp.experts.64.up_proj", "model.language_model.layers.32.mlp.experts.65.up_proj", "model.language_model.layers.32.mlp.experts.66.up_proj", "model.language_model.layers.32.mlp.experts.67.up_proj", "model.language_model.layers.32.mlp.experts.68.up_proj", "model.language_model.layers.32.mlp.experts.69.up_proj", "model.language_model.layers.32.mlp.experts.70.up_proj", "model.language_model.layers.32.mlp.experts.71.up_proj", "model.language_model.layers.32.mlp.experts.72.up_proj", "model.language_model.layers.32.mlp.experts.73.up_proj", "model.language_model.layers.32.mlp.experts.74.up_proj", "model.language_model.layers.32.mlp.experts.75.up_proj", "model.language_model.layers.32.mlp.experts.76.up_proj", "model.language_model.layers.32.mlp.experts.77.up_proj", "model.language_model.layers.32.mlp.experts.78.up_proj", "model.language_model.layers.32.mlp.experts.79.up_proj", "model.language_model.layers.32.mlp.experts.80.up_proj", "model.language_model.layers.32.mlp.experts.81.up_proj", "model.language_model.layers.32.mlp.experts.82.up_proj", "model.language_model.layers.32.mlp.experts.83.up_proj", "model.language_model.layers.32.mlp.experts.84.up_proj", "model.language_model.layers.32.mlp.experts.85.up_proj", "model.language_model.layers.32.mlp.experts.86.up_proj", "model.language_model.layers.32.mlp.experts.87.up_proj", "model.language_model.layers.32.mlp.experts.88.up_proj", "model.language_model.layers.32.mlp.experts.89.up_proj", "model.language_model.layers.32.mlp.experts.90.up_proj", "model.language_model.layers.32.mlp.experts.91.up_proj", "model.language_model.layers.32.mlp.experts.92.up_proj", "model.language_model.layers.32.mlp.experts.93.up_proj", "model.language_model.layers.32.mlp.experts.94.up_proj", "model.language_model.layers.32.mlp.experts.95.up_proj", "model.language_model.layers.32.mlp.experts.96.up_proj", "model.language_model.layers.32.mlp.experts.97.up_proj", "model.language_model.layers.32.mlp.experts.98.up_proj", "model.language_model.layers.32.mlp.experts.99.up_proj", "model.language_model.layers.32.mlp.experts.100.up_proj", "model.language_model.layers.32.mlp.experts.101.up_proj", "model.language_model.layers.32.mlp.experts.102.up_proj", "model.language_model.layers.32.mlp.experts.103.up_proj", "model.language_model.layers.32.mlp.experts.104.up_proj", "model.language_model.layers.32.mlp.experts.105.up_proj", "model.language_model.layers.32.mlp.experts.106.up_proj", "model.language_model.layers.32.mlp.experts.107.up_proj", "model.language_model.layers.32.mlp.experts.108.up_proj", "model.language_model.layers.32.mlp.experts.109.up_proj", "model.language_model.layers.32.mlp.experts.110.up_proj", "model.language_model.layers.32.mlp.experts.111.up_proj", "model.language_model.layers.32.mlp.experts.112.up_proj", "model.language_model.layers.32.mlp.experts.113.up_proj", "model.language_model.layers.32.mlp.experts.114.up_proj", "model.language_model.layers.32.mlp.experts.115.up_proj", "model.language_model.layers.32.mlp.experts.116.up_proj", "model.language_model.layers.32.mlp.experts.117.up_proj", "model.language_model.layers.32.mlp.experts.118.up_proj", "model.language_model.layers.32.mlp.experts.119.up_proj", "model.language_model.layers.32.mlp.experts.120.up_proj", "model.language_model.layers.32.mlp.experts.121.up_proj", "model.language_model.layers.32.mlp.experts.122.up_proj", "model.language_model.layers.32.mlp.experts.123.up_proj", "model.language_model.layers.32.mlp.experts.124.up_proj", "model.language_model.layers.32.mlp.experts.125.up_proj", "model.language_model.layers.32.mlp.experts.126.up_proj", "model.language_model.layers.32.mlp.experts.127.up_proj", "model.language_model.layers.32.mlp.experts.0.down_proj", "model.language_model.layers.32.mlp.experts.1.down_proj", "model.language_model.layers.32.mlp.experts.2.down_proj", "model.language_model.layers.32.mlp.experts.3.down_proj", "model.language_model.layers.32.mlp.experts.4.down_proj", "model.language_model.layers.32.mlp.experts.5.down_proj", "model.language_model.layers.32.mlp.experts.6.down_proj", "model.language_model.layers.32.mlp.experts.7.down_proj", "model.language_model.layers.32.mlp.experts.8.down_proj", "model.language_model.layers.32.mlp.experts.9.down_proj", "model.language_model.layers.32.mlp.experts.10.down_proj", "model.language_model.layers.32.mlp.experts.11.down_proj", "model.language_model.layers.32.mlp.experts.12.down_proj", "model.language_model.layers.32.mlp.experts.13.down_proj", "model.language_model.layers.32.mlp.experts.14.down_proj", "model.language_model.layers.32.mlp.experts.15.down_proj", "model.language_model.layers.32.mlp.experts.16.down_proj", "model.language_model.layers.32.mlp.experts.17.down_proj", "model.language_model.layers.32.mlp.experts.18.down_proj", "model.language_model.layers.32.mlp.experts.19.down_proj", "model.language_model.layers.32.mlp.experts.20.down_proj", "model.language_model.layers.32.mlp.experts.21.down_proj", "model.language_model.layers.32.mlp.experts.22.down_proj", "model.language_model.layers.32.mlp.experts.23.down_proj", "model.language_model.layers.32.mlp.experts.24.down_proj", "model.language_model.layers.32.mlp.experts.25.down_proj", "model.language_model.layers.32.mlp.experts.26.down_proj", "model.language_model.layers.32.mlp.experts.27.down_proj", "model.language_model.layers.32.mlp.experts.28.down_proj", "model.language_model.layers.32.mlp.experts.29.down_proj", "model.language_model.layers.32.mlp.experts.30.down_proj", "model.language_model.layers.32.mlp.experts.31.down_proj", "model.language_model.layers.32.mlp.experts.32.down_proj", "model.language_model.layers.32.mlp.experts.33.down_proj", "model.language_model.layers.32.mlp.experts.34.down_proj", "model.language_model.layers.32.mlp.experts.35.down_proj", "model.language_model.layers.32.mlp.experts.36.down_proj", "model.language_model.layers.32.mlp.experts.37.down_proj", "model.language_model.layers.32.mlp.experts.38.down_proj", "model.language_model.layers.32.mlp.experts.39.down_proj", "model.language_model.layers.32.mlp.experts.40.down_proj", "model.language_model.layers.32.mlp.experts.41.down_proj", "model.language_model.layers.32.mlp.experts.42.down_proj", "model.language_model.layers.32.mlp.experts.43.down_proj", "model.language_model.layers.32.mlp.experts.44.down_proj", "model.language_model.layers.32.mlp.experts.45.down_proj", "model.language_model.layers.32.mlp.experts.46.down_proj", "model.language_model.layers.32.mlp.experts.47.down_proj", "model.language_model.layers.32.mlp.experts.48.down_proj", "model.language_model.layers.32.mlp.experts.49.down_proj", "model.language_model.layers.32.mlp.experts.50.down_proj", "model.language_model.layers.32.mlp.experts.51.down_proj", "model.language_model.layers.32.mlp.experts.52.down_proj", "model.language_model.layers.32.mlp.experts.53.down_proj", "model.language_model.layers.32.mlp.experts.54.down_proj", "model.language_model.layers.32.mlp.experts.55.down_proj", "model.language_model.layers.32.mlp.experts.56.down_proj", "model.language_model.layers.32.mlp.experts.57.down_proj", "model.language_model.layers.32.mlp.experts.58.down_proj", "model.language_model.layers.32.mlp.experts.59.down_proj", "model.language_model.layers.32.mlp.experts.60.down_proj", "model.language_model.layers.32.mlp.experts.61.down_proj", "model.language_model.layers.32.mlp.experts.62.down_proj", "model.language_model.layers.32.mlp.experts.63.down_proj", "model.language_model.layers.32.mlp.experts.64.down_proj", "model.language_model.layers.32.mlp.experts.65.down_proj", "model.language_model.layers.32.mlp.experts.66.down_proj", "model.language_model.layers.32.mlp.experts.67.down_proj", "model.language_model.layers.32.mlp.experts.68.down_proj", "model.language_model.layers.32.mlp.experts.69.down_proj", "model.language_model.layers.32.mlp.experts.70.down_proj", "model.language_model.layers.32.mlp.experts.71.down_proj", "model.language_model.layers.32.mlp.experts.72.down_proj", "model.language_model.layers.32.mlp.experts.73.down_proj", "model.language_model.layers.32.mlp.experts.74.down_proj", "model.language_model.layers.32.mlp.experts.75.down_proj", "model.language_model.layers.32.mlp.experts.76.down_proj", "model.language_model.layers.32.mlp.experts.77.down_proj", "model.language_model.layers.32.mlp.experts.78.down_proj", "model.language_model.layers.32.mlp.experts.79.down_proj", "model.language_model.layers.32.mlp.experts.80.down_proj", "model.language_model.layers.32.mlp.experts.81.down_proj", "model.language_model.layers.32.mlp.experts.82.down_proj", "model.language_model.layers.32.mlp.experts.83.down_proj", "model.language_model.layers.32.mlp.experts.84.down_proj", "model.language_model.layers.32.mlp.experts.85.down_proj", "model.language_model.layers.32.mlp.experts.86.down_proj", "model.language_model.layers.32.mlp.experts.87.down_proj", "model.language_model.layers.32.mlp.experts.88.down_proj", "model.language_model.layers.32.mlp.experts.89.down_proj", "model.language_model.layers.32.mlp.experts.90.down_proj", "model.language_model.layers.32.mlp.experts.91.down_proj", "model.language_model.layers.32.mlp.experts.92.down_proj", "model.language_model.layers.32.mlp.experts.93.down_proj", "model.language_model.layers.32.mlp.experts.94.down_proj", "model.language_model.layers.32.mlp.experts.95.down_proj", "model.language_model.layers.32.mlp.experts.96.down_proj", "model.language_model.layers.32.mlp.experts.97.down_proj", "model.language_model.layers.32.mlp.experts.98.down_proj", "model.language_model.layers.32.mlp.experts.99.down_proj", "model.language_model.layers.32.mlp.experts.100.down_proj", "model.language_model.layers.32.mlp.experts.101.down_proj", "model.language_model.layers.32.mlp.experts.102.down_proj", "model.language_model.layers.32.mlp.experts.103.down_proj", "model.language_model.layers.32.mlp.experts.104.down_proj", "model.language_model.layers.32.mlp.experts.105.down_proj", "model.language_model.layers.32.mlp.experts.106.down_proj", "model.language_model.layers.32.mlp.experts.107.down_proj", "model.language_model.layers.32.mlp.experts.108.down_proj", "model.language_model.layers.32.mlp.experts.109.down_proj", "model.language_model.layers.32.mlp.experts.110.down_proj", "model.language_model.layers.32.mlp.experts.111.down_proj", "model.language_model.layers.32.mlp.experts.112.down_proj", "model.language_model.layers.32.mlp.experts.113.down_proj", "model.language_model.layers.32.mlp.experts.114.down_proj", "model.language_model.layers.32.mlp.experts.115.down_proj", "model.language_model.layers.32.mlp.experts.116.down_proj", "model.language_model.layers.32.mlp.experts.117.down_proj", "model.language_model.layers.32.mlp.experts.118.down_proj", "model.language_model.layers.32.mlp.experts.119.down_proj", "model.language_model.layers.32.mlp.experts.120.down_proj", "model.language_model.layers.32.mlp.experts.121.down_proj", "model.language_model.layers.32.mlp.experts.122.down_proj", "model.language_model.layers.32.mlp.experts.123.down_proj", "model.language_model.layers.32.mlp.experts.124.down_proj", "model.language_model.layers.32.mlp.experts.125.down_proj", "model.language_model.layers.32.mlp.experts.126.down_proj", "model.language_model.layers.32.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.0012587582692503985, "dbits": 603979776 } ] }, { "idx": 66, "layers": [ "model.language_model.layers.33.self_attn.q_proj", "model.language_model.layers.33.self_attn.k_proj", "model.language_model.layers.33.self_attn.v_proj", "model.language_model.layers.33.self_attn.o_proj" ], "candidates": [ { "dkld": -0.00022627571597695628, "dbits": 18874368 } ] }, { "idx": 67, "layers": [ "model.language_model.layers.33.mlp.experts.0.gate_proj", "model.language_model.layers.33.mlp.experts.1.gate_proj", "model.language_model.layers.33.mlp.experts.2.gate_proj", "model.language_model.layers.33.mlp.experts.3.gate_proj", "model.language_model.layers.33.mlp.experts.4.gate_proj", "model.language_model.layers.33.mlp.experts.5.gate_proj", "model.language_model.layers.33.mlp.experts.6.gate_proj", "model.language_model.layers.33.mlp.experts.7.gate_proj", "model.language_model.layers.33.mlp.experts.8.gate_proj", "model.language_model.layers.33.mlp.experts.9.gate_proj", "model.language_model.layers.33.mlp.experts.10.gate_proj", "model.language_model.layers.33.mlp.experts.11.gate_proj", "model.language_model.layers.33.mlp.experts.12.gate_proj", "model.language_model.layers.33.mlp.experts.13.gate_proj", "model.language_model.layers.33.mlp.experts.14.gate_proj", "model.language_model.layers.33.mlp.experts.15.gate_proj", "model.language_model.layers.33.mlp.experts.16.gate_proj", "model.language_model.layers.33.mlp.experts.17.gate_proj", "model.language_model.layers.33.mlp.experts.18.gate_proj", "model.language_model.layers.33.mlp.experts.19.gate_proj", "model.language_model.layers.33.mlp.experts.20.gate_proj", "model.language_model.layers.33.mlp.experts.21.gate_proj", "model.language_model.layers.33.mlp.experts.22.gate_proj", "model.language_model.layers.33.mlp.experts.23.gate_proj", "model.language_model.layers.33.mlp.experts.24.gate_proj", "model.language_model.layers.33.mlp.experts.25.gate_proj", "model.language_model.layers.33.mlp.experts.26.gate_proj", "model.language_model.layers.33.mlp.experts.27.gate_proj", "model.language_model.layers.33.mlp.experts.28.gate_proj", "model.language_model.layers.33.mlp.experts.29.gate_proj", "model.language_model.layers.33.mlp.experts.30.gate_proj", "model.language_model.layers.33.mlp.experts.31.gate_proj", "model.language_model.layers.33.mlp.experts.32.gate_proj", "model.language_model.layers.33.mlp.experts.33.gate_proj", "model.language_model.layers.33.mlp.experts.34.gate_proj", "model.language_model.layers.33.mlp.experts.35.gate_proj", "model.language_model.layers.33.mlp.experts.36.gate_proj", "model.language_model.layers.33.mlp.experts.37.gate_proj", "model.language_model.layers.33.mlp.experts.38.gate_proj", "model.language_model.layers.33.mlp.experts.39.gate_proj", "model.language_model.layers.33.mlp.experts.40.gate_proj", "model.language_model.layers.33.mlp.experts.41.gate_proj", "model.language_model.layers.33.mlp.experts.42.gate_proj", "model.language_model.layers.33.mlp.experts.43.gate_proj", "model.language_model.layers.33.mlp.experts.44.gate_proj", "model.language_model.layers.33.mlp.experts.45.gate_proj", "model.language_model.layers.33.mlp.experts.46.gate_proj", "model.language_model.layers.33.mlp.experts.47.gate_proj", "model.language_model.layers.33.mlp.experts.48.gate_proj", "model.language_model.layers.33.mlp.experts.49.gate_proj", "model.language_model.layers.33.mlp.experts.50.gate_proj", "model.language_model.layers.33.mlp.experts.51.gate_proj", "model.language_model.layers.33.mlp.experts.52.gate_proj", "model.language_model.layers.33.mlp.experts.53.gate_proj", "model.language_model.layers.33.mlp.experts.54.gate_proj", "model.language_model.layers.33.mlp.experts.55.gate_proj", "model.language_model.layers.33.mlp.experts.56.gate_proj", "model.language_model.layers.33.mlp.experts.57.gate_proj", "model.language_model.layers.33.mlp.experts.58.gate_proj", "model.language_model.layers.33.mlp.experts.59.gate_proj", "model.language_model.layers.33.mlp.experts.60.gate_proj", "model.language_model.layers.33.mlp.experts.61.gate_proj", "model.language_model.layers.33.mlp.experts.62.gate_proj", "model.language_model.layers.33.mlp.experts.63.gate_proj", "model.language_model.layers.33.mlp.experts.64.gate_proj", "model.language_model.layers.33.mlp.experts.65.gate_proj", "model.language_model.layers.33.mlp.experts.66.gate_proj", "model.language_model.layers.33.mlp.experts.67.gate_proj", "model.language_model.layers.33.mlp.experts.68.gate_proj", "model.language_model.layers.33.mlp.experts.69.gate_proj", "model.language_model.layers.33.mlp.experts.70.gate_proj", "model.language_model.layers.33.mlp.experts.71.gate_proj", "model.language_model.layers.33.mlp.experts.72.gate_proj", "model.language_model.layers.33.mlp.experts.73.gate_proj", "model.language_model.layers.33.mlp.experts.74.gate_proj", "model.language_model.layers.33.mlp.experts.75.gate_proj", "model.language_model.layers.33.mlp.experts.76.gate_proj", "model.language_model.layers.33.mlp.experts.77.gate_proj", "model.language_model.layers.33.mlp.experts.78.gate_proj", "model.language_model.layers.33.mlp.experts.79.gate_proj", "model.language_model.layers.33.mlp.experts.80.gate_proj", "model.language_model.layers.33.mlp.experts.81.gate_proj", "model.language_model.layers.33.mlp.experts.82.gate_proj", "model.language_model.layers.33.mlp.experts.83.gate_proj", "model.language_model.layers.33.mlp.experts.84.gate_proj", "model.language_model.layers.33.mlp.experts.85.gate_proj", "model.language_model.layers.33.mlp.experts.86.gate_proj", "model.language_model.layers.33.mlp.experts.87.gate_proj", "model.language_model.layers.33.mlp.experts.88.gate_proj", "model.language_model.layers.33.mlp.experts.89.gate_proj", "model.language_model.layers.33.mlp.experts.90.gate_proj", "model.language_model.layers.33.mlp.experts.91.gate_proj", "model.language_model.layers.33.mlp.experts.92.gate_proj", "model.language_model.layers.33.mlp.experts.93.gate_proj", "model.language_model.layers.33.mlp.experts.94.gate_proj", "model.language_model.layers.33.mlp.experts.95.gate_proj", "model.language_model.layers.33.mlp.experts.96.gate_proj", "model.language_model.layers.33.mlp.experts.97.gate_proj", "model.language_model.layers.33.mlp.experts.98.gate_proj", "model.language_model.layers.33.mlp.experts.99.gate_proj", "model.language_model.layers.33.mlp.experts.100.gate_proj", "model.language_model.layers.33.mlp.experts.101.gate_proj", "model.language_model.layers.33.mlp.experts.102.gate_proj", "model.language_model.layers.33.mlp.experts.103.gate_proj", "model.language_model.layers.33.mlp.experts.104.gate_proj", "model.language_model.layers.33.mlp.experts.105.gate_proj", "model.language_model.layers.33.mlp.experts.106.gate_proj", "model.language_model.layers.33.mlp.experts.107.gate_proj", "model.language_model.layers.33.mlp.experts.108.gate_proj", "model.language_model.layers.33.mlp.experts.109.gate_proj", "model.language_model.layers.33.mlp.experts.110.gate_proj", "model.language_model.layers.33.mlp.experts.111.gate_proj", "model.language_model.layers.33.mlp.experts.112.gate_proj", "model.language_model.layers.33.mlp.experts.113.gate_proj", "model.language_model.layers.33.mlp.experts.114.gate_proj", "model.language_model.layers.33.mlp.experts.115.gate_proj", "model.language_model.layers.33.mlp.experts.116.gate_proj", "model.language_model.layers.33.mlp.experts.117.gate_proj", "model.language_model.layers.33.mlp.experts.118.gate_proj", "model.language_model.layers.33.mlp.experts.119.gate_proj", "model.language_model.layers.33.mlp.experts.120.gate_proj", "model.language_model.layers.33.mlp.experts.121.gate_proj", "model.language_model.layers.33.mlp.experts.122.gate_proj", "model.language_model.layers.33.mlp.experts.123.gate_proj", "model.language_model.layers.33.mlp.experts.124.gate_proj", "model.language_model.layers.33.mlp.experts.125.gate_proj", "model.language_model.layers.33.mlp.experts.126.gate_proj", "model.language_model.layers.33.mlp.experts.127.gate_proj", "model.language_model.layers.33.mlp.experts.0.up_proj", "model.language_model.layers.33.mlp.experts.1.up_proj", "model.language_model.layers.33.mlp.experts.2.up_proj", "model.language_model.layers.33.mlp.experts.3.up_proj", "model.language_model.layers.33.mlp.experts.4.up_proj", "model.language_model.layers.33.mlp.experts.5.up_proj", "model.language_model.layers.33.mlp.experts.6.up_proj", "model.language_model.layers.33.mlp.experts.7.up_proj", "model.language_model.layers.33.mlp.experts.8.up_proj", "model.language_model.layers.33.mlp.experts.9.up_proj", "model.language_model.layers.33.mlp.experts.10.up_proj", "model.language_model.layers.33.mlp.experts.11.up_proj", "model.language_model.layers.33.mlp.experts.12.up_proj", "model.language_model.layers.33.mlp.experts.13.up_proj", "model.language_model.layers.33.mlp.experts.14.up_proj", "model.language_model.layers.33.mlp.experts.15.up_proj", "model.language_model.layers.33.mlp.experts.16.up_proj", "model.language_model.layers.33.mlp.experts.17.up_proj", "model.language_model.layers.33.mlp.experts.18.up_proj", "model.language_model.layers.33.mlp.experts.19.up_proj", "model.language_model.layers.33.mlp.experts.20.up_proj", "model.language_model.layers.33.mlp.experts.21.up_proj", "model.language_model.layers.33.mlp.experts.22.up_proj", "model.language_model.layers.33.mlp.experts.23.up_proj", "model.language_model.layers.33.mlp.experts.24.up_proj", "model.language_model.layers.33.mlp.experts.25.up_proj", "model.language_model.layers.33.mlp.experts.26.up_proj", "model.language_model.layers.33.mlp.experts.27.up_proj", "model.language_model.layers.33.mlp.experts.28.up_proj", "model.language_model.layers.33.mlp.experts.29.up_proj", "model.language_model.layers.33.mlp.experts.30.up_proj", "model.language_model.layers.33.mlp.experts.31.up_proj", "model.language_model.layers.33.mlp.experts.32.up_proj", "model.language_model.layers.33.mlp.experts.33.up_proj", "model.language_model.layers.33.mlp.experts.34.up_proj", "model.language_model.layers.33.mlp.experts.35.up_proj", "model.language_model.layers.33.mlp.experts.36.up_proj", "model.language_model.layers.33.mlp.experts.37.up_proj", "model.language_model.layers.33.mlp.experts.38.up_proj", "model.language_model.layers.33.mlp.experts.39.up_proj", "model.language_model.layers.33.mlp.experts.40.up_proj", "model.language_model.layers.33.mlp.experts.41.up_proj", "model.language_model.layers.33.mlp.experts.42.up_proj", "model.language_model.layers.33.mlp.experts.43.up_proj", "model.language_model.layers.33.mlp.experts.44.up_proj", "model.language_model.layers.33.mlp.experts.45.up_proj", "model.language_model.layers.33.mlp.experts.46.up_proj", "model.language_model.layers.33.mlp.experts.47.up_proj", "model.language_model.layers.33.mlp.experts.48.up_proj", "model.language_model.layers.33.mlp.experts.49.up_proj", "model.language_model.layers.33.mlp.experts.50.up_proj", "model.language_model.layers.33.mlp.experts.51.up_proj", "model.language_model.layers.33.mlp.experts.52.up_proj", "model.language_model.layers.33.mlp.experts.53.up_proj", "model.language_model.layers.33.mlp.experts.54.up_proj", "model.language_model.layers.33.mlp.experts.55.up_proj", "model.language_model.layers.33.mlp.experts.56.up_proj", "model.language_model.layers.33.mlp.experts.57.up_proj", "model.language_model.layers.33.mlp.experts.58.up_proj", "model.language_model.layers.33.mlp.experts.59.up_proj", "model.language_model.layers.33.mlp.experts.60.up_proj", "model.language_model.layers.33.mlp.experts.61.up_proj", "model.language_model.layers.33.mlp.experts.62.up_proj", "model.language_model.layers.33.mlp.experts.63.up_proj", "model.language_model.layers.33.mlp.experts.64.up_proj", "model.language_model.layers.33.mlp.experts.65.up_proj", "model.language_model.layers.33.mlp.experts.66.up_proj", "model.language_model.layers.33.mlp.experts.67.up_proj", "model.language_model.layers.33.mlp.experts.68.up_proj", "model.language_model.layers.33.mlp.experts.69.up_proj", "model.language_model.layers.33.mlp.experts.70.up_proj", "model.language_model.layers.33.mlp.experts.71.up_proj", "model.language_model.layers.33.mlp.experts.72.up_proj", "model.language_model.layers.33.mlp.experts.73.up_proj", "model.language_model.layers.33.mlp.experts.74.up_proj", "model.language_model.layers.33.mlp.experts.75.up_proj", "model.language_model.layers.33.mlp.experts.76.up_proj", "model.language_model.layers.33.mlp.experts.77.up_proj", "model.language_model.layers.33.mlp.experts.78.up_proj", "model.language_model.layers.33.mlp.experts.79.up_proj", "model.language_model.layers.33.mlp.experts.80.up_proj", "model.language_model.layers.33.mlp.experts.81.up_proj", "model.language_model.layers.33.mlp.experts.82.up_proj", "model.language_model.layers.33.mlp.experts.83.up_proj", "model.language_model.layers.33.mlp.experts.84.up_proj", "model.language_model.layers.33.mlp.experts.85.up_proj", "model.language_model.layers.33.mlp.experts.86.up_proj", "model.language_model.layers.33.mlp.experts.87.up_proj", "model.language_model.layers.33.mlp.experts.88.up_proj", "model.language_model.layers.33.mlp.experts.89.up_proj", "model.language_model.layers.33.mlp.experts.90.up_proj", "model.language_model.layers.33.mlp.experts.91.up_proj", "model.language_model.layers.33.mlp.experts.92.up_proj", "model.language_model.layers.33.mlp.experts.93.up_proj", "model.language_model.layers.33.mlp.experts.94.up_proj", "model.language_model.layers.33.mlp.experts.95.up_proj", "model.language_model.layers.33.mlp.experts.96.up_proj", "model.language_model.layers.33.mlp.experts.97.up_proj", "model.language_model.layers.33.mlp.experts.98.up_proj", "model.language_model.layers.33.mlp.experts.99.up_proj", "model.language_model.layers.33.mlp.experts.100.up_proj", "model.language_model.layers.33.mlp.experts.101.up_proj", "model.language_model.layers.33.mlp.experts.102.up_proj", "model.language_model.layers.33.mlp.experts.103.up_proj", "model.language_model.layers.33.mlp.experts.104.up_proj", "model.language_model.layers.33.mlp.experts.105.up_proj", "model.language_model.layers.33.mlp.experts.106.up_proj", "model.language_model.layers.33.mlp.experts.107.up_proj", "model.language_model.layers.33.mlp.experts.108.up_proj", "model.language_model.layers.33.mlp.experts.109.up_proj", "model.language_model.layers.33.mlp.experts.110.up_proj", "model.language_model.layers.33.mlp.experts.111.up_proj", "model.language_model.layers.33.mlp.experts.112.up_proj", "model.language_model.layers.33.mlp.experts.113.up_proj", "model.language_model.layers.33.mlp.experts.114.up_proj", "model.language_model.layers.33.mlp.experts.115.up_proj", "model.language_model.layers.33.mlp.experts.116.up_proj", "model.language_model.layers.33.mlp.experts.117.up_proj", "model.language_model.layers.33.mlp.experts.118.up_proj", "model.language_model.layers.33.mlp.experts.119.up_proj", "model.language_model.layers.33.mlp.experts.120.up_proj", "model.language_model.layers.33.mlp.experts.121.up_proj", "model.language_model.layers.33.mlp.experts.122.up_proj", "model.language_model.layers.33.mlp.experts.123.up_proj", "model.language_model.layers.33.mlp.experts.124.up_proj", "model.language_model.layers.33.mlp.experts.125.up_proj", "model.language_model.layers.33.mlp.experts.126.up_proj", "model.language_model.layers.33.mlp.experts.127.up_proj", "model.language_model.layers.33.mlp.experts.0.down_proj", "model.language_model.layers.33.mlp.experts.1.down_proj", "model.language_model.layers.33.mlp.experts.2.down_proj", "model.language_model.layers.33.mlp.experts.3.down_proj", "model.language_model.layers.33.mlp.experts.4.down_proj", "model.language_model.layers.33.mlp.experts.5.down_proj", "model.language_model.layers.33.mlp.experts.6.down_proj", "model.language_model.layers.33.mlp.experts.7.down_proj", "model.language_model.layers.33.mlp.experts.8.down_proj", "model.language_model.layers.33.mlp.experts.9.down_proj", "model.language_model.layers.33.mlp.experts.10.down_proj", "model.language_model.layers.33.mlp.experts.11.down_proj", "model.language_model.layers.33.mlp.experts.12.down_proj", "model.language_model.layers.33.mlp.experts.13.down_proj", "model.language_model.layers.33.mlp.experts.14.down_proj", "model.language_model.layers.33.mlp.experts.15.down_proj", "model.language_model.layers.33.mlp.experts.16.down_proj", "model.language_model.layers.33.mlp.experts.17.down_proj", "model.language_model.layers.33.mlp.experts.18.down_proj", "model.language_model.layers.33.mlp.experts.19.down_proj", "model.language_model.layers.33.mlp.experts.20.down_proj", "model.language_model.layers.33.mlp.experts.21.down_proj", "model.language_model.layers.33.mlp.experts.22.down_proj", "model.language_model.layers.33.mlp.experts.23.down_proj", "model.language_model.layers.33.mlp.experts.24.down_proj", "model.language_model.layers.33.mlp.experts.25.down_proj", "model.language_model.layers.33.mlp.experts.26.down_proj", "model.language_model.layers.33.mlp.experts.27.down_proj", "model.language_model.layers.33.mlp.experts.28.down_proj", "model.language_model.layers.33.mlp.experts.29.down_proj", "model.language_model.layers.33.mlp.experts.30.down_proj", "model.language_model.layers.33.mlp.experts.31.down_proj", "model.language_model.layers.33.mlp.experts.32.down_proj", "model.language_model.layers.33.mlp.experts.33.down_proj", "model.language_model.layers.33.mlp.experts.34.down_proj", "model.language_model.layers.33.mlp.experts.35.down_proj", "model.language_model.layers.33.mlp.experts.36.down_proj", "model.language_model.layers.33.mlp.experts.37.down_proj", "model.language_model.layers.33.mlp.experts.38.down_proj", "model.language_model.layers.33.mlp.experts.39.down_proj", "model.language_model.layers.33.mlp.experts.40.down_proj", "model.language_model.layers.33.mlp.experts.41.down_proj", "model.language_model.layers.33.mlp.experts.42.down_proj", "model.language_model.layers.33.mlp.experts.43.down_proj", "model.language_model.layers.33.mlp.experts.44.down_proj", "model.language_model.layers.33.mlp.experts.45.down_proj", "model.language_model.layers.33.mlp.experts.46.down_proj", "model.language_model.layers.33.mlp.experts.47.down_proj", "model.language_model.layers.33.mlp.experts.48.down_proj", "model.language_model.layers.33.mlp.experts.49.down_proj", "model.language_model.layers.33.mlp.experts.50.down_proj", "model.language_model.layers.33.mlp.experts.51.down_proj", "model.language_model.layers.33.mlp.experts.52.down_proj", "model.language_model.layers.33.mlp.experts.53.down_proj", "model.language_model.layers.33.mlp.experts.54.down_proj", "model.language_model.layers.33.mlp.experts.55.down_proj", "model.language_model.layers.33.mlp.experts.56.down_proj", "model.language_model.layers.33.mlp.experts.57.down_proj", "model.language_model.layers.33.mlp.experts.58.down_proj", "model.language_model.layers.33.mlp.experts.59.down_proj", "model.language_model.layers.33.mlp.experts.60.down_proj", "model.language_model.layers.33.mlp.experts.61.down_proj", "model.language_model.layers.33.mlp.experts.62.down_proj", "model.language_model.layers.33.mlp.experts.63.down_proj", "model.language_model.layers.33.mlp.experts.64.down_proj", "model.language_model.layers.33.mlp.experts.65.down_proj", "model.language_model.layers.33.mlp.experts.66.down_proj", "model.language_model.layers.33.mlp.experts.67.down_proj", "model.language_model.layers.33.mlp.experts.68.down_proj", "model.language_model.layers.33.mlp.experts.69.down_proj", "model.language_model.layers.33.mlp.experts.70.down_proj", "model.language_model.layers.33.mlp.experts.71.down_proj", "model.language_model.layers.33.mlp.experts.72.down_proj", "model.language_model.layers.33.mlp.experts.73.down_proj", "model.language_model.layers.33.mlp.experts.74.down_proj", "model.language_model.layers.33.mlp.experts.75.down_proj", "model.language_model.layers.33.mlp.experts.76.down_proj", "model.language_model.layers.33.mlp.experts.77.down_proj", "model.language_model.layers.33.mlp.experts.78.down_proj", "model.language_model.layers.33.mlp.experts.79.down_proj", "model.language_model.layers.33.mlp.experts.80.down_proj", "model.language_model.layers.33.mlp.experts.81.down_proj", "model.language_model.layers.33.mlp.experts.82.down_proj", "model.language_model.layers.33.mlp.experts.83.down_proj", "model.language_model.layers.33.mlp.experts.84.down_proj", "model.language_model.layers.33.mlp.experts.85.down_proj", "model.language_model.layers.33.mlp.experts.86.down_proj", "model.language_model.layers.33.mlp.experts.87.down_proj", "model.language_model.layers.33.mlp.experts.88.down_proj", "model.language_model.layers.33.mlp.experts.89.down_proj", "model.language_model.layers.33.mlp.experts.90.down_proj", "model.language_model.layers.33.mlp.experts.91.down_proj", "model.language_model.layers.33.mlp.experts.92.down_proj", "model.language_model.layers.33.mlp.experts.93.down_proj", "model.language_model.layers.33.mlp.experts.94.down_proj", "model.language_model.layers.33.mlp.experts.95.down_proj", "model.language_model.layers.33.mlp.experts.96.down_proj", "model.language_model.layers.33.mlp.experts.97.down_proj", "model.language_model.layers.33.mlp.experts.98.down_proj", "model.language_model.layers.33.mlp.experts.99.down_proj", "model.language_model.layers.33.mlp.experts.100.down_proj", "model.language_model.layers.33.mlp.experts.101.down_proj", "model.language_model.layers.33.mlp.experts.102.down_proj", "model.language_model.layers.33.mlp.experts.103.down_proj", "model.language_model.layers.33.mlp.experts.104.down_proj", "model.language_model.layers.33.mlp.experts.105.down_proj", "model.language_model.layers.33.mlp.experts.106.down_proj", "model.language_model.layers.33.mlp.experts.107.down_proj", "model.language_model.layers.33.mlp.experts.108.down_proj", "model.language_model.layers.33.mlp.experts.109.down_proj", "model.language_model.layers.33.mlp.experts.110.down_proj", "model.language_model.layers.33.mlp.experts.111.down_proj", "model.language_model.layers.33.mlp.experts.112.down_proj", "model.language_model.layers.33.mlp.experts.113.down_proj", "model.language_model.layers.33.mlp.experts.114.down_proj", "model.language_model.layers.33.mlp.experts.115.down_proj", "model.language_model.layers.33.mlp.experts.116.down_proj", "model.language_model.layers.33.mlp.experts.117.down_proj", "model.language_model.layers.33.mlp.experts.118.down_proj", "model.language_model.layers.33.mlp.experts.119.down_proj", "model.language_model.layers.33.mlp.experts.120.down_proj", "model.language_model.layers.33.mlp.experts.121.down_proj", "model.language_model.layers.33.mlp.experts.122.down_proj", "model.language_model.layers.33.mlp.experts.123.down_proj", "model.language_model.layers.33.mlp.experts.124.down_proj", "model.language_model.layers.33.mlp.experts.125.down_proj", "model.language_model.layers.33.mlp.experts.126.down_proj", "model.language_model.layers.33.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.0014060846529901083, "dbits": 603979776 } ] }, { "idx": 68, "layers": [ "model.language_model.layers.34.self_attn.q_proj", "model.language_model.layers.34.self_attn.k_proj", "model.language_model.layers.34.self_attn.v_proj", "model.language_model.layers.34.self_attn.o_proj" ], "candidates": [ { "dkld": 2.099759876728058e-05, "dbits": 18874368 } ] }, { "idx": 69, "layers": [ "model.language_model.layers.34.mlp.experts.0.gate_proj", "model.language_model.layers.34.mlp.experts.1.gate_proj", "model.language_model.layers.34.mlp.experts.2.gate_proj", "model.language_model.layers.34.mlp.experts.3.gate_proj", "model.language_model.layers.34.mlp.experts.4.gate_proj", "model.language_model.layers.34.mlp.experts.5.gate_proj", "model.language_model.layers.34.mlp.experts.6.gate_proj", "model.language_model.layers.34.mlp.experts.7.gate_proj", "model.language_model.layers.34.mlp.experts.8.gate_proj", "model.language_model.layers.34.mlp.experts.9.gate_proj", "model.language_model.layers.34.mlp.experts.10.gate_proj", "model.language_model.layers.34.mlp.experts.11.gate_proj", "model.language_model.layers.34.mlp.experts.12.gate_proj", "model.language_model.layers.34.mlp.experts.13.gate_proj", "model.language_model.layers.34.mlp.experts.14.gate_proj", "model.language_model.layers.34.mlp.experts.15.gate_proj", "model.language_model.layers.34.mlp.experts.16.gate_proj", "model.language_model.layers.34.mlp.experts.17.gate_proj", "model.language_model.layers.34.mlp.experts.18.gate_proj", "model.language_model.layers.34.mlp.experts.19.gate_proj", "model.language_model.layers.34.mlp.experts.20.gate_proj", "model.language_model.layers.34.mlp.experts.21.gate_proj", "model.language_model.layers.34.mlp.experts.22.gate_proj", "model.language_model.layers.34.mlp.experts.23.gate_proj", "model.language_model.layers.34.mlp.experts.24.gate_proj", "model.language_model.layers.34.mlp.experts.25.gate_proj", "model.language_model.layers.34.mlp.experts.26.gate_proj", "model.language_model.layers.34.mlp.experts.27.gate_proj", "model.language_model.layers.34.mlp.experts.28.gate_proj", "model.language_model.layers.34.mlp.experts.29.gate_proj", "model.language_model.layers.34.mlp.experts.30.gate_proj", "model.language_model.layers.34.mlp.experts.31.gate_proj", "model.language_model.layers.34.mlp.experts.32.gate_proj", "model.language_model.layers.34.mlp.experts.33.gate_proj", "model.language_model.layers.34.mlp.experts.34.gate_proj", "model.language_model.layers.34.mlp.experts.35.gate_proj", "model.language_model.layers.34.mlp.experts.36.gate_proj", "model.language_model.layers.34.mlp.experts.37.gate_proj", "model.language_model.layers.34.mlp.experts.38.gate_proj", "model.language_model.layers.34.mlp.experts.39.gate_proj", "model.language_model.layers.34.mlp.experts.40.gate_proj", "model.language_model.layers.34.mlp.experts.41.gate_proj", "model.language_model.layers.34.mlp.experts.42.gate_proj", "model.language_model.layers.34.mlp.experts.43.gate_proj", "model.language_model.layers.34.mlp.experts.44.gate_proj", "model.language_model.layers.34.mlp.experts.45.gate_proj", "model.language_model.layers.34.mlp.experts.46.gate_proj", "model.language_model.layers.34.mlp.experts.47.gate_proj", "model.language_model.layers.34.mlp.experts.48.gate_proj", "model.language_model.layers.34.mlp.experts.49.gate_proj", "model.language_model.layers.34.mlp.experts.50.gate_proj", "model.language_model.layers.34.mlp.experts.51.gate_proj", "model.language_model.layers.34.mlp.experts.52.gate_proj", "model.language_model.layers.34.mlp.experts.53.gate_proj", "model.language_model.layers.34.mlp.experts.54.gate_proj", "model.language_model.layers.34.mlp.experts.55.gate_proj", "model.language_model.layers.34.mlp.experts.56.gate_proj", "model.language_model.layers.34.mlp.experts.57.gate_proj", "model.language_model.layers.34.mlp.experts.58.gate_proj", "model.language_model.layers.34.mlp.experts.59.gate_proj", "model.language_model.layers.34.mlp.experts.60.gate_proj", "model.language_model.layers.34.mlp.experts.61.gate_proj", "model.language_model.layers.34.mlp.experts.62.gate_proj", "model.language_model.layers.34.mlp.experts.63.gate_proj", "model.language_model.layers.34.mlp.experts.64.gate_proj", "model.language_model.layers.34.mlp.experts.65.gate_proj", "model.language_model.layers.34.mlp.experts.66.gate_proj", "model.language_model.layers.34.mlp.experts.67.gate_proj", "model.language_model.layers.34.mlp.experts.68.gate_proj", "model.language_model.layers.34.mlp.experts.69.gate_proj", "model.language_model.layers.34.mlp.experts.70.gate_proj", "model.language_model.layers.34.mlp.experts.71.gate_proj", "model.language_model.layers.34.mlp.experts.72.gate_proj", "model.language_model.layers.34.mlp.experts.73.gate_proj", "model.language_model.layers.34.mlp.experts.74.gate_proj", "model.language_model.layers.34.mlp.experts.75.gate_proj", "model.language_model.layers.34.mlp.experts.76.gate_proj", "model.language_model.layers.34.mlp.experts.77.gate_proj", "model.language_model.layers.34.mlp.experts.78.gate_proj", "model.language_model.layers.34.mlp.experts.79.gate_proj", "model.language_model.layers.34.mlp.experts.80.gate_proj", "model.language_model.layers.34.mlp.experts.81.gate_proj", "model.language_model.layers.34.mlp.experts.82.gate_proj", "model.language_model.layers.34.mlp.experts.83.gate_proj", "model.language_model.layers.34.mlp.experts.84.gate_proj", "model.language_model.layers.34.mlp.experts.85.gate_proj", "model.language_model.layers.34.mlp.experts.86.gate_proj", "model.language_model.layers.34.mlp.experts.87.gate_proj", "model.language_model.layers.34.mlp.experts.88.gate_proj", "model.language_model.layers.34.mlp.experts.89.gate_proj", "model.language_model.layers.34.mlp.experts.90.gate_proj", "model.language_model.layers.34.mlp.experts.91.gate_proj", "model.language_model.layers.34.mlp.experts.92.gate_proj", "model.language_model.layers.34.mlp.experts.93.gate_proj", "model.language_model.layers.34.mlp.experts.94.gate_proj", "model.language_model.layers.34.mlp.experts.95.gate_proj", "model.language_model.layers.34.mlp.experts.96.gate_proj", "model.language_model.layers.34.mlp.experts.97.gate_proj", "model.language_model.layers.34.mlp.experts.98.gate_proj", "model.language_model.layers.34.mlp.experts.99.gate_proj", "model.language_model.layers.34.mlp.experts.100.gate_proj", "model.language_model.layers.34.mlp.experts.101.gate_proj", "model.language_model.layers.34.mlp.experts.102.gate_proj", "model.language_model.layers.34.mlp.experts.103.gate_proj", "model.language_model.layers.34.mlp.experts.104.gate_proj", "model.language_model.layers.34.mlp.experts.105.gate_proj", "model.language_model.layers.34.mlp.experts.106.gate_proj", "model.language_model.layers.34.mlp.experts.107.gate_proj", "model.language_model.layers.34.mlp.experts.108.gate_proj", "model.language_model.layers.34.mlp.experts.109.gate_proj", "model.language_model.layers.34.mlp.experts.110.gate_proj", "model.language_model.layers.34.mlp.experts.111.gate_proj", "model.language_model.layers.34.mlp.experts.112.gate_proj", "model.language_model.layers.34.mlp.experts.113.gate_proj", "model.language_model.layers.34.mlp.experts.114.gate_proj", "model.language_model.layers.34.mlp.experts.115.gate_proj", "model.language_model.layers.34.mlp.experts.116.gate_proj", "model.language_model.layers.34.mlp.experts.117.gate_proj", "model.language_model.layers.34.mlp.experts.118.gate_proj", "model.language_model.layers.34.mlp.experts.119.gate_proj", "model.language_model.layers.34.mlp.experts.120.gate_proj", "model.language_model.layers.34.mlp.experts.121.gate_proj", "model.language_model.layers.34.mlp.experts.122.gate_proj", "model.language_model.layers.34.mlp.experts.123.gate_proj", "model.language_model.layers.34.mlp.experts.124.gate_proj", "model.language_model.layers.34.mlp.experts.125.gate_proj", "model.language_model.layers.34.mlp.experts.126.gate_proj", "model.language_model.layers.34.mlp.experts.127.gate_proj", "model.language_model.layers.34.mlp.experts.0.up_proj", "model.language_model.layers.34.mlp.experts.1.up_proj", "model.language_model.layers.34.mlp.experts.2.up_proj", "model.language_model.layers.34.mlp.experts.3.up_proj", "model.language_model.layers.34.mlp.experts.4.up_proj", "model.language_model.layers.34.mlp.experts.5.up_proj", "model.language_model.layers.34.mlp.experts.6.up_proj", "model.language_model.layers.34.mlp.experts.7.up_proj", "model.language_model.layers.34.mlp.experts.8.up_proj", "model.language_model.layers.34.mlp.experts.9.up_proj", "model.language_model.layers.34.mlp.experts.10.up_proj", "model.language_model.layers.34.mlp.experts.11.up_proj", "model.language_model.layers.34.mlp.experts.12.up_proj", "model.language_model.layers.34.mlp.experts.13.up_proj", "model.language_model.layers.34.mlp.experts.14.up_proj", "model.language_model.layers.34.mlp.experts.15.up_proj", "model.language_model.layers.34.mlp.experts.16.up_proj", "model.language_model.layers.34.mlp.experts.17.up_proj", "model.language_model.layers.34.mlp.experts.18.up_proj", "model.language_model.layers.34.mlp.experts.19.up_proj", "model.language_model.layers.34.mlp.experts.20.up_proj", "model.language_model.layers.34.mlp.experts.21.up_proj", "model.language_model.layers.34.mlp.experts.22.up_proj", "model.language_model.layers.34.mlp.experts.23.up_proj", "model.language_model.layers.34.mlp.experts.24.up_proj", "model.language_model.layers.34.mlp.experts.25.up_proj", "model.language_model.layers.34.mlp.experts.26.up_proj", "model.language_model.layers.34.mlp.experts.27.up_proj", "model.language_model.layers.34.mlp.experts.28.up_proj", "model.language_model.layers.34.mlp.experts.29.up_proj", "model.language_model.layers.34.mlp.experts.30.up_proj", "model.language_model.layers.34.mlp.experts.31.up_proj", "model.language_model.layers.34.mlp.experts.32.up_proj", "model.language_model.layers.34.mlp.experts.33.up_proj", "model.language_model.layers.34.mlp.experts.34.up_proj", "model.language_model.layers.34.mlp.experts.35.up_proj", "model.language_model.layers.34.mlp.experts.36.up_proj", "model.language_model.layers.34.mlp.experts.37.up_proj", "model.language_model.layers.34.mlp.experts.38.up_proj", "model.language_model.layers.34.mlp.experts.39.up_proj", "model.language_model.layers.34.mlp.experts.40.up_proj", "model.language_model.layers.34.mlp.experts.41.up_proj", "model.language_model.layers.34.mlp.experts.42.up_proj", "model.language_model.layers.34.mlp.experts.43.up_proj", "model.language_model.layers.34.mlp.experts.44.up_proj", "model.language_model.layers.34.mlp.experts.45.up_proj", "model.language_model.layers.34.mlp.experts.46.up_proj", "model.language_model.layers.34.mlp.experts.47.up_proj", "model.language_model.layers.34.mlp.experts.48.up_proj", "model.language_model.layers.34.mlp.experts.49.up_proj", "model.language_model.layers.34.mlp.experts.50.up_proj", "model.language_model.layers.34.mlp.experts.51.up_proj", "model.language_model.layers.34.mlp.experts.52.up_proj", "model.language_model.layers.34.mlp.experts.53.up_proj", "model.language_model.layers.34.mlp.experts.54.up_proj", "model.language_model.layers.34.mlp.experts.55.up_proj", "model.language_model.layers.34.mlp.experts.56.up_proj", "model.language_model.layers.34.mlp.experts.57.up_proj", "model.language_model.layers.34.mlp.experts.58.up_proj", "model.language_model.layers.34.mlp.experts.59.up_proj", "model.language_model.layers.34.mlp.experts.60.up_proj", "model.language_model.layers.34.mlp.experts.61.up_proj", "model.language_model.layers.34.mlp.experts.62.up_proj", "model.language_model.layers.34.mlp.experts.63.up_proj", "model.language_model.layers.34.mlp.experts.64.up_proj", "model.language_model.layers.34.mlp.experts.65.up_proj", "model.language_model.layers.34.mlp.experts.66.up_proj", "model.language_model.layers.34.mlp.experts.67.up_proj", "model.language_model.layers.34.mlp.experts.68.up_proj", "model.language_model.layers.34.mlp.experts.69.up_proj", "model.language_model.layers.34.mlp.experts.70.up_proj", "model.language_model.layers.34.mlp.experts.71.up_proj", "model.language_model.layers.34.mlp.experts.72.up_proj", "model.language_model.layers.34.mlp.experts.73.up_proj", "model.language_model.layers.34.mlp.experts.74.up_proj", "model.language_model.layers.34.mlp.experts.75.up_proj", "model.language_model.layers.34.mlp.experts.76.up_proj", "model.language_model.layers.34.mlp.experts.77.up_proj", "model.language_model.layers.34.mlp.experts.78.up_proj", "model.language_model.layers.34.mlp.experts.79.up_proj", "model.language_model.layers.34.mlp.experts.80.up_proj", "model.language_model.layers.34.mlp.experts.81.up_proj", "model.language_model.layers.34.mlp.experts.82.up_proj", "model.language_model.layers.34.mlp.experts.83.up_proj", "model.language_model.layers.34.mlp.experts.84.up_proj", "model.language_model.layers.34.mlp.experts.85.up_proj", "model.language_model.layers.34.mlp.experts.86.up_proj", "model.language_model.layers.34.mlp.experts.87.up_proj", "model.language_model.layers.34.mlp.experts.88.up_proj", "model.language_model.layers.34.mlp.experts.89.up_proj", "model.language_model.layers.34.mlp.experts.90.up_proj", "model.language_model.layers.34.mlp.experts.91.up_proj", "model.language_model.layers.34.mlp.experts.92.up_proj", "model.language_model.layers.34.mlp.experts.93.up_proj", "model.language_model.layers.34.mlp.experts.94.up_proj", "model.language_model.layers.34.mlp.experts.95.up_proj", "model.language_model.layers.34.mlp.experts.96.up_proj", "model.language_model.layers.34.mlp.experts.97.up_proj", "model.language_model.layers.34.mlp.experts.98.up_proj", "model.language_model.layers.34.mlp.experts.99.up_proj", "model.language_model.layers.34.mlp.experts.100.up_proj", "model.language_model.layers.34.mlp.experts.101.up_proj", "model.language_model.layers.34.mlp.experts.102.up_proj", "model.language_model.layers.34.mlp.experts.103.up_proj", "model.language_model.layers.34.mlp.experts.104.up_proj", "model.language_model.layers.34.mlp.experts.105.up_proj", "model.language_model.layers.34.mlp.experts.106.up_proj", "model.language_model.layers.34.mlp.experts.107.up_proj", "model.language_model.layers.34.mlp.experts.108.up_proj", "model.language_model.layers.34.mlp.experts.109.up_proj", "model.language_model.layers.34.mlp.experts.110.up_proj", "model.language_model.layers.34.mlp.experts.111.up_proj", "model.language_model.layers.34.mlp.experts.112.up_proj", "model.language_model.layers.34.mlp.experts.113.up_proj", "model.language_model.layers.34.mlp.experts.114.up_proj", "model.language_model.layers.34.mlp.experts.115.up_proj", "model.language_model.layers.34.mlp.experts.116.up_proj", "model.language_model.layers.34.mlp.experts.117.up_proj", "model.language_model.layers.34.mlp.experts.118.up_proj", "model.language_model.layers.34.mlp.experts.119.up_proj", "model.language_model.layers.34.mlp.experts.120.up_proj", "model.language_model.layers.34.mlp.experts.121.up_proj", "model.language_model.layers.34.mlp.experts.122.up_proj", "model.language_model.layers.34.mlp.experts.123.up_proj", "model.language_model.layers.34.mlp.experts.124.up_proj", "model.language_model.layers.34.mlp.experts.125.up_proj", "model.language_model.layers.34.mlp.experts.126.up_proj", "model.language_model.layers.34.mlp.experts.127.up_proj", "model.language_model.layers.34.mlp.experts.0.down_proj", "model.language_model.layers.34.mlp.experts.1.down_proj", "model.language_model.layers.34.mlp.experts.2.down_proj", "model.language_model.layers.34.mlp.experts.3.down_proj", "model.language_model.layers.34.mlp.experts.4.down_proj", "model.language_model.layers.34.mlp.experts.5.down_proj", "model.language_model.layers.34.mlp.experts.6.down_proj", "model.language_model.layers.34.mlp.experts.7.down_proj", "model.language_model.layers.34.mlp.experts.8.down_proj", "model.language_model.layers.34.mlp.experts.9.down_proj", "model.language_model.layers.34.mlp.experts.10.down_proj", "model.language_model.layers.34.mlp.experts.11.down_proj", "model.language_model.layers.34.mlp.experts.12.down_proj", "model.language_model.layers.34.mlp.experts.13.down_proj", "model.language_model.layers.34.mlp.experts.14.down_proj", "model.language_model.layers.34.mlp.experts.15.down_proj", "model.language_model.layers.34.mlp.experts.16.down_proj", "model.language_model.layers.34.mlp.experts.17.down_proj", "model.language_model.layers.34.mlp.experts.18.down_proj", "model.language_model.layers.34.mlp.experts.19.down_proj", "model.language_model.layers.34.mlp.experts.20.down_proj", "model.language_model.layers.34.mlp.experts.21.down_proj", "model.language_model.layers.34.mlp.experts.22.down_proj", "model.language_model.layers.34.mlp.experts.23.down_proj", "model.language_model.layers.34.mlp.experts.24.down_proj", "model.language_model.layers.34.mlp.experts.25.down_proj", "model.language_model.layers.34.mlp.experts.26.down_proj", "model.language_model.layers.34.mlp.experts.27.down_proj", "model.language_model.layers.34.mlp.experts.28.down_proj", "model.language_model.layers.34.mlp.experts.29.down_proj", "model.language_model.layers.34.mlp.experts.30.down_proj", "model.language_model.layers.34.mlp.experts.31.down_proj", "model.language_model.layers.34.mlp.experts.32.down_proj", "model.language_model.layers.34.mlp.experts.33.down_proj", "model.language_model.layers.34.mlp.experts.34.down_proj", "model.language_model.layers.34.mlp.experts.35.down_proj", "model.language_model.layers.34.mlp.experts.36.down_proj", "model.language_model.layers.34.mlp.experts.37.down_proj", "model.language_model.layers.34.mlp.experts.38.down_proj", "model.language_model.layers.34.mlp.experts.39.down_proj", "model.language_model.layers.34.mlp.experts.40.down_proj", "model.language_model.layers.34.mlp.experts.41.down_proj", "model.language_model.layers.34.mlp.experts.42.down_proj", "model.language_model.layers.34.mlp.experts.43.down_proj", "model.language_model.layers.34.mlp.experts.44.down_proj", "model.language_model.layers.34.mlp.experts.45.down_proj", "model.language_model.layers.34.mlp.experts.46.down_proj", "model.language_model.layers.34.mlp.experts.47.down_proj", "model.language_model.layers.34.mlp.experts.48.down_proj", "model.language_model.layers.34.mlp.experts.49.down_proj", "model.language_model.layers.34.mlp.experts.50.down_proj", "model.language_model.layers.34.mlp.experts.51.down_proj", "model.language_model.layers.34.mlp.experts.52.down_proj", "model.language_model.layers.34.mlp.experts.53.down_proj", "model.language_model.layers.34.mlp.experts.54.down_proj", "model.language_model.layers.34.mlp.experts.55.down_proj", "model.language_model.layers.34.mlp.experts.56.down_proj", "model.language_model.layers.34.mlp.experts.57.down_proj", "model.language_model.layers.34.mlp.experts.58.down_proj", "model.language_model.layers.34.mlp.experts.59.down_proj", "model.language_model.layers.34.mlp.experts.60.down_proj", "model.language_model.layers.34.mlp.experts.61.down_proj", "model.language_model.layers.34.mlp.experts.62.down_proj", "model.language_model.layers.34.mlp.experts.63.down_proj", "model.language_model.layers.34.mlp.experts.64.down_proj", "model.language_model.layers.34.mlp.experts.65.down_proj", "model.language_model.layers.34.mlp.experts.66.down_proj", "model.language_model.layers.34.mlp.experts.67.down_proj", "model.language_model.layers.34.mlp.experts.68.down_proj", "model.language_model.layers.34.mlp.experts.69.down_proj", "model.language_model.layers.34.mlp.experts.70.down_proj", "model.language_model.layers.34.mlp.experts.71.down_proj", "model.language_model.layers.34.mlp.experts.72.down_proj", "model.language_model.layers.34.mlp.experts.73.down_proj", "model.language_model.layers.34.mlp.experts.74.down_proj", "model.language_model.layers.34.mlp.experts.75.down_proj", "model.language_model.layers.34.mlp.experts.76.down_proj", "model.language_model.layers.34.mlp.experts.77.down_proj", "model.language_model.layers.34.mlp.experts.78.down_proj", "model.language_model.layers.34.mlp.experts.79.down_proj", "model.language_model.layers.34.mlp.experts.80.down_proj", "model.language_model.layers.34.mlp.experts.81.down_proj", "model.language_model.layers.34.mlp.experts.82.down_proj", "model.language_model.layers.34.mlp.experts.83.down_proj", "model.language_model.layers.34.mlp.experts.84.down_proj", "model.language_model.layers.34.mlp.experts.85.down_proj", "model.language_model.layers.34.mlp.experts.86.down_proj", "model.language_model.layers.34.mlp.experts.87.down_proj", "model.language_model.layers.34.mlp.experts.88.down_proj", "model.language_model.layers.34.mlp.experts.89.down_proj", "model.language_model.layers.34.mlp.experts.90.down_proj", "model.language_model.layers.34.mlp.experts.91.down_proj", "model.language_model.layers.34.mlp.experts.92.down_proj", "model.language_model.layers.34.mlp.experts.93.down_proj", "model.language_model.layers.34.mlp.experts.94.down_proj", "model.language_model.layers.34.mlp.experts.95.down_proj", "model.language_model.layers.34.mlp.experts.96.down_proj", "model.language_model.layers.34.mlp.experts.97.down_proj", "model.language_model.layers.34.mlp.experts.98.down_proj", "model.language_model.layers.34.mlp.experts.99.down_proj", "model.language_model.layers.34.mlp.experts.100.down_proj", "model.language_model.layers.34.mlp.experts.101.down_proj", "model.language_model.layers.34.mlp.experts.102.down_proj", "model.language_model.layers.34.mlp.experts.103.down_proj", "model.language_model.layers.34.mlp.experts.104.down_proj", "model.language_model.layers.34.mlp.experts.105.down_proj", "model.language_model.layers.34.mlp.experts.106.down_proj", "model.language_model.layers.34.mlp.experts.107.down_proj", "model.language_model.layers.34.mlp.experts.108.down_proj", "model.language_model.layers.34.mlp.experts.109.down_proj", "model.language_model.layers.34.mlp.experts.110.down_proj", "model.language_model.layers.34.mlp.experts.111.down_proj", "model.language_model.layers.34.mlp.experts.112.down_proj", "model.language_model.layers.34.mlp.experts.113.down_proj", "model.language_model.layers.34.mlp.experts.114.down_proj", "model.language_model.layers.34.mlp.experts.115.down_proj", "model.language_model.layers.34.mlp.experts.116.down_proj", "model.language_model.layers.34.mlp.experts.117.down_proj", "model.language_model.layers.34.mlp.experts.118.down_proj", "model.language_model.layers.34.mlp.experts.119.down_proj", "model.language_model.layers.34.mlp.experts.120.down_proj", "model.language_model.layers.34.mlp.experts.121.down_proj", "model.language_model.layers.34.mlp.experts.122.down_proj", "model.language_model.layers.34.mlp.experts.123.down_proj", "model.language_model.layers.34.mlp.experts.124.down_proj", "model.language_model.layers.34.mlp.experts.125.down_proj", "model.language_model.layers.34.mlp.experts.126.down_proj", "model.language_model.layers.34.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.0009261626750230734, "dbits": 603979776 } ] }, { "idx": 70, "layers": [ "model.language_model.layers.35.self_attn.q_proj", "model.language_model.layers.35.self_attn.k_proj", "model.language_model.layers.35.self_attn.v_proj", "model.language_model.layers.35.self_attn.o_proj" ], "candidates": [ { "dkld": -1.2130942195653915e-05, "dbits": 18874368 } ] }, { "idx": 71, "layers": [ "model.language_model.layers.35.mlp.experts.0.gate_proj", "model.language_model.layers.35.mlp.experts.1.gate_proj", "model.language_model.layers.35.mlp.experts.2.gate_proj", "model.language_model.layers.35.mlp.experts.3.gate_proj", "model.language_model.layers.35.mlp.experts.4.gate_proj", "model.language_model.layers.35.mlp.experts.5.gate_proj", "model.language_model.layers.35.mlp.experts.6.gate_proj", "model.language_model.layers.35.mlp.experts.7.gate_proj", "model.language_model.layers.35.mlp.experts.8.gate_proj", "model.language_model.layers.35.mlp.experts.9.gate_proj", "model.language_model.layers.35.mlp.experts.10.gate_proj", "model.language_model.layers.35.mlp.experts.11.gate_proj", "model.language_model.layers.35.mlp.experts.12.gate_proj", "model.language_model.layers.35.mlp.experts.13.gate_proj", "model.language_model.layers.35.mlp.experts.14.gate_proj", "model.language_model.layers.35.mlp.experts.15.gate_proj", "model.language_model.layers.35.mlp.experts.16.gate_proj", "model.language_model.layers.35.mlp.experts.17.gate_proj", "model.language_model.layers.35.mlp.experts.18.gate_proj", "model.language_model.layers.35.mlp.experts.19.gate_proj", "model.language_model.layers.35.mlp.experts.20.gate_proj", "model.language_model.layers.35.mlp.experts.21.gate_proj", "model.language_model.layers.35.mlp.experts.22.gate_proj", "model.language_model.layers.35.mlp.experts.23.gate_proj", "model.language_model.layers.35.mlp.experts.24.gate_proj", "model.language_model.layers.35.mlp.experts.25.gate_proj", "model.language_model.layers.35.mlp.experts.26.gate_proj", "model.language_model.layers.35.mlp.experts.27.gate_proj", "model.language_model.layers.35.mlp.experts.28.gate_proj", "model.language_model.layers.35.mlp.experts.29.gate_proj", "model.language_model.layers.35.mlp.experts.30.gate_proj", "model.language_model.layers.35.mlp.experts.31.gate_proj", "model.language_model.layers.35.mlp.experts.32.gate_proj", "model.language_model.layers.35.mlp.experts.33.gate_proj", "model.language_model.layers.35.mlp.experts.34.gate_proj", "model.language_model.layers.35.mlp.experts.35.gate_proj", "model.language_model.layers.35.mlp.experts.36.gate_proj", "model.language_model.layers.35.mlp.experts.37.gate_proj", "model.language_model.layers.35.mlp.experts.38.gate_proj", "model.language_model.layers.35.mlp.experts.39.gate_proj", "model.language_model.layers.35.mlp.experts.40.gate_proj", "model.language_model.layers.35.mlp.experts.41.gate_proj", "model.language_model.layers.35.mlp.experts.42.gate_proj", "model.language_model.layers.35.mlp.experts.43.gate_proj", "model.language_model.layers.35.mlp.experts.44.gate_proj", "model.language_model.layers.35.mlp.experts.45.gate_proj", "model.language_model.layers.35.mlp.experts.46.gate_proj", "model.language_model.layers.35.mlp.experts.47.gate_proj", "model.language_model.layers.35.mlp.experts.48.gate_proj", "model.language_model.layers.35.mlp.experts.49.gate_proj", "model.language_model.layers.35.mlp.experts.50.gate_proj", "model.language_model.layers.35.mlp.experts.51.gate_proj", "model.language_model.layers.35.mlp.experts.52.gate_proj", "model.language_model.layers.35.mlp.experts.53.gate_proj", "model.language_model.layers.35.mlp.experts.54.gate_proj", "model.language_model.layers.35.mlp.experts.55.gate_proj", "model.language_model.layers.35.mlp.experts.56.gate_proj", "model.language_model.layers.35.mlp.experts.57.gate_proj", "model.language_model.layers.35.mlp.experts.58.gate_proj", "model.language_model.layers.35.mlp.experts.59.gate_proj", "model.language_model.layers.35.mlp.experts.60.gate_proj", "model.language_model.layers.35.mlp.experts.61.gate_proj", "model.language_model.layers.35.mlp.experts.62.gate_proj", "model.language_model.layers.35.mlp.experts.63.gate_proj", "model.language_model.layers.35.mlp.experts.64.gate_proj", "model.language_model.layers.35.mlp.experts.65.gate_proj", "model.language_model.layers.35.mlp.experts.66.gate_proj", "model.language_model.layers.35.mlp.experts.67.gate_proj", "model.language_model.layers.35.mlp.experts.68.gate_proj", "model.language_model.layers.35.mlp.experts.69.gate_proj", "model.language_model.layers.35.mlp.experts.70.gate_proj", "model.language_model.layers.35.mlp.experts.71.gate_proj", "model.language_model.layers.35.mlp.experts.72.gate_proj", "model.language_model.layers.35.mlp.experts.73.gate_proj", "model.language_model.layers.35.mlp.experts.74.gate_proj", "model.language_model.layers.35.mlp.experts.75.gate_proj", "model.language_model.layers.35.mlp.experts.76.gate_proj", "model.language_model.layers.35.mlp.experts.77.gate_proj", "model.language_model.layers.35.mlp.experts.78.gate_proj", "model.language_model.layers.35.mlp.experts.79.gate_proj", "model.language_model.layers.35.mlp.experts.80.gate_proj", "model.language_model.layers.35.mlp.experts.81.gate_proj", "model.language_model.layers.35.mlp.experts.82.gate_proj", "model.language_model.layers.35.mlp.experts.83.gate_proj", "model.language_model.layers.35.mlp.experts.84.gate_proj", "model.language_model.layers.35.mlp.experts.85.gate_proj", "model.language_model.layers.35.mlp.experts.86.gate_proj", "model.language_model.layers.35.mlp.experts.87.gate_proj", "model.language_model.layers.35.mlp.experts.88.gate_proj", "model.language_model.layers.35.mlp.experts.89.gate_proj", "model.language_model.layers.35.mlp.experts.90.gate_proj", "model.language_model.layers.35.mlp.experts.91.gate_proj", "model.language_model.layers.35.mlp.experts.92.gate_proj", "model.language_model.layers.35.mlp.experts.93.gate_proj", "model.language_model.layers.35.mlp.experts.94.gate_proj", "model.language_model.layers.35.mlp.experts.95.gate_proj", "model.language_model.layers.35.mlp.experts.96.gate_proj", "model.language_model.layers.35.mlp.experts.97.gate_proj", "model.language_model.layers.35.mlp.experts.98.gate_proj", "model.language_model.layers.35.mlp.experts.99.gate_proj", "model.language_model.layers.35.mlp.experts.100.gate_proj", "model.language_model.layers.35.mlp.experts.101.gate_proj", "model.language_model.layers.35.mlp.experts.102.gate_proj", "model.language_model.layers.35.mlp.experts.103.gate_proj", "model.language_model.layers.35.mlp.experts.104.gate_proj", "model.language_model.layers.35.mlp.experts.105.gate_proj", "model.language_model.layers.35.mlp.experts.106.gate_proj", "model.language_model.layers.35.mlp.experts.107.gate_proj", "model.language_model.layers.35.mlp.experts.108.gate_proj", "model.language_model.layers.35.mlp.experts.109.gate_proj", "model.language_model.layers.35.mlp.experts.110.gate_proj", "model.language_model.layers.35.mlp.experts.111.gate_proj", "model.language_model.layers.35.mlp.experts.112.gate_proj", "model.language_model.layers.35.mlp.experts.113.gate_proj", "model.language_model.layers.35.mlp.experts.114.gate_proj", "model.language_model.layers.35.mlp.experts.115.gate_proj", "model.language_model.layers.35.mlp.experts.116.gate_proj", "model.language_model.layers.35.mlp.experts.117.gate_proj", "model.language_model.layers.35.mlp.experts.118.gate_proj", "model.language_model.layers.35.mlp.experts.119.gate_proj", "model.language_model.layers.35.mlp.experts.120.gate_proj", "model.language_model.layers.35.mlp.experts.121.gate_proj", "model.language_model.layers.35.mlp.experts.122.gate_proj", "model.language_model.layers.35.mlp.experts.123.gate_proj", "model.language_model.layers.35.mlp.experts.124.gate_proj", "model.language_model.layers.35.mlp.experts.125.gate_proj", "model.language_model.layers.35.mlp.experts.126.gate_proj", "model.language_model.layers.35.mlp.experts.127.gate_proj", "model.language_model.layers.35.mlp.experts.0.up_proj", "model.language_model.layers.35.mlp.experts.1.up_proj", "model.language_model.layers.35.mlp.experts.2.up_proj", "model.language_model.layers.35.mlp.experts.3.up_proj", "model.language_model.layers.35.mlp.experts.4.up_proj", "model.language_model.layers.35.mlp.experts.5.up_proj", "model.language_model.layers.35.mlp.experts.6.up_proj", "model.language_model.layers.35.mlp.experts.7.up_proj", "model.language_model.layers.35.mlp.experts.8.up_proj", "model.language_model.layers.35.mlp.experts.9.up_proj", "model.language_model.layers.35.mlp.experts.10.up_proj", "model.language_model.layers.35.mlp.experts.11.up_proj", "model.language_model.layers.35.mlp.experts.12.up_proj", "model.language_model.layers.35.mlp.experts.13.up_proj", "model.language_model.layers.35.mlp.experts.14.up_proj", "model.language_model.layers.35.mlp.experts.15.up_proj", "model.language_model.layers.35.mlp.experts.16.up_proj", "model.language_model.layers.35.mlp.experts.17.up_proj", "model.language_model.layers.35.mlp.experts.18.up_proj", "model.language_model.layers.35.mlp.experts.19.up_proj", "model.language_model.layers.35.mlp.experts.20.up_proj", "model.language_model.layers.35.mlp.experts.21.up_proj", "model.language_model.layers.35.mlp.experts.22.up_proj", "model.language_model.layers.35.mlp.experts.23.up_proj", "model.language_model.layers.35.mlp.experts.24.up_proj", "model.language_model.layers.35.mlp.experts.25.up_proj", "model.language_model.layers.35.mlp.experts.26.up_proj", "model.language_model.layers.35.mlp.experts.27.up_proj", "model.language_model.layers.35.mlp.experts.28.up_proj", "model.language_model.layers.35.mlp.experts.29.up_proj", "model.language_model.layers.35.mlp.experts.30.up_proj", "model.language_model.layers.35.mlp.experts.31.up_proj", "model.language_model.layers.35.mlp.experts.32.up_proj", "model.language_model.layers.35.mlp.experts.33.up_proj", "model.language_model.layers.35.mlp.experts.34.up_proj", "model.language_model.layers.35.mlp.experts.35.up_proj", "model.language_model.layers.35.mlp.experts.36.up_proj", "model.language_model.layers.35.mlp.experts.37.up_proj", "model.language_model.layers.35.mlp.experts.38.up_proj", "model.language_model.layers.35.mlp.experts.39.up_proj", "model.language_model.layers.35.mlp.experts.40.up_proj", "model.language_model.layers.35.mlp.experts.41.up_proj", "model.language_model.layers.35.mlp.experts.42.up_proj", "model.language_model.layers.35.mlp.experts.43.up_proj", "model.language_model.layers.35.mlp.experts.44.up_proj", "model.language_model.layers.35.mlp.experts.45.up_proj", "model.language_model.layers.35.mlp.experts.46.up_proj", "model.language_model.layers.35.mlp.experts.47.up_proj", "model.language_model.layers.35.mlp.experts.48.up_proj", "model.language_model.layers.35.mlp.experts.49.up_proj", "model.language_model.layers.35.mlp.experts.50.up_proj", "model.language_model.layers.35.mlp.experts.51.up_proj", "model.language_model.layers.35.mlp.experts.52.up_proj", "model.language_model.layers.35.mlp.experts.53.up_proj", "model.language_model.layers.35.mlp.experts.54.up_proj", "model.language_model.layers.35.mlp.experts.55.up_proj", "model.language_model.layers.35.mlp.experts.56.up_proj", "model.language_model.layers.35.mlp.experts.57.up_proj", "model.language_model.layers.35.mlp.experts.58.up_proj", "model.language_model.layers.35.mlp.experts.59.up_proj", "model.language_model.layers.35.mlp.experts.60.up_proj", "model.language_model.layers.35.mlp.experts.61.up_proj", "model.language_model.layers.35.mlp.experts.62.up_proj", "model.language_model.layers.35.mlp.experts.63.up_proj", "model.language_model.layers.35.mlp.experts.64.up_proj", "model.language_model.layers.35.mlp.experts.65.up_proj", "model.language_model.layers.35.mlp.experts.66.up_proj", "model.language_model.layers.35.mlp.experts.67.up_proj", "model.language_model.layers.35.mlp.experts.68.up_proj", "model.language_model.layers.35.mlp.experts.69.up_proj", "model.language_model.layers.35.mlp.experts.70.up_proj", "model.language_model.layers.35.mlp.experts.71.up_proj", "model.language_model.layers.35.mlp.experts.72.up_proj", "model.language_model.layers.35.mlp.experts.73.up_proj", "model.language_model.layers.35.mlp.experts.74.up_proj", "model.language_model.layers.35.mlp.experts.75.up_proj", "model.language_model.layers.35.mlp.experts.76.up_proj", "model.language_model.layers.35.mlp.experts.77.up_proj", "model.language_model.layers.35.mlp.experts.78.up_proj", "model.language_model.layers.35.mlp.experts.79.up_proj", "model.language_model.layers.35.mlp.experts.80.up_proj", "model.language_model.layers.35.mlp.experts.81.up_proj", "model.language_model.layers.35.mlp.experts.82.up_proj", "model.language_model.layers.35.mlp.experts.83.up_proj", "model.language_model.layers.35.mlp.experts.84.up_proj", "model.language_model.layers.35.mlp.experts.85.up_proj", "model.language_model.layers.35.mlp.experts.86.up_proj", "model.language_model.layers.35.mlp.experts.87.up_proj", "model.language_model.layers.35.mlp.experts.88.up_proj", "model.language_model.layers.35.mlp.experts.89.up_proj", "model.language_model.layers.35.mlp.experts.90.up_proj", "model.language_model.layers.35.mlp.experts.91.up_proj", "model.language_model.layers.35.mlp.experts.92.up_proj", "model.language_model.layers.35.mlp.experts.93.up_proj", "model.language_model.layers.35.mlp.experts.94.up_proj", "model.language_model.layers.35.mlp.experts.95.up_proj", "model.language_model.layers.35.mlp.experts.96.up_proj", "model.language_model.layers.35.mlp.experts.97.up_proj", "model.language_model.layers.35.mlp.experts.98.up_proj", "model.language_model.layers.35.mlp.experts.99.up_proj", "model.language_model.layers.35.mlp.experts.100.up_proj", "model.language_model.layers.35.mlp.experts.101.up_proj", "model.language_model.layers.35.mlp.experts.102.up_proj", "model.language_model.layers.35.mlp.experts.103.up_proj", "model.language_model.layers.35.mlp.experts.104.up_proj", "model.language_model.layers.35.mlp.experts.105.up_proj", "model.language_model.layers.35.mlp.experts.106.up_proj", "model.language_model.layers.35.mlp.experts.107.up_proj", "model.language_model.layers.35.mlp.experts.108.up_proj", "model.language_model.layers.35.mlp.experts.109.up_proj", "model.language_model.layers.35.mlp.experts.110.up_proj", "model.language_model.layers.35.mlp.experts.111.up_proj", "model.language_model.layers.35.mlp.experts.112.up_proj", "model.language_model.layers.35.mlp.experts.113.up_proj", "model.language_model.layers.35.mlp.experts.114.up_proj", "model.language_model.layers.35.mlp.experts.115.up_proj", "model.language_model.layers.35.mlp.experts.116.up_proj", "model.language_model.layers.35.mlp.experts.117.up_proj", "model.language_model.layers.35.mlp.experts.118.up_proj", "model.language_model.layers.35.mlp.experts.119.up_proj", "model.language_model.layers.35.mlp.experts.120.up_proj", "model.language_model.layers.35.mlp.experts.121.up_proj", "model.language_model.layers.35.mlp.experts.122.up_proj", "model.language_model.layers.35.mlp.experts.123.up_proj", "model.language_model.layers.35.mlp.experts.124.up_proj", "model.language_model.layers.35.mlp.experts.125.up_proj", "model.language_model.layers.35.mlp.experts.126.up_proj", "model.language_model.layers.35.mlp.experts.127.up_proj", "model.language_model.layers.35.mlp.experts.0.down_proj", "model.language_model.layers.35.mlp.experts.1.down_proj", "model.language_model.layers.35.mlp.experts.2.down_proj", "model.language_model.layers.35.mlp.experts.3.down_proj", "model.language_model.layers.35.mlp.experts.4.down_proj", "model.language_model.layers.35.mlp.experts.5.down_proj", "model.language_model.layers.35.mlp.experts.6.down_proj", "model.language_model.layers.35.mlp.experts.7.down_proj", "model.language_model.layers.35.mlp.experts.8.down_proj", "model.language_model.layers.35.mlp.experts.9.down_proj", "model.language_model.layers.35.mlp.experts.10.down_proj", "model.language_model.layers.35.mlp.experts.11.down_proj", "model.language_model.layers.35.mlp.experts.12.down_proj", "model.language_model.layers.35.mlp.experts.13.down_proj", "model.language_model.layers.35.mlp.experts.14.down_proj", "model.language_model.layers.35.mlp.experts.15.down_proj", "model.language_model.layers.35.mlp.experts.16.down_proj", "model.language_model.layers.35.mlp.experts.17.down_proj", "model.language_model.layers.35.mlp.experts.18.down_proj", "model.language_model.layers.35.mlp.experts.19.down_proj", "model.language_model.layers.35.mlp.experts.20.down_proj", "model.language_model.layers.35.mlp.experts.21.down_proj", "model.language_model.layers.35.mlp.experts.22.down_proj", "model.language_model.layers.35.mlp.experts.23.down_proj", "model.language_model.layers.35.mlp.experts.24.down_proj", "model.language_model.layers.35.mlp.experts.25.down_proj", "model.language_model.layers.35.mlp.experts.26.down_proj", "model.language_model.layers.35.mlp.experts.27.down_proj", "model.language_model.layers.35.mlp.experts.28.down_proj", "model.language_model.layers.35.mlp.experts.29.down_proj", "model.language_model.layers.35.mlp.experts.30.down_proj", "model.language_model.layers.35.mlp.experts.31.down_proj", "model.language_model.layers.35.mlp.experts.32.down_proj", "model.language_model.layers.35.mlp.experts.33.down_proj", "model.language_model.layers.35.mlp.experts.34.down_proj", "model.language_model.layers.35.mlp.experts.35.down_proj", "model.language_model.layers.35.mlp.experts.36.down_proj", "model.language_model.layers.35.mlp.experts.37.down_proj", "model.language_model.layers.35.mlp.experts.38.down_proj", "model.language_model.layers.35.mlp.experts.39.down_proj", "model.language_model.layers.35.mlp.experts.40.down_proj", "model.language_model.layers.35.mlp.experts.41.down_proj", "model.language_model.layers.35.mlp.experts.42.down_proj", "model.language_model.layers.35.mlp.experts.43.down_proj", "model.language_model.layers.35.mlp.experts.44.down_proj", "model.language_model.layers.35.mlp.experts.45.down_proj", "model.language_model.layers.35.mlp.experts.46.down_proj", "model.language_model.layers.35.mlp.experts.47.down_proj", "model.language_model.layers.35.mlp.experts.48.down_proj", "model.language_model.layers.35.mlp.experts.49.down_proj", "model.language_model.layers.35.mlp.experts.50.down_proj", "model.language_model.layers.35.mlp.experts.51.down_proj", "model.language_model.layers.35.mlp.experts.52.down_proj", "model.language_model.layers.35.mlp.experts.53.down_proj", "model.language_model.layers.35.mlp.experts.54.down_proj", "model.language_model.layers.35.mlp.experts.55.down_proj", "model.language_model.layers.35.mlp.experts.56.down_proj", "model.language_model.layers.35.mlp.experts.57.down_proj", "model.language_model.layers.35.mlp.experts.58.down_proj", "model.language_model.layers.35.mlp.experts.59.down_proj", "model.language_model.layers.35.mlp.experts.60.down_proj", "model.language_model.layers.35.mlp.experts.61.down_proj", "model.language_model.layers.35.mlp.experts.62.down_proj", "model.language_model.layers.35.mlp.experts.63.down_proj", "model.language_model.layers.35.mlp.experts.64.down_proj", "model.language_model.layers.35.mlp.experts.65.down_proj", "model.language_model.layers.35.mlp.experts.66.down_proj", "model.language_model.layers.35.mlp.experts.67.down_proj", "model.language_model.layers.35.mlp.experts.68.down_proj", "model.language_model.layers.35.mlp.experts.69.down_proj", "model.language_model.layers.35.mlp.experts.70.down_proj", "model.language_model.layers.35.mlp.experts.71.down_proj", "model.language_model.layers.35.mlp.experts.72.down_proj", "model.language_model.layers.35.mlp.experts.73.down_proj", "model.language_model.layers.35.mlp.experts.74.down_proj", "model.language_model.layers.35.mlp.experts.75.down_proj", "model.language_model.layers.35.mlp.experts.76.down_proj", "model.language_model.layers.35.mlp.experts.77.down_proj", "model.language_model.layers.35.mlp.experts.78.down_proj", "model.language_model.layers.35.mlp.experts.79.down_proj", "model.language_model.layers.35.mlp.experts.80.down_proj", "model.language_model.layers.35.mlp.experts.81.down_proj", "model.language_model.layers.35.mlp.experts.82.down_proj", "model.language_model.layers.35.mlp.experts.83.down_proj", "model.language_model.layers.35.mlp.experts.84.down_proj", "model.language_model.layers.35.mlp.experts.85.down_proj", "model.language_model.layers.35.mlp.experts.86.down_proj", "model.language_model.layers.35.mlp.experts.87.down_proj", "model.language_model.layers.35.mlp.experts.88.down_proj", "model.language_model.layers.35.mlp.experts.89.down_proj", "model.language_model.layers.35.mlp.experts.90.down_proj", "model.language_model.layers.35.mlp.experts.91.down_proj", "model.language_model.layers.35.mlp.experts.92.down_proj", "model.language_model.layers.35.mlp.experts.93.down_proj", "model.language_model.layers.35.mlp.experts.94.down_proj", "model.language_model.layers.35.mlp.experts.95.down_proj", "model.language_model.layers.35.mlp.experts.96.down_proj", "model.language_model.layers.35.mlp.experts.97.down_proj", "model.language_model.layers.35.mlp.experts.98.down_proj", "model.language_model.layers.35.mlp.experts.99.down_proj", "model.language_model.layers.35.mlp.experts.100.down_proj", "model.language_model.layers.35.mlp.experts.101.down_proj", "model.language_model.layers.35.mlp.experts.102.down_proj", "model.language_model.layers.35.mlp.experts.103.down_proj", "model.language_model.layers.35.mlp.experts.104.down_proj", "model.language_model.layers.35.mlp.experts.105.down_proj", "model.language_model.layers.35.mlp.experts.106.down_proj", "model.language_model.layers.35.mlp.experts.107.down_proj", "model.language_model.layers.35.mlp.experts.108.down_proj", "model.language_model.layers.35.mlp.experts.109.down_proj", "model.language_model.layers.35.mlp.experts.110.down_proj", "model.language_model.layers.35.mlp.experts.111.down_proj", "model.language_model.layers.35.mlp.experts.112.down_proj", "model.language_model.layers.35.mlp.experts.113.down_proj", "model.language_model.layers.35.mlp.experts.114.down_proj", "model.language_model.layers.35.mlp.experts.115.down_proj", "model.language_model.layers.35.mlp.experts.116.down_proj", "model.language_model.layers.35.mlp.experts.117.down_proj", "model.language_model.layers.35.mlp.experts.118.down_proj", "model.language_model.layers.35.mlp.experts.119.down_proj", "model.language_model.layers.35.mlp.experts.120.down_proj", "model.language_model.layers.35.mlp.experts.121.down_proj", "model.language_model.layers.35.mlp.experts.122.down_proj", "model.language_model.layers.35.mlp.experts.123.down_proj", "model.language_model.layers.35.mlp.experts.124.down_proj", "model.language_model.layers.35.mlp.experts.125.down_proj", "model.language_model.layers.35.mlp.experts.126.down_proj", "model.language_model.layers.35.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.0005161253735423144, "dbits": 603979776 } ] }, { "idx": 72, "layers": [ "model.language_model.layers.36.self_attn.q_proj", "model.language_model.layers.36.self_attn.k_proj", "model.language_model.layers.36.self_attn.v_proj", "model.language_model.layers.36.self_attn.o_proj" ], "candidates": [ { "dkld": -0.00036950679495931227, "dbits": 18874368 } ] }, { "idx": 73, "layers": [ "model.language_model.layers.36.mlp.experts.0.gate_proj", "model.language_model.layers.36.mlp.experts.1.gate_proj", "model.language_model.layers.36.mlp.experts.2.gate_proj", "model.language_model.layers.36.mlp.experts.3.gate_proj", "model.language_model.layers.36.mlp.experts.4.gate_proj", "model.language_model.layers.36.mlp.experts.5.gate_proj", "model.language_model.layers.36.mlp.experts.6.gate_proj", "model.language_model.layers.36.mlp.experts.7.gate_proj", "model.language_model.layers.36.mlp.experts.8.gate_proj", "model.language_model.layers.36.mlp.experts.9.gate_proj", "model.language_model.layers.36.mlp.experts.10.gate_proj", "model.language_model.layers.36.mlp.experts.11.gate_proj", "model.language_model.layers.36.mlp.experts.12.gate_proj", "model.language_model.layers.36.mlp.experts.13.gate_proj", "model.language_model.layers.36.mlp.experts.14.gate_proj", "model.language_model.layers.36.mlp.experts.15.gate_proj", "model.language_model.layers.36.mlp.experts.16.gate_proj", "model.language_model.layers.36.mlp.experts.17.gate_proj", "model.language_model.layers.36.mlp.experts.18.gate_proj", "model.language_model.layers.36.mlp.experts.19.gate_proj", "model.language_model.layers.36.mlp.experts.20.gate_proj", "model.language_model.layers.36.mlp.experts.21.gate_proj", "model.language_model.layers.36.mlp.experts.22.gate_proj", "model.language_model.layers.36.mlp.experts.23.gate_proj", "model.language_model.layers.36.mlp.experts.24.gate_proj", "model.language_model.layers.36.mlp.experts.25.gate_proj", "model.language_model.layers.36.mlp.experts.26.gate_proj", "model.language_model.layers.36.mlp.experts.27.gate_proj", "model.language_model.layers.36.mlp.experts.28.gate_proj", "model.language_model.layers.36.mlp.experts.29.gate_proj", "model.language_model.layers.36.mlp.experts.30.gate_proj", "model.language_model.layers.36.mlp.experts.31.gate_proj", "model.language_model.layers.36.mlp.experts.32.gate_proj", "model.language_model.layers.36.mlp.experts.33.gate_proj", "model.language_model.layers.36.mlp.experts.34.gate_proj", "model.language_model.layers.36.mlp.experts.35.gate_proj", "model.language_model.layers.36.mlp.experts.36.gate_proj", "model.language_model.layers.36.mlp.experts.37.gate_proj", "model.language_model.layers.36.mlp.experts.38.gate_proj", "model.language_model.layers.36.mlp.experts.39.gate_proj", "model.language_model.layers.36.mlp.experts.40.gate_proj", "model.language_model.layers.36.mlp.experts.41.gate_proj", "model.language_model.layers.36.mlp.experts.42.gate_proj", "model.language_model.layers.36.mlp.experts.43.gate_proj", "model.language_model.layers.36.mlp.experts.44.gate_proj", "model.language_model.layers.36.mlp.experts.45.gate_proj", "model.language_model.layers.36.mlp.experts.46.gate_proj", "model.language_model.layers.36.mlp.experts.47.gate_proj", "model.language_model.layers.36.mlp.experts.48.gate_proj", "model.language_model.layers.36.mlp.experts.49.gate_proj", "model.language_model.layers.36.mlp.experts.50.gate_proj", "model.language_model.layers.36.mlp.experts.51.gate_proj", "model.language_model.layers.36.mlp.experts.52.gate_proj", "model.language_model.layers.36.mlp.experts.53.gate_proj", "model.language_model.layers.36.mlp.experts.54.gate_proj", "model.language_model.layers.36.mlp.experts.55.gate_proj", "model.language_model.layers.36.mlp.experts.56.gate_proj", "model.language_model.layers.36.mlp.experts.57.gate_proj", "model.language_model.layers.36.mlp.experts.58.gate_proj", "model.language_model.layers.36.mlp.experts.59.gate_proj", "model.language_model.layers.36.mlp.experts.60.gate_proj", "model.language_model.layers.36.mlp.experts.61.gate_proj", "model.language_model.layers.36.mlp.experts.62.gate_proj", "model.language_model.layers.36.mlp.experts.63.gate_proj", "model.language_model.layers.36.mlp.experts.64.gate_proj", "model.language_model.layers.36.mlp.experts.65.gate_proj", "model.language_model.layers.36.mlp.experts.66.gate_proj", "model.language_model.layers.36.mlp.experts.67.gate_proj", "model.language_model.layers.36.mlp.experts.68.gate_proj", "model.language_model.layers.36.mlp.experts.69.gate_proj", "model.language_model.layers.36.mlp.experts.70.gate_proj", "model.language_model.layers.36.mlp.experts.71.gate_proj", "model.language_model.layers.36.mlp.experts.72.gate_proj", "model.language_model.layers.36.mlp.experts.73.gate_proj", "model.language_model.layers.36.mlp.experts.74.gate_proj", "model.language_model.layers.36.mlp.experts.75.gate_proj", "model.language_model.layers.36.mlp.experts.76.gate_proj", "model.language_model.layers.36.mlp.experts.77.gate_proj", "model.language_model.layers.36.mlp.experts.78.gate_proj", "model.language_model.layers.36.mlp.experts.79.gate_proj", "model.language_model.layers.36.mlp.experts.80.gate_proj", "model.language_model.layers.36.mlp.experts.81.gate_proj", "model.language_model.layers.36.mlp.experts.82.gate_proj", "model.language_model.layers.36.mlp.experts.83.gate_proj", "model.language_model.layers.36.mlp.experts.84.gate_proj", "model.language_model.layers.36.mlp.experts.85.gate_proj", "model.language_model.layers.36.mlp.experts.86.gate_proj", "model.language_model.layers.36.mlp.experts.87.gate_proj", "model.language_model.layers.36.mlp.experts.88.gate_proj", "model.language_model.layers.36.mlp.experts.89.gate_proj", "model.language_model.layers.36.mlp.experts.90.gate_proj", "model.language_model.layers.36.mlp.experts.91.gate_proj", "model.language_model.layers.36.mlp.experts.92.gate_proj", "model.language_model.layers.36.mlp.experts.93.gate_proj", "model.language_model.layers.36.mlp.experts.94.gate_proj", "model.language_model.layers.36.mlp.experts.95.gate_proj", "model.language_model.layers.36.mlp.experts.96.gate_proj", "model.language_model.layers.36.mlp.experts.97.gate_proj", "model.language_model.layers.36.mlp.experts.98.gate_proj", "model.language_model.layers.36.mlp.experts.99.gate_proj", "model.language_model.layers.36.mlp.experts.100.gate_proj", "model.language_model.layers.36.mlp.experts.101.gate_proj", "model.language_model.layers.36.mlp.experts.102.gate_proj", "model.language_model.layers.36.mlp.experts.103.gate_proj", "model.language_model.layers.36.mlp.experts.104.gate_proj", "model.language_model.layers.36.mlp.experts.105.gate_proj", "model.language_model.layers.36.mlp.experts.106.gate_proj", "model.language_model.layers.36.mlp.experts.107.gate_proj", "model.language_model.layers.36.mlp.experts.108.gate_proj", "model.language_model.layers.36.mlp.experts.109.gate_proj", "model.language_model.layers.36.mlp.experts.110.gate_proj", "model.language_model.layers.36.mlp.experts.111.gate_proj", "model.language_model.layers.36.mlp.experts.112.gate_proj", "model.language_model.layers.36.mlp.experts.113.gate_proj", "model.language_model.layers.36.mlp.experts.114.gate_proj", "model.language_model.layers.36.mlp.experts.115.gate_proj", "model.language_model.layers.36.mlp.experts.116.gate_proj", "model.language_model.layers.36.mlp.experts.117.gate_proj", "model.language_model.layers.36.mlp.experts.118.gate_proj", "model.language_model.layers.36.mlp.experts.119.gate_proj", "model.language_model.layers.36.mlp.experts.120.gate_proj", "model.language_model.layers.36.mlp.experts.121.gate_proj", "model.language_model.layers.36.mlp.experts.122.gate_proj", "model.language_model.layers.36.mlp.experts.123.gate_proj", "model.language_model.layers.36.mlp.experts.124.gate_proj", "model.language_model.layers.36.mlp.experts.125.gate_proj", "model.language_model.layers.36.mlp.experts.126.gate_proj", "model.language_model.layers.36.mlp.experts.127.gate_proj", "model.language_model.layers.36.mlp.experts.0.up_proj", "model.language_model.layers.36.mlp.experts.1.up_proj", "model.language_model.layers.36.mlp.experts.2.up_proj", "model.language_model.layers.36.mlp.experts.3.up_proj", "model.language_model.layers.36.mlp.experts.4.up_proj", "model.language_model.layers.36.mlp.experts.5.up_proj", "model.language_model.layers.36.mlp.experts.6.up_proj", "model.language_model.layers.36.mlp.experts.7.up_proj", "model.language_model.layers.36.mlp.experts.8.up_proj", "model.language_model.layers.36.mlp.experts.9.up_proj", "model.language_model.layers.36.mlp.experts.10.up_proj", "model.language_model.layers.36.mlp.experts.11.up_proj", "model.language_model.layers.36.mlp.experts.12.up_proj", "model.language_model.layers.36.mlp.experts.13.up_proj", "model.language_model.layers.36.mlp.experts.14.up_proj", "model.language_model.layers.36.mlp.experts.15.up_proj", "model.language_model.layers.36.mlp.experts.16.up_proj", "model.language_model.layers.36.mlp.experts.17.up_proj", "model.language_model.layers.36.mlp.experts.18.up_proj", "model.language_model.layers.36.mlp.experts.19.up_proj", "model.language_model.layers.36.mlp.experts.20.up_proj", "model.language_model.layers.36.mlp.experts.21.up_proj", "model.language_model.layers.36.mlp.experts.22.up_proj", "model.language_model.layers.36.mlp.experts.23.up_proj", "model.language_model.layers.36.mlp.experts.24.up_proj", "model.language_model.layers.36.mlp.experts.25.up_proj", "model.language_model.layers.36.mlp.experts.26.up_proj", "model.language_model.layers.36.mlp.experts.27.up_proj", "model.language_model.layers.36.mlp.experts.28.up_proj", "model.language_model.layers.36.mlp.experts.29.up_proj", "model.language_model.layers.36.mlp.experts.30.up_proj", "model.language_model.layers.36.mlp.experts.31.up_proj", "model.language_model.layers.36.mlp.experts.32.up_proj", "model.language_model.layers.36.mlp.experts.33.up_proj", "model.language_model.layers.36.mlp.experts.34.up_proj", "model.language_model.layers.36.mlp.experts.35.up_proj", "model.language_model.layers.36.mlp.experts.36.up_proj", "model.language_model.layers.36.mlp.experts.37.up_proj", "model.language_model.layers.36.mlp.experts.38.up_proj", "model.language_model.layers.36.mlp.experts.39.up_proj", "model.language_model.layers.36.mlp.experts.40.up_proj", "model.language_model.layers.36.mlp.experts.41.up_proj", "model.language_model.layers.36.mlp.experts.42.up_proj", "model.language_model.layers.36.mlp.experts.43.up_proj", "model.language_model.layers.36.mlp.experts.44.up_proj", "model.language_model.layers.36.mlp.experts.45.up_proj", "model.language_model.layers.36.mlp.experts.46.up_proj", "model.language_model.layers.36.mlp.experts.47.up_proj", "model.language_model.layers.36.mlp.experts.48.up_proj", "model.language_model.layers.36.mlp.experts.49.up_proj", "model.language_model.layers.36.mlp.experts.50.up_proj", "model.language_model.layers.36.mlp.experts.51.up_proj", "model.language_model.layers.36.mlp.experts.52.up_proj", "model.language_model.layers.36.mlp.experts.53.up_proj", "model.language_model.layers.36.mlp.experts.54.up_proj", "model.language_model.layers.36.mlp.experts.55.up_proj", "model.language_model.layers.36.mlp.experts.56.up_proj", "model.language_model.layers.36.mlp.experts.57.up_proj", "model.language_model.layers.36.mlp.experts.58.up_proj", "model.language_model.layers.36.mlp.experts.59.up_proj", "model.language_model.layers.36.mlp.experts.60.up_proj", "model.language_model.layers.36.mlp.experts.61.up_proj", "model.language_model.layers.36.mlp.experts.62.up_proj", "model.language_model.layers.36.mlp.experts.63.up_proj", "model.language_model.layers.36.mlp.experts.64.up_proj", "model.language_model.layers.36.mlp.experts.65.up_proj", "model.language_model.layers.36.mlp.experts.66.up_proj", "model.language_model.layers.36.mlp.experts.67.up_proj", "model.language_model.layers.36.mlp.experts.68.up_proj", "model.language_model.layers.36.mlp.experts.69.up_proj", "model.language_model.layers.36.mlp.experts.70.up_proj", "model.language_model.layers.36.mlp.experts.71.up_proj", "model.language_model.layers.36.mlp.experts.72.up_proj", "model.language_model.layers.36.mlp.experts.73.up_proj", "model.language_model.layers.36.mlp.experts.74.up_proj", "model.language_model.layers.36.mlp.experts.75.up_proj", "model.language_model.layers.36.mlp.experts.76.up_proj", "model.language_model.layers.36.mlp.experts.77.up_proj", "model.language_model.layers.36.mlp.experts.78.up_proj", "model.language_model.layers.36.mlp.experts.79.up_proj", "model.language_model.layers.36.mlp.experts.80.up_proj", "model.language_model.layers.36.mlp.experts.81.up_proj", "model.language_model.layers.36.mlp.experts.82.up_proj", "model.language_model.layers.36.mlp.experts.83.up_proj", "model.language_model.layers.36.mlp.experts.84.up_proj", "model.language_model.layers.36.mlp.experts.85.up_proj", "model.language_model.layers.36.mlp.experts.86.up_proj", "model.language_model.layers.36.mlp.experts.87.up_proj", "model.language_model.layers.36.mlp.experts.88.up_proj", "model.language_model.layers.36.mlp.experts.89.up_proj", "model.language_model.layers.36.mlp.experts.90.up_proj", "model.language_model.layers.36.mlp.experts.91.up_proj", "model.language_model.layers.36.mlp.experts.92.up_proj", "model.language_model.layers.36.mlp.experts.93.up_proj", "model.language_model.layers.36.mlp.experts.94.up_proj", "model.language_model.layers.36.mlp.experts.95.up_proj", "model.language_model.layers.36.mlp.experts.96.up_proj", "model.language_model.layers.36.mlp.experts.97.up_proj", "model.language_model.layers.36.mlp.experts.98.up_proj", "model.language_model.layers.36.mlp.experts.99.up_proj", "model.language_model.layers.36.mlp.experts.100.up_proj", "model.language_model.layers.36.mlp.experts.101.up_proj", "model.language_model.layers.36.mlp.experts.102.up_proj", "model.language_model.layers.36.mlp.experts.103.up_proj", "model.language_model.layers.36.mlp.experts.104.up_proj", "model.language_model.layers.36.mlp.experts.105.up_proj", "model.language_model.layers.36.mlp.experts.106.up_proj", "model.language_model.layers.36.mlp.experts.107.up_proj", "model.language_model.layers.36.mlp.experts.108.up_proj", "model.language_model.layers.36.mlp.experts.109.up_proj", "model.language_model.layers.36.mlp.experts.110.up_proj", "model.language_model.layers.36.mlp.experts.111.up_proj", "model.language_model.layers.36.mlp.experts.112.up_proj", "model.language_model.layers.36.mlp.experts.113.up_proj", "model.language_model.layers.36.mlp.experts.114.up_proj", "model.language_model.layers.36.mlp.experts.115.up_proj", "model.language_model.layers.36.mlp.experts.116.up_proj", "model.language_model.layers.36.mlp.experts.117.up_proj", "model.language_model.layers.36.mlp.experts.118.up_proj", "model.language_model.layers.36.mlp.experts.119.up_proj", "model.language_model.layers.36.mlp.experts.120.up_proj", "model.language_model.layers.36.mlp.experts.121.up_proj", "model.language_model.layers.36.mlp.experts.122.up_proj", "model.language_model.layers.36.mlp.experts.123.up_proj", "model.language_model.layers.36.mlp.experts.124.up_proj", "model.language_model.layers.36.mlp.experts.125.up_proj", "model.language_model.layers.36.mlp.experts.126.up_proj", "model.language_model.layers.36.mlp.experts.127.up_proj", "model.language_model.layers.36.mlp.experts.0.down_proj", "model.language_model.layers.36.mlp.experts.1.down_proj", "model.language_model.layers.36.mlp.experts.2.down_proj", "model.language_model.layers.36.mlp.experts.3.down_proj", "model.language_model.layers.36.mlp.experts.4.down_proj", "model.language_model.layers.36.mlp.experts.5.down_proj", "model.language_model.layers.36.mlp.experts.6.down_proj", "model.language_model.layers.36.mlp.experts.7.down_proj", "model.language_model.layers.36.mlp.experts.8.down_proj", "model.language_model.layers.36.mlp.experts.9.down_proj", "model.language_model.layers.36.mlp.experts.10.down_proj", "model.language_model.layers.36.mlp.experts.11.down_proj", "model.language_model.layers.36.mlp.experts.12.down_proj", "model.language_model.layers.36.mlp.experts.13.down_proj", "model.language_model.layers.36.mlp.experts.14.down_proj", "model.language_model.layers.36.mlp.experts.15.down_proj", "model.language_model.layers.36.mlp.experts.16.down_proj", "model.language_model.layers.36.mlp.experts.17.down_proj", "model.language_model.layers.36.mlp.experts.18.down_proj", "model.language_model.layers.36.mlp.experts.19.down_proj", "model.language_model.layers.36.mlp.experts.20.down_proj", "model.language_model.layers.36.mlp.experts.21.down_proj", "model.language_model.layers.36.mlp.experts.22.down_proj", "model.language_model.layers.36.mlp.experts.23.down_proj", "model.language_model.layers.36.mlp.experts.24.down_proj", "model.language_model.layers.36.mlp.experts.25.down_proj", "model.language_model.layers.36.mlp.experts.26.down_proj", "model.language_model.layers.36.mlp.experts.27.down_proj", "model.language_model.layers.36.mlp.experts.28.down_proj", "model.language_model.layers.36.mlp.experts.29.down_proj", "model.language_model.layers.36.mlp.experts.30.down_proj", "model.language_model.layers.36.mlp.experts.31.down_proj", "model.language_model.layers.36.mlp.experts.32.down_proj", "model.language_model.layers.36.mlp.experts.33.down_proj", "model.language_model.layers.36.mlp.experts.34.down_proj", "model.language_model.layers.36.mlp.experts.35.down_proj", "model.language_model.layers.36.mlp.experts.36.down_proj", "model.language_model.layers.36.mlp.experts.37.down_proj", "model.language_model.layers.36.mlp.experts.38.down_proj", "model.language_model.layers.36.mlp.experts.39.down_proj", "model.language_model.layers.36.mlp.experts.40.down_proj", "model.language_model.layers.36.mlp.experts.41.down_proj", "model.language_model.layers.36.mlp.experts.42.down_proj", "model.language_model.layers.36.mlp.experts.43.down_proj", "model.language_model.layers.36.mlp.experts.44.down_proj", "model.language_model.layers.36.mlp.experts.45.down_proj", "model.language_model.layers.36.mlp.experts.46.down_proj", "model.language_model.layers.36.mlp.experts.47.down_proj", "model.language_model.layers.36.mlp.experts.48.down_proj", "model.language_model.layers.36.mlp.experts.49.down_proj", "model.language_model.layers.36.mlp.experts.50.down_proj", "model.language_model.layers.36.mlp.experts.51.down_proj", "model.language_model.layers.36.mlp.experts.52.down_proj", "model.language_model.layers.36.mlp.experts.53.down_proj", "model.language_model.layers.36.mlp.experts.54.down_proj", "model.language_model.layers.36.mlp.experts.55.down_proj", "model.language_model.layers.36.mlp.experts.56.down_proj", "model.language_model.layers.36.mlp.experts.57.down_proj", "model.language_model.layers.36.mlp.experts.58.down_proj", "model.language_model.layers.36.mlp.experts.59.down_proj", "model.language_model.layers.36.mlp.experts.60.down_proj", "model.language_model.layers.36.mlp.experts.61.down_proj", "model.language_model.layers.36.mlp.experts.62.down_proj", "model.language_model.layers.36.mlp.experts.63.down_proj", "model.language_model.layers.36.mlp.experts.64.down_proj", "model.language_model.layers.36.mlp.experts.65.down_proj", "model.language_model.layers.36.mlp.experts.66.down_proj", "model.language_model.layers.36.mlp.experts.67.down_proj", "model.language_model.layers.36.mlp.experts.68.down_proj", "model.language_model.layers.36.mlp.experts.69.down_proj", "model.language_model.layers.36.mlp.experts.70.down_proj", "model.language_model.layers.36.mlp.experts.71.down_proj", "model.language_model.layers.36.mlp.experts.72.down_proj", "model.language_model.layers.36.mlp.experts.73.down_proj", "model.language_model.layers.36.mlp.experts.74.down_proj", "model.language_model.layers.36.mlp.experts.75.down_proj", "model.language_model.layers.36.mlp.experts.76.down_proj", "model.language_model.layers.36.mlp.experts.77.down_proj", "model.language_model.layers.36.mlp.experts.78.down_proj", "model.language_model.layers.36.mlp.experts.79.down_proj", "model.language_model.layers.36.mlp.experts.80.down_proj", "model.language_model.layers.36.mlp.experts.81.down_proj", "model.language_model.layers.36.mlp.experts.82.down_proj", "model.language_model.layers.36.mlp.experts.83.down_proj", "model.language_model.layers.36.mlp.experts.84.down_proj", "model.language_model.layers.36.mlp.experts.85.down_proj", "model.language_model.layers.36.mlp.experts.86.down_proj", "model.language_model.layers.36.mlp.experts.87.down_proj", "model.language_model.layers.36.mlp.experts.88.down_proj", "model.language_model.layers.36.mlp.experts.89.down_proj", "model.language_model.layers.36.mlp.experts.90.down_proj", "model.language_model.layers.36.mlp.experts.91.down_proj", "model.language_model.layers.36.mlp.experts.92.down_proj", "model.language_model.layers.36.mlp.experts.93.down_proj", "model.language_model.layers.36.mlp.experts.94.down_proj", "model.language_model.layers.36.mlp.experts.95.down_proj", "model.language_model.layers.36.mlp.experts.96.down_proj", "model.language_model.layers.36.mlp.experts.97.down_proj", "model.language_model.layers.36.mlp.experts.98.down_proj", "model.language_model.layers.36.mlp.experts.99.down_proj", "model.language_model.layers.36.mlp.experts.100.down_proj", "model.language_model.layers.36.mlp.experts.101.down_proj", "model.language_model.layers.36.mlp.experts.102.down_proj", "model.language_model.layers.36.mlp.experts.103.down_proj", "model.language_model.layers.36.mlp.experts.104.down_proj", "model.language_model.layers.36.mlp.experts.105.down_proj", "model.language_model.layers.36.mlp.experts.106.down_proj", "model.language_model.layers.36.mlp.experts.107.down_proj", "model.language_model.layers.36.mlp.experts.108.down_proj", "model.language_model.layers.36.mlp.experts.109.down_proj", "model.language_model.layers.36.mlp.experts.110.down_proj", "model.language_model.layers.36.mlp.experts.111.down_proj", "model.language_model.layers.36.mlp.experts.112.down_proj", "model.language_model.layers.36.mlp.experts.113.down_proj", "model.language_model.layers.36.mlp.experts.114.down_proj", "model.language_model.layers.36.mlp.experts.115.down_proj", "model.language_model.layers.36.mlp.experts.116.down_proj", "model.language_model.layers.36.mlp.experts.117.down_proj", "model.language_model.layers.36.mlp.experts.118.down_proj", "model.language_model.layers.36.mlp.experts.119.down_proj", "model.language_model.layers.36.mlp.experts.120.down_proj", "model.language_model.layers.36.mlp.experts.121.down_proj", "model.language_model.layers.36.mlp.experts.122.down_proj", "model.language_model.layers.36.mlp.experts.123.down_proj", "model.language_model.layers.36.mlp.experts.124.down_proj", "model.language_model.layers.36.mlp.experts.125.down_proj", "model.language_model.layers.36.mlp.experts.126.down_proj", "model.language_model.layers.36.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.0014683872461318942, "dbits": 603979776 } ] }, { "idx": 74, "layers": [ "model.language_model.layers.37.self_attn.q_proj", "model.language_model.layers.37.self_attn.k_proj", "model.language_model.layers.37.self_attn.v_proj", "model.language_model.layers.37.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0005391523241996821, "dbits": 18874368 } ] }, { "idx": 75, "layers": [ "model.language_model.layers.37.mlp.experts.0.gate_proj", "model.language_model.layers.37.mlp.experts.1.gate_proj", "model.language_model.layers.37.mlp.experts.2.gate_proj", "model.language_model.layers.37.mlp.experts.3.gate_proj", "model.language_model.layers.37.mlp.experts.4.gate_proj", "model.language_model.layers.37.mlp.experts.5.gate_proj", "model.language_model.layers.37.mlp.experts.6.gate_proj", "model.language_model.layers.37.mlp.experts.7.gate_proj", "model.language_model.layers.37.mlp.experts.8.gate_proj", "model.language_model.layers.37.mlp.experts.9.gate_proj", "model.language_model.layers.37.mlp.experts.10.gate_proj", "model.language_model.layers.37.mlp.experts.11.gate_proj", "model.language_model.layers.37.mlp.experts.12.gate_proj", "model.language_model.layers.37.mlp.experts.13.gate_proj", "model.language_model.layers.37.mlp.experts.14.gate_proj", "model.language_model.layers.37.mlp.experts.15.gate_proj", "model.language_model.layers.37.mlp.experts.16.gate_proj", "model.language_model.layers.37.mlp.experts.17.gate_proj", "model.language_model.layers.37.mlp.experts.18.gate_proj", "model.language_model.layers.37.mlp.experts.19.gate_proj", "model.language_model.layers.37.mlp.experts.20.gate_proj", "model.language_model.layers.37.mlp.experts.21.gate_proj", "model.language_model.layers.37.mlp.experts.22.gate_proj", "model.language_model.layers.37.mlp.experts.23.gate_proj", "model.language_model.layers.37.mlp.experts.24.gate_proj", "model.language_model.layers.37.mlp.experts.25.gate_proj", "model.language_model.layers.37.mlp.experts.26.gate_proj", "model.language_model.layers.37.mlp.experts.27.gate_proj", "model.language_model.layers.37.mlp.experts.28.gate_proj", "model.language_model.layers.37.mlp.experts.29.gate_proj", "model.language_model.layers.37.mlp.experts.30.gate_proj", "model.language_model.layers.37.mlp.experts.31.gate_proj", "model.language_model.layers.37.mlp.experts.32.gate_proj", "model.language_model.layers.37.mlp.experts.33.gate_proj", "model.language_model.layers.37.mlp.experts.34.gate_proj", "model.language_model.layers.37.mlp.experts.35.gate_proj", "model.language_model.layers.37.mlp.experts.36.gate_proj", "model.language_model.layers.37.mlp.experts.37.gate_proj", "model.language_model.layers.37.mlp.experts.38.gate_proj", "model.language_model.layers.37.mlp.experts.39.gate_proj", "model.language_model.layers.37.mlp.experts.40.gate_proj", "model.language_model.layers.37.mlp.experts.41.gate_proj", "model.language_model.layers.37.mlp.experts.42.gate_proj", "model.language_model.layers.37.mlp.experts.43.gate_proj", "model.language_model.layers.37.mlp.experts.44.gate_proj", "model.language_model.layers.37.mlp.experts.45.gate_proj", "model.language_model.layers.37.mlp.experts.46.gate_proj", "model.language_model.layers.37.mlp.experts.47.gate_proj", "model.language_model.layers.37.mlp.experts.48.gate_proj", "model.language_model.layers.37.mlp.experts.49.gate_proj", "model.language_model.layers.37.mlp.experts.50.gate_proj", "model.language_model.layers.37.mlp.experts.51.gate_proj", "model.language_model.layers.37.mlp.experts.52.gate_proj", "model.language_model.layers.37.mlp.experts.53.gate_proj", "model.language_model.layers.37.mlp.experts.54.gate_proj", "model.language_model.layers.37.mlp.experts.55.gate_proj", "model.language_model.layers.37.mlp.experts.56.gate_proj", "model.language_model.layers.37.mlp.experts.57.gate_proj", "model.language_model.layers.37.mlp.experts.58.gate_proj", "model.language_model.layers.37.mlp.experts.59.gate_proj", "model.language_model.layers.37.mlp.experts.60.gate_proj", "model.language_model.layers.37.mlp.experts.61.gate_proj", "model.language_model.layers.37.mlp.experts.62.gate_proj", "model.language_model.layers.37.mlp.experts.63.gate_proj", "model.language_model.layers.37.mlp.experts.64.gate_proj", "model.language_model.layers.37.mlp.experts.65.gate_proj", "model.language_model.layers.37.mlp.experts.66.gate_proj", "model.language_model.layers.37.mlp.experts.67.gate_proj", "model.language_model.layers.37.mlp.experts.68.gate_proj", "model.language_model.layers.37.mlp.experts.69.gate_proj", "model.language_model.layers.37.mlp.experts.70.gate_proj", "model.language_model.layers.37.mlp.experts.71.gate_proj", "model.language_model.layers.37.mlp.experts.72.gate_proj", "model.language_model.layers.37.mlp.experts.73.gate_proj", "model.language_model.layers.37.mlp.experts.74.gate_proj", "model.language_model.layers.37.mlp.experts.75.gate_proj", "model.language_model.layers.37.mlp.experts.76.gate_proj", "model.language_model.layers.37.mlp.experts.77.gate_proj", "model.language_model.layers.37.mlp.experts.78.gate_proj", "model.language_model.layers.37.mlp.experts.79.gate_proj", "model.language_model.layers.37.mlp.experts.80.gate_proj", "model.language_model.layers.37.mlp.experts.81.gate_proj", "model.language_model.layers.37.mlp.experts.82.gate_proj", "model.language_model.layers.37.mlp.experts.83.gate_proj", "model.language_model.layers.37.mlp.experts.84.gate_proj", "model.language_model.layers.37.mlp.experts.85.gate_proj", "model.language_model.layers.37.mlp.experts.86.gate_proj", "model.language_model.layers.37.mlp.experts.87.gate_proj", "model.language_model.layers.37.mlp.experts.88.gate_proj", "model.language_model.layers.37.mlp.experts.89.gate_proj", "model.language_model.layers.37.mlp.experts.90.gate_proj", "model.language_model.layers.37.mlp.experts.91.gate_proj", "model.language_model.layers.37.mlp.experts.92.gate_proj", "model.language_model.layers.37.mlp.experts.93.gate_proj", "model.language_model.layers.37.mlp.experts.94.gate_proj", "model.language_model.layers.37.mlp.experts.95.gate_proj", "model.language_model.layers.37.mlp.experts.96.gate_proj", "model.language_model.layers.37.mlp.experts.97.gate_proj", "model.language_model.layers.37.mlp.experts.98.gate_proj", "model.language_model.layers.37.mlp.experts.99.gate_proj", "model.language_model.layers.37.mlp.experts.100.gate_proj", "model.language_model.layers.37.mlp.experts.101.gate_proj", "model.language_model.layers.37.mlp.experts.102.gate_proj", "model.language_model.layers.37.mlp.experts.103.gate_proj", "model.language_model.layers.37.mlp.experts.104.gate_proj", "model.language_model.layers.37.mlp.experts.105.gate_proj", "model.language_model.layers.37.mlp.experts.106.gate_proj", "model.language_model.layers.37.mlp.experts.107.gate_proj", "model.language_model.layers.37.mlp.experts.108.gate_proj", "model.language_model.layers.37.mlp.experts.109.gate_proj", "model.language_model.layers.37.mlp.experts.110.gate_proj", "model.language_model.layers.37.mlp.experts.111.gate_proj", "model.language_model.layers.37.mlp.experts.112.gate_proj", "model.language_model.layers.37.mlp.experts.113.gate_proj", "model.language_model.layers.37.mlp.experts.114.gate_proj", "model.language_model.layers.37.mlp.experts.115.gate_proj", "model.language_model.layers.37.mlp.experts.116.gate_proj", "model.language_model.layers.37.mlp.experts.117.gate_proj", "model.language_model.layers.37.mlp.experts.118.gate_proj", "model.language_model.layers.37.mlp.experts.119.gate_proj", "model.language_model.layers.37.mlp.experts.120.gate_proj", "model.language_model.layers.37.mlp.experts.121.gate_proj", "model.language_model.layers.37.mlp.experts.122.gate_proj", "model.language_model.layers.37.mlp.experts.123.gate_proj", "model.language_model.layers.37.mlp.experts.124.gate_proj", "model.language_model.layers.37.mlp.experts.125.gate_proj", "model.language_model.layers.37.mlp.experts.126.gate_proj", "model.language_model.layers.37.mlp.experts.127.gate_proj", "model.language_model.layers.37.mlp.experts.0.up_proj", "model.language_model.layers.37.mlp.experts.1.up_proj", "model.language_model.layers.37.mlp.experts.2.up_proj", "model.language_model.layers.37.mlp.experts.3.up_proj", "model.language_model.layers.37.mlp.experts.4.up_proj", "model.language_model.layers.37.mlp.experts.5.up_proj", "model.language_model.layers.37.mlp.experts.6.up_proj", "model.language_model.layers.37.mlp.experts.7.up_proj", "model.language_model.layers.37.mlp.experts.8.up_proj", "model.language_model.layers.37.mlp.experts.9.up_proj", "model.language_model.layers.37.mlp.experts.10.up_proj", "model.language_model.layers.37.mlp.experts.11.up_proj", "model.language_model.layers.37.mlp.experts.12.up_proj", "model.language_model.layers.37.mlp.experts.13.up_proj", "model.language_model.layers.37.mlp.experts.14.up_proj", "model.language_model.layers.37.mlp.experts.15.up_proj", "model.language_model.layers.37.mlp.experts.16.up_proj", "model.language_model.layers.37.mlp.experts.17.up_proj", "model.language_model.layers.37.mlp.experts.18.up_proj", "model.language_model.layers.37.mlp.experts.19.up_proj", "model.language_model.layers.37.mlp.experts.20.up_proj", "model.language_model.layers.37.mlp.experts.21.up_proj", "model.language_model.layers.37.mlp.experts.22.up_proj", "model.language_model.layers.37.mlp.experts.23.up_proj", "model.language_model.layers.37.mlp.experts.24.up_proj", "model.language_model.layers.37.mlp.experts.25.up_proj", "model.language_model.layers.37.mlp.experts.26.up_proj", "model.language_model.layers.37.mlp.experts.27.up_proj", "model.language_model.layers.37.mlp.experts.28.up_proj", "model.language_model.layers.37.mlp.experts.29.up_proj", "model.language_model.layers.37.mlp.experts.30.up_proj", "model.language_model.layers.37.mlp.experts.31.up_proj", "model.language_model.layers.37.mlp.experts.32.up_proj", "model.language_model.layers.37.mlp.experts.33.up_proj", "model.language_model.layers.37.mlp.experts.34.up_proj", "model.language_model.layers.37.mlp.experts.35.up_proj", "model.language_model.layers.37.mlp.experts.36.up_proj", "model.language_model.layers.37.mlp.experts.37.up_proj", "model.language_model.layers.37.mlp.experts.38.up_proj", "model.language_model.layers.37.mlp.experts.39.up_proj", "model.language_model.layers.37.mlp.experts.40.up_proj", "model.language_model.layers.37.mlp.experts.41.up_proj", "model.language_model.layers.37.mlp.experts.42.up_proj", "model.language_model.layers.37.mlp.experts.43.up_proj", "model.language_model.layers.37.mlp.experts.44.up_proj", "model.language_model.layers.37.mlp.experts.45.up_proj", "model.language_model.layers.37.mlp.experts.46.up_proj", "model.language_model.layers.37.mlp.experts.47.up_proj", "model.language_model.layers.37.mlp.experts.48.up_proj", "model.language_model.layers.37.mlp.experts.49.up_proj", "model.language_model.layers.37.mlp.experts.50.up_proj", "model.language_model.layers.37.mlp.experts.51.up_proj", "model.language_model.layers.37.mlp.experts.52.up_proj", "model.language_model.layers.37.mlp.experts.53.up_proj", "model.language_model.layers.37.mlp.experts.54.up_proj", "model.language_model.layers.37.mlp.experts.55.up_proj", "model.language_model.layers.37.mlp.experts.56.up_proj", "model.language_model.layers.37.mlp.experts.57.up_proj", "model.language_model.layers.37.mlp.experts.58.up_proj", "model.language_model.layers.37.mlp.experts.59.up_proj", "model.language_model.layers.37.mlp.experts.60.up_proj", "model.language_model.layers.37.mlp.experts.61.up_proj", "model.language_model.layers.37.mlp.experts.62.up_proj", "model.language_model.layers.37.mlp.experts.63.up_proj", "model.language_model.layers.37.mlp.experts.64.up_proj", "model.language_model.layers.37.mlp.experts.65.up_proj", "model.language_model.layers.37.mlp.experts.66.up_proj", "model.language_model.layers.37.mlp.experts.67.up_proj", "model.language_model.layers.37.mlp.experts.68.up_proj", "model.language_model.layers.37.mlp.experts.69.up_proj", "model.language_model.layers.37.mlp.experts.70.up_proj", "model.language_model.layers.37.mlp.experts.71.up_proj", "model.language_model.layers.37.mlp.experts.72.up_proj", "model.language_model.layers.37.mlp.experts.73.up_proj", "model.language_model.layers.37.mlp.experts.74.up_proj", "model.language_model.layers.37.mlp.experts.75.up_proj", "model.language_model.layers.37.mlp.experts.76.up_proj", "model.language_model.layers.37.mlp.experts.77.up_proj", "model.language_model.layers.37.mlp.experts.78.up_proj", "model.language_model.layers.37.mlp.experts.79.up_proj", "model.language_model.layers.37.mlp.experts.80.up_proj", "model.language_model.layers.37.mlp.experts.81.up_proj", "model.language_model.layers.37.mlp.experts.82.up_proj", "model.language_model.layers.37.mlp.experts.83.up_proj", "model.language_model.layers.37.mlp.experts.84.up_proj", "model.language_model.layers.37.mlp.experts.85.up_proj", "model.language_model.layers.37.mlp.experts.86.up_proj", "model.language_model.layers.37.mlp.experts.87.up_proj", "model.language_model.layers.37.mlp.experts.88.up_proj", "model.language_model.layers.37.mlp.experts.89.up_proj", "model.language_model.layers.37.mlp.experts.90.up_proj", "model.language_model.layers.37.mlp.experts.91.up_proj", "model.language_model.layers.37.mlp.experts.92.up_proj", "model.language_model.layers.37.mlp.experts.93.up_proj", "model.language_model.layers.37.mlp.experts.94.up_proj", "model.language_model.layers.37.mlp.experts.95.up_proj", "model.language_model.layers.37.mlp.experts.96.up_proj", "model.language_model.layers.37.mlp.experts.97.up_proj", "model.language_model.layers.37.mlp.experts.98.up_proj", "model.language_model.layers.37.mlp.experts.99.up_proj", "model.language_model.layers.37.mlp.experts.100.up_proj", "model.language_model.layers.37.mlp.experts.101.up_proj", "model.language_model.layers.37.mlp.experts.102.up_proj", "model.language_model.layers.37.mlp.experts.103.up_proj", "model.language_model.layers.37.mlp.experts.104.up_proj", "model.language_model.layers.37.mlp.experts.105.up_proj", "model.language_model.layers.37.mlp.experts.106.up_proj", "model.language_model.layers.37.mlp.experts.107.up_proj", "model.language_model.layers.37.mlp.experts.108.up_proj", "model.language_model.layers.37.mlp.experts.109.up_proj", "model.language_model.layers.37.mlp.experts.110.up_proj", "model.language_model.layers.37.mlp.experts.111.up_proj", "model.language_model.layers.37.mlp.experts.112.up_proj", "model.language_model.layers.37.mlp.experts.113.up_proj", "model.language_model.layers.37.mlp.experts.114.up_proj", "model.language_model.layers.37.mlp.experts.115.up_proj", "model.language_model.layers.37.mlp.experts.116.up_proj", "model.language_model.layers.37.mlp.experts.117.up_proj", "model.language_model.layers.37.mlp.experts.118.up_proj", "model.language_model.layers.37.mlp.experts.119.up_proj", "model.language_model.layers.37.mlp.experts.120.up_proj", "model.language_model.layers.37.mlp.experts.121.up_proj", "model.language_model.layers.37.mlp.experts.122.up_proj", "model.language_model.layers.37.mlp.experts.123.up_proj", "model.language_model.layers.37.mlp.experts.124.up_proj", "model.language_model.layers.37.mlp.experts.125.up_proj", "model.language_model.layers.37.mlp.experts.126.up_proj", "model.language_model.layers.37.mlp.experts.127.up_proj", "model.language_model.layers.37.mlp.experts.0.down_proj", "model.language_model.layers.37.mlp.experts.1.down_proj", "model.language_model.layers.37.mlp.experts.2.down_proj", "model.language_model.layers.37.mlp.experts.3.down_proj", "model.language_model.layers.37.mlp.experts.4.down_proj", "model.language_model.layers.37.mlp.experts.5.down_proj", "model.language_model.layers.37.mlp.experts.6.down_proj", "model.language_model.layers.37.mlp.experts.7.down_proj", "model.language_model.layers.37.mlp.experts.8.down_proj", "model.language_model.layers.37.mlp.experts.9.down_proj", "model.language_model.layers.37.mlp.experts.10.down_proj", "model.language_model.layers.37.mlp.experts.11.down_proj", "model.language_model.layers.37.mlp.experts.12.down_proj", "model.language_model.layers.37.mlp.experts.13.down_proj", "model.language_model.layers.37.mlp.experts.14.down_proj", "model.language_model.layers.37.mlp.experts.15.down_proj", "model.language_model.layers.37.mlp.experts.16.down_proj", "model.language_model.layers.37.mlp.experts.17.down_proj", "model.language_model.layers.37.mlp.experts.18.down_proj", "model.language_model.layers.37.mlp.experts.19.down_proj", "model.language_model.layers.37.mlp.experts.20.down_proj", "model.language_model.layers.37.mlp.experts.21.down_proj", "model.language_model.layers.37.mlp.experts.22.down_proj", "model.language_model.layers.37.mlp.experts.23.down_proj", "model.language_model.layers.37.mlp.experts.24.down_proj", "model.language_model.layers.37.mlp.experts.25.down_proj", "model.language_model.layers.37.mlp.experts.26.down_proj", "model.language_model.layers.37.mlp.experts.27.down_proj", "model.language_model.layers.37.mlp.experts.28.down_proj", "model.language_model.layers.37.mlp.experts.29.down_proj", "model.language_model.layers.37.mlp.experts.30.down_proj", "model.language_model.layers.37.mlp.experts.31.down_proj", "model.language_model.layers.37.mlp.experts.32.down_proj", "model.language_model.layers.37.mlp.experts.33.down_proj", "model.language_model.layers.37.mlp.experts.34.down_proj", "model.language_model.layers.37.mlp.experts.35.down_proj", "model.language_model.layers.37.mlp.experts.36.down_proj", "model.language_model.layers.37.mlp.experts.37.down_proj", "model.language_model.layers.37.mlp.experts.38.down_proj", "model.language_model.layers.37.mlp.experts.39.down_proj", "model.language_model.layers.37.mlp.experts.40.down_proj", "model.language_model.layers.37.mlp.experts.41.down_proj", "model.language_model.layers.37.mlp.experts.42.down_proj", "model.language_model.layers.37.mlp.experts.43.down_proj", "model.language_model.layers.37.mlp.experts.44.down_proj", "model.language_model.layers.37.mlp.experts.45.down_proj", "model.language_model.layers.37.mlp.experts.46.down_proj", "model.language_model.layers.37.mlp.experts.47.down_proj", "model.language_model.layers.37.mlp.experts.48.down_proj", "model.language_model.layers.37.mlp.experts.49.down_proj", "model.language_model.layers.37.mlp.experts.50.down_proj", "model.language_model.layers.37.mlp.experts.51.down_proj", "model.language_model.layers.37.mlp.experts.52.down_proj", "model.language_model.layers.37.mlp.experts.53.down_proj", "model.language_model.layers.37.mlp.experts.54.down_proj", "model.language_model.layers.37.mlp.experts.55.down_proj", "model.language_model.layers.37.mlp.experts.56.down_proj", "model.language_model.layers.37.mlp.experts.57.down_proj", "model.language_model.layers.37.mlp.experts.58.down_proj", "model.language_model.layers.37.mlp.experts.59.down_proj", "model.language_model.layers.37.mlp.experts.60.down_proj", "model.language_model.layers.37.mlp.experts.61.down_proj", "model.language_model.layers.37.mlp.experts.62.down_proj", "model.language_model.layers.37.mlp.experts.63.down_proj", "model.language_model.layers.37.mlp.experts.64.down_proj", "model.language_model.layers.37.mlp.experts.65.down_proj", "model.language_model.layers.37.mlp.experts.66.down_proj", "model.language_model.layers.37.mlp.experts.67.down_proj", "model.language_model.layers.37.mlp.experts.68.down_proj", "model.language_model.layers.37.mlp.experts.69.down_proj", "model.language_model.layers.37.mlp.experts.70.down_proj", "model.language_model.layers.37.mlp.experts.71.down_proj", "model.language_model.layers.37.mlp.experts.72.down_proj", "model.language_model.layers.37.mlp.experts.73.down_proj", "model.language_model.layers.37.mlp.experts.74.down_proj", "model.language_model.layers.37.mlp.experts.75.down_proj", "model.language_model.layers.37.mlp.experts.76.down_proj", "model.language_model.layers.37.mlp.experts.77.down_proj", "model.language_model.layers.37.mlp.experts.78.down_proj", "model.language_model.layers.37.mlp.experts.79.down_proj", "model.language_model.layers.37.mlp.experts.80.down_proj", "model.language_model.layers.37.mlp.experts.81.down_proj", "model.language_model.layers.37.mlp.experts.82.down_proj", "model.language_model.layers.37.mlp.experts.83.down_proj", "model.language_model.layers.37.mlp.experts.84.down_proj", "model.language_model.layers.37.mlp.experts.85.down_proj", "model.language_model.layers.37.mlp.experts.86.down_proj", "model.language_model.layers.37.mlp.experts.87.down_proj", "model.language_model.layers.37.mlp.experts.88.down_proj", "model.language_model.layers.37.mlp.experts.89.down_proj", "model.language_model.layers.37.mlp.experts.90.down_proj", "model.language_model.layers.37.mlp.experts.91.down_proj", "model.language_model.layers.37.mlp.experts.92.down_proj", "model.language_model.layers.37.mlp.experts.93.down_proj", "model.language_model.layers.37.mlp.experts.94.down_proj", "model.language_model.layers.37.mlp.experts.95.down_proj", "model.language_model.layers.37.mlp.experts.96.down_proj", "model.language_model.layers.37.mlp.experts.97.down_proj", "model.language_model.layers.37.mlp.experts.98.down_proj", "model.language_model.layers.37.mlp.experts.99.down_proj", "model.language_model.layers.37.mlp.experts.100.down_proj", "model.language_model.layers.37.mlp.experts.101.down_proj", "model.language_model.layers.37.mlp.experts.102.down_proj", "model.language_model.layers.37.mlp.experts.103.down_proj", "model.language_model.layers.37.mlp.experts.104.down_proj", "model.language_model.layers.37.mlp.experts.105.down_proj", "model.language_model.layers.37.mlp.experts.106.down_proj", "model.language_model.layers.37.mlp.experts.107.down_proj", "model.language_model.layers.37.mlp.experts.108.down_proj", "model.language_model.layers.37.mlp.experts.109.down_proj", "model.language_model.layers.37.mlp.experts.110.down_proj", "model.language_model.layers.37.mlp.experts.111.down_proj", "model.language_model.layers.37.mlp.experts.112.down_proj", "model.language_model.layers.37.mlp.experts.113.down_proj", "model.language_model.layers.37.mlp.experts.114.down_proj", "model.language_model.layers.37.mlp.experts.115.down_proj", "model.language_model.layers.37.mlp.experts.116.down_proj", "model.language_model.layers.37.mlp.experts.117.down_proj", "model.language_model.layers.37.mlp.experts.118.down_proj", "model.language_model.layers.37.mlp.experts.119.down_proj", "model.language_model.layers.37.mlp.experts.120.down_proj", "model.language_model.layers.37.mlp.experts.121.down_proj", "model.language_model.layers.37.mlp.experts.122.down_proj", "model.language_model.layers.37.mlp.experts.123.down_proj", "model.language_model.layers.37.mlp.experts.124.down_proj", "model.language_model.layers.37.mlp.experts.125.down_proj", "model.language_model.layers.37.mlp.experts.126.down_proj", "model.language_model.layers.37.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.00047669699415564537, "dbits": 603979776 } ] }, { "idx": 76, "layers": [ "model.language_model.layers.38.self_attn.q_proj", "model.language_model.layers.38.self_attn.k_proj", "model.language_model.layers.38.self_attn.v_proj", "model.language_model.layers.38.self_attn.o_proj" ], "candidates": [ { "dkld": 0.00013495562598109245, "dbits": 18874368 } ] }, { "idx": 77, "layers": [ "model.language_model.layers.38.mlp.experts.0.gate_proj", "model.language_model.layers.38.mlp.experts.1.gate_proj", "model.language_model.layers.38.mlp.experts.2.gate_proj", "model.language_model.layers.38.mlp.experts.3.gate_proj", "model.language_model.layers.38.mlp.experts.4.gate_proj", "model.language_model.layers.38.mlp.experts.5.gate_proj", "model.language_model.layers.38.mlp.experts.6.gate_proj", "model.language_model.layers.38.mlp.experts.7.gate_proj", "model.language_model.layers.38.mlp.experts.8.gate_proj", "model.language_model.layers.38.mlp.experts.9.gate_proj", "model.language_model.layers.38.mlp.experts.10.gate_proj", "model.language_model.layers.38.mlp.experts.11.gate_proj", "model.language_model.layers.38.mlp.experts.12.gate_proj", "model.language_model.layers.38.mlp.experts.13.gate_proj", "model.language_model.layers.38.mlp.experts.14.gate_proj", "model.language_model.layers.38.mlp.experts.15.gate_proj", "model.language_model.layers.38.mlp.experts.16.gate_proj", "model.language_model.layers.38.mlp.experts.17.gate_proj", "model.language_model.layers.38.mlp.experts.18.gate_proj", "model.language_model.layers.38.mlp.experts.19.gate_proj", "model.language_model.layers.38.mlp.experts.20.gate_proj", "model.language_model.layers.38.mlp.experts.21.gate_proj", "model.language_model.layers.38.mlp.experts.22.gate_proj", "model.language_model.layers.38.mlp.experts.23.gate_proj", "model.language_model.layers.38.mlp.experts.24.gate_proj", "model.language_model.layers.38.mlp.experts.25.gate_proj", "model.language_model.layers.38.mlp.experts.26.gate_proj", "model.language_model.layers.38.mlp.experts.27.gate_proj", "model.language_model.layers.38.mlp.experts.28.gate_proj", "model.language_model.layers.38.mlp.experts.29.gate_proj", "model.language_model.layers.38.mlp.experts.30.gate_proj", "model.language_model.layers.38.mlp.experts.31.gate_proj", "model.language_model.layers.38.mlp.experts.32.gate_proj", "model.language_model.layers.38.mlp.experts.33.gate_proj", "model.language_model.layers.38.mlp.experts.34.gate_proj", "model.language_model.layers.38.mlp.experts.35.gate_proj", "model.language_model.layers.38.mlp.experts.36.gate_proj", "model.language_model.layers.38.mlp.experts.37.gate_proj", "model.language_model.layers.38.mlp.experts.38.gate_proj", "model.language_model.layers.38.mlp.experts.39.gate_proj", "model.language_model.layers.38.mlp.experts.40.gate_proj", "model.language_model.layers.38.mlp.experts.41.gate_proj", "model.language_model.layers.38.mlp.experts.42.gate_proj", "model.language_model.layers.38.mlp.experts.43.gate_proj", "model.language_model.layers.38.mlp.experts.44.gate_proj", "model.language_model.layers.38.mlp.experts.45.gate_proj", "model.language_model.layers.38.mlp.experts.46.gate_proj", "model.language_model.layers.38.mlp.experts.47.gate_proj", "model.language_model.layers.38.mlp.experts.48.gate_proj", "model.language_model.layers.38.mlp.experts.49.gate_proj", "model.language_model.layers.38.mlp.experts.50.gate_proj", "model.language_model.layers.38.mlp.experts.51.gate_proj", "model.language_model.layers.38.mlp.experts.52.gate_proj", "model.language_model.layers.38.mlp.experts.53.gate_proj", "model.language_model.layers.38.mlp.experts.54.gate_proj", "model.language_model.layers.38.mlp.experts.55.gate_proj", "model.language_model.layers.38.mlp.experts.56.gate_proj", "model.language_model.layers.38.mlp.experts.57.gate_proj", "model.language_model.layers.38.mlp.experts.58.gate_proj", "model.language_model.layers.38.mlp.experts.59.gate_proj", "model.language_model.layers.38.mlp.experts.60.gate_proj", "model.language_model.layers.38.mlp.experts.61.gate_proj", "model.language_model.layers.38.mlp.experts.62.gate_proj", "model.language_model.layers.38.mlp.experts.63.gate_proj", "model.language_model.layers.38.mlp.experts.64.gate_proj", "model.language_model.layers.38.mlp.experts.65.gate_proj", "model.language_model.layers.38.mlp.experts.66.gate_proj", "model.language_model.layers.38.mlp.experts.67.gate_proj", "model.language_model.layers.38.mlp.experts.68.gate_proj", "model.language_model.layers.38.mlp.experts.69.gate_proj", "model.language_model.layers.38.mlp.experts.70.gate_proj", "model.language_model.layers.38.mlp.experts.71.gate_proj", "model.language_model.layers.38.mlp.experts.72.gate_proj", "model.language_model.layers.38.mlp.experts.73.gate_proj", "model.language_model.layers.38.mlp.experts.74.gate_proj", "model.language_model.layers.38.mlp.experts.75.gate_proj", "model.language_model.layers.38.mlp.experts.76.gate_proj", "model.language_model.layers.38.mlp.experts.77.gate_proj", "model.language_model.layers.38.mlp.experts.78.gate_proj", "model.language_model.layers.38.mlp.experts.79.gate_proj", "model.language_model.layers.38.mlp.experts.80.gate_proj", "model.language_model.layers.38.mlp.experts.81.gate_proj", "model.language_model.layers.38.mlp.experts.82.gate_proj", "model.language_model.layers.38.mlp.experts.83.gate_proj", "model.language_model.layers.38.mlp.experts.84.gate_proj", "model.language_model.layers.38.mlp.experts.85.gate_proj", "model.language_model.layers.38.mlp.experts.86.gate_proj", "model.language_model.layers.38.mlp.experts.87.gate_proj", "model.language_model.layers.38.mlp.experts.88.gate_proj", "model.language_model.layers.38.mlp.experts.89.gate_proj", "model.language_model.layers.38.mlp.experts.90.gate_proj", "model.language_model.layers.38.mlp.experts.91.gate_proj", "model.language_model.layers.38.mlp.experts.92.gate_proj", "model.language_model.layers.38.mlp.experts.93.gate_proj", "model.language_model.layers.38.mlp.experts.94.gate_proj", "model.language_model.layers.38.mlp.experts.95.gate_proj", "model.language_model.layers.38.mlp.experts.96.gate_proj", "model.language_model.layers.38.mlp.experts.97.gate_proj", "model.language_model.layers.38.mlp.experts.98.gate_proj", "model.language_model.layers.38.mlp.experts.99.gate_proj", "model.language_model.layers.38.mlp.experts.100.gate_proj", "model.language_model.layers.38.mlp.experts.101.gate_proj", "model.language_model.layers.38.mlp.experts.102.gate_proj", "model.language_model.layers.38.mlp.experts.103.gate_proj", "model.language_model.layers.38.mlp.experts.104.gate_proj", "model.language_model.layers.38.mlp.experts.105.gate_proj", "model.language_model.layers.38.mlp.experts.106.gate_proj", "model.language_model.layers.38.mlp.experts.107.gate_proj", "model.language_model.layers.38.mlp.experts.108.gate_proj", "model.language_model.layers.38.mlp.experts.109.gate_proj", "model.language_model.layers.38.mlp.experts.110.gate_proj", "model.language_model.layers.38.mlp.experts.111.gate_proj", "model.language_model.layers.38.mlp.experts.112.gate_proj", "model.language_model.layers.38.mlp.experts.113.gate_proj", "model.language_model.layers.38.mlp.experts.114.gate_proj", "model.language_model.layers.38.mlp.experts.115.gate_proj", "model.language_model.layers.38.mlp.experts.116.gate_proj", "model.language_model.layers.38.mlp.experts.117.gate_proj", "model.language_model.layers.38.mlp.experts.118.gate_proj", "model.language_model.layers.38.mlp.experts.119.gate_proj", "model.language_model.layers.38.mlp.experts.120.gate_proj", "model.language_model.layers.38.mlp.experts.121.gate_proj", "model.language_model.layers.38.mlp.experts.122.gate_proj", "model.language_model.layers.38.mlp.experts.123.gate_proj", "model.language_model.layers.38.mlp.experts.124.gate_proj", "model.language_model.layers.38.mlp.experts.125.gate_proj", "model.language_model.layers.38.mlp.experts.126.gate_proj", "model.language_model.layers.38.mlp.experts.127.gate_proj", "model.language_model.layers.38.mlp.experts.0.up_proj", "model.language_model.layers.38.mlp.experts.1.up_proj", "model.language_model.layers.38.mlp.experts.2.up_proj", "model.language_model.layers.38.mlp.experts.3.up_proj", "model.language_model.layers.38.mlp.experts.4.up_proj", "model.language_model.layers.38.mlp.experts.5.up_proj", "model.language_model.layers.38.mlp.experts.6.up_proj", "model.language_model.layers.38.mlp.experts.7.up_proj", "model.language_model.layers.38.mlp.experts.8.up_proj", "model.language_model.layers.38.mlp.experts.9.up_proj", "model.language_model.layers.38.mlp.experts.10.up_proj", "model.language_model.layers.38.mlp.experts.11.up_proj", "model.language_model.layers.38.mlp.experts.12.up_proj", "model.language_model.layers.38.mlp.experts.13.up_proj", "model.language_model.layers.38.mlp.experts.14.up_proj", "model.language_model.layers.38.mlp.experts.15.up_proj", "model.language_model.layers.38.mlp.experts.16.up_proj", "model.language_model.layers.38.mlp.experts.17.up_proj", "model.language_model.layers.38.mlp.experts.18.up_proj", "model.language_model.layers.38.mlp.experts.19.up_proj", "model.language_model.layers.38.mlp.experts.20.up_proj", "model.language_model.layers.38.mlp.experts.21.up_proj", "model.language_model.layers.38.mlp.experts.22.up_proj", "model.language_model.layers.38.mlp.experts.23.up_proj", "model.language_model.layers.38.mlp.experts.24.up_proj", "model.language_model.layers.38.mlp.experts.25.up_proj", "model.language_model.layers.38.mlp.experts.26.up_proj", "model.language_model.layers.38.mlp.experts.27.up_proj", "model.language_model.layers.38.mlp.experts.28.up_proj", "model.language_model.layers.38.mlp.experts.29.up_proj", "model.language_model.layers.38.mlp.experts.30.up_proj", "model.language_model.layers.38.mlp.experts.31.up_proj", "model.language_model.layers.38.mlp.experts.32.up_proj", "model.language_model.layers.38.mlp.experts.33.up_proj", "model.language_model.layers.38.mlp.experts.34.up_proj", "model.language_model.layers.38.mlp.experts.35.up_proj", "model.language_model.layers.38.mlp.experts.36.up_proj", "model.language_model.layers.38.mlp.experts.37.up_proj", "model.language_model.layers.38.mlp.experts.38.up_proj", "model.language_model.layers.38.mlp.experts.39.up_proj", "model.language_model.layers.38.mlp.experts.40.up_proj", "model.language_model.layers.38.mlp.experts.41.up_proj", "model.language_model.layers.38.mlp.experts.42.up_proj", "model.language_model.layers.38.mlp.experts.43.up_proj", "model.language_model.layers.38.mlp.experts.44.up_proj", "model.language_model.layers.38.mlp.experts.45.up_proj", "model.language_model.layers.38.mlp.experts.46.up_proj", "model.language_model.layers.38.mlp.experts.47.up_proj", "model.language_model.layers.38.mlp.experts.48.up_proj", "model.language_model.layers.38.mlp.experts.49.up_proj", "model.language_model.layers.38.mlp.experts.50.up_proj", "model.language_model.layers.38.mlp.experts.51.up_proj", "model.language_model.layers.38.mlp.experts.52.up_proj", "model.language_model.layers.38.mlp.experts.53.up_proj", "model.language_model.layers.38.mlp.experts.54.up_proj", "model.language_model.layers.38.mlp.experts.55.up_proj", "model.language_model.layers.38.mlp.experts.56.up_proj", "model.language_model.layers.38.mlp.experts.57.up_proj", "model.language_model.layers.38.mlp.experts.58.up_proj", "model.language_model.layers.38.mlp.experts.59.up_proj", "model.language_model.layers.38.mlp.experts.60.up_proj", "model.language_model.layers.38.mlp.experts.61.up_proj", "model.language_model.layers.38.mlp.experts.62.up_proj", "model.language_model.layers.38.mlp.experts.63.up_proj", "model.language_model.layers.38.mlp.experts.64.up_proj", "model.language_model.layers.38.mlp.experts.65.up_proj", "model.language_model.layers.38.mlp.experts.66.up_proj", "model.language_model.layers.38.mlp.experts.67.up_proj", "model.language_model.layers.38.mlp.experts.68.up_proj", "model.language_model.layers.38.mlp.experts.69.up_proj", "model.language_model.layers.38.mlp.experts.70.up_proj", "model.language_model.layers.38.mlp.experts.71.up_proj", "model.language_model.layers.38.mlp.experts.72.up_proj", "model.language_model.layers.38.mlp.experts.73.up_proj", "model.language_model.layers.38.mlp.experts.74.up_proj", "model.language_model.layers.38.mlp.experts.75.up_proj", "model.language_model.layers.38.mlp.experts.76.up_proj", "model.language_model.layers.38.mlp.experts.77.up_proj", "model.language_model.layers.38.mlp.experts.78.up_proj", "model.language_model.layers.38.mlp.experts.79.up_proj", "model.language_model.layers.38.mlp.experts.80.up_proj", "model.language_model.layers.38.mlp.experts.81.up_proj", "model.language_model.layers.38.mlp.experts.82.up_proj", "model.language_model.layers.38.mlp.experts.83.up_proj", "model.language_model.layers.38.mlp.experts.84.up_proj", "model.language_model.layers.38.mlp.experts.85.up_proj", "model.language_model.layers.38.mlp.experts.86.up_proj", "model.language_model.layers.38.mlp.experts.87.up_proj", "model.language_model.layers.38.mlp.experts.88.up_proj", "model.language_model.layers.38.mlp.experts.89.up_proj", "model.language_model.layers.38.mlp.experts.90.up_proj", "model.language_model.layers.38.mlp.experts.91.up_proj", "model.language_model.layers.38.mlp.experts.92.up_proj", "model.language_model.layers.38.mlp.experts.93.up_proj", "model.language_model.layers.38.mlp.experts.94.up_proj", "model.language_model.layers.38.mlp.experts.95.up_proj", "model.language_model.layers.38.mlp.experts.96.up_proj", "model.language_model.layers.38.mlp.experts.97.up_proj", "model.language_model.layers.38.mlp.experts.98.up_proj", "model.language_model.layers.38.mlp.experts.99.up_proj", "model.language_model.layers.38.mlp.experts.100.up_proj", "model.language_model.layers.38.mlp.experts.101.up_proj", "model.language_model.layers.38.mlp.experts.102.up_proj", "model.language_model.layers.38.mlp.experts.103.up_proj", "model.language_model.layers.38.mlp.experts.104.up_proj", "model.language_model.layers.38.mlp.experts.105.up_proj", "model.language_model.layers.38.mlp.experts.106.up_proj", "model.language_model.layers.38.mlp.experts.107.up_proj", "model.language_model.layers.38.mlp.experts.108.up_proj", "model.language_model.layers.38.mlp.experts.109.up_proj", "model.language_model.layers.38.mlp.experts.110.up_proj", "model.language_model.layers.38.mlp.experts.111.up_proj", "model.language_model.layers.38.mlp.experts.112.up_proj", "model.language_model.layers.38.mlp.experts.113.up_proj", "model.language_model.layers.38.mlp.experts.114.up_proj", "model.language_model.layers.38.mlp.experts.115.up_proj", "model.language_model.layers.38.mlp.experts.116.up_proj", "model.language_model.layers.38.mlp.experts.117.up_proj", "model.language_model.layers.38.mlp.experts.118.up_proj", "model.language_model.layers.38.mlp.experts.119.up_proj", "model.language_model.layers.38.mlp.experts.120.up_proj", "model.language_model.layers.38.mlp.experts.121.up_proj", "model.language_model.layers.38.mlp.experts.122.up_proj", "model.language_model.layers.38.mlp.experts.123.up_proj", "model.language_model.layers.38.mlp.experts.124.up_proj", "model.language_model.layers.38.mlp.experts.125.up_proj", "model.language_model.layers.38.mlp.experts.126.up_proj", "model.language_model.layers.38.mlp.experts.127.up_proj", "model.language_model.layers.38.mlp.experts.0.down_proj", "model.language_model.layers.38.mlp.experts.1.down_proj", "model.language_model.layers.38.mlp.experts.2.down_proj", "model.language_model.layers.38.mlp.experts.3.down_proj", "model.language_model.layers.38.mlp.experts.4.down_proj", "model.language_model.layers.38.mlp.experts.5.down_proj", "model.language_model.layers.38.mlp.experts.6.down_proj", "model.language_model.layers.38.mlp.experts.7.down_proj", "model.language_model.layers.38.mlp.experts.8.down_proj", "model.language_model.layers.38.mlp.experts.9.down_proj", "model.language_model.layers.38.mlp.experts.10.down_proj", "model.language_model.layers.38.mlp.experts.11.down_proj", "model.language_model.layers.38.mlp.experts.12.down_proj", "model.language_model.layers.38.mlp.experts.13.down_proj", "model.language_model.layers.38.mlp.experts.14.down_proj", "model.language_model.layers.38.mlp.experts.15.down_proj", "model.language_model.layers.38.mlp.experts.16.down_proj", "model.language_model.layers.38.mlp.experts.17.down_proj", "model.language_model.layers.38.mlp.experts.18.down_proj", "model.language_model.layers.38.mlp.experts.19.down_proj", "model.language_model.layers.38.mlp.experts.20.down_proj", "model.language_model.layers.38.mlp.experts.21.down_proj", "model.language_model.layers.38.mlp.experts.22.down_proj", "model.language_model.layers.38.mlp.experts.23.down_proj", "model.language_model.layers.38.mlp.experts.24.down_proj", "model.language_model.layers.38.mlp.experts.25.down_proj", "model.language_model.layers.38.mlp.experts.26.down_proj", "model.language_model.layers.38.mlp.experts.27.down_proj", "model.language_model.layers.38.mlp.experts.28.down_proj", "model.language_model.layers.38.mlp.experts.29.down_proj", "model.language_model.layers.38.mlp.experts.30.down_proj", "model.language_model.layers.38.mlp.experts.31.down_proj", "model.language_model.layers.38.mlp.experts.32.down_proj", "model.language_model.layers.38.mlp.experts.33.down_proj", "model.language_model.layers.38.mlp.experts.34.down_proj", "model.language_model.layers.38.mlp.experts.35.down_proj", "model.language_model.layers.38.mlp.experts.36.down_proj", "model.language_model.layers.38.mlp.experts.37.down_proj", "model.language_model.layers.38.mlp.experts.38.down_proj", "model.language_model.layers.38.mlp.experts.39.down_proj", "model.language_model.layers.38.mlp.experts.40.down_proj", "model.language_model.layers.38.mlp.experts.41.down_proj", "model.language_model.layers.38.mlp.experts.42.down_proj", "model.language_model.layers.38.mlp.experts.43.down_proj", "model.language_model.layers.38.mlp.experts.44.down_proj", "model.language_model.layers.38.mlp.experts.45.down_proj", "model.language_model.layers.38.mlp.experts.46.down_proj", "model.language_model.layers.38.mlp.experts.47.down_proj", "model.language_model.layers.38.mlp.experts.48.down_proj", "model.language_model.layers.38.mlp.experts.49.down_proj", "model.language_model.layers.38.mlp.experts.50.down_proj", "model.language_model.layers.38.mlp.experts.51.down_proj", "model.language_model.layers.38.mlp.experts.52.down_proj", "model.language_model.layers.38.mlp.experts.53.down_proj", "model.language_model.layers.38.mlp.experts.54.down_proj", "model.language_model.layers.38.mlp.experts.55.down_proj", "model.language_model.layers.38.mlp.experts.56.down_proj", "model.language_model.layers.38.mlp.experts.57.down_proj", "model.language_model.layers.38.mlp.experts.58.down_proj", "model.language_model.layers.38.mlp.experts.59.down_proj", "model.language_model.layers.38.mlp.experts.60.down_proj", "model.language_model.layers.38.mlp.experts.61.down_proj", "model.language_model.layers.38.mlp.experts.62.down_proj", "model.language_model.layers.38.mlp.experts.63.down_proj", "model.language_model.layers.38.mlp.experts.64.down_proj", "model.language_model.layers.38.mlp.experts.65.down_proj", "model.language_model.layers.38.mlp.experts.66.down_proj", "model.language_model.layers.38.mlp.experts.67.down_proj", "model.language_model.layers.38.mlp.experts.68.down_proj", "model.language_model.layers.38.mlp.experts.69.down_proj", "model.language_model.layers.38.mlp.experts.70.down_proj", "model.language_model.layers.38.mlp.experts.71.down_proj", "model.language_model.layers.38.mlp.experts.72.down_proj", "model.language_model.layers.38.mlp.experts.73.down_proj", "model.language_model.layers.38.mlp.experts.74.down_proj", "model.language_model.layers.38.mlp.experts.75.down_proj", "model.language_model.layers.38.mlp.experts.76.down_proj", "model.language_model.layers.38.mlp.experts.77.down_proj", "model.language_model.layers.38.mlp.experts.78.down_proj", "model.language_model.layers.38.mlp.experts.79.down_proj", "model.language_model.layers.38.mlp.experts.80.down_proj", "model.language_model.layers.38.mlp.experts.81.down_proj", "model.language_model.layers.38.mlp.experts.82.down_proj", "model.language_model.layers.38.mlp.experts.83.down_proj", "model.language_model.layers.38.mlp.experts.84.down_proj", "model.language_model.layers.38.mlp.experts.85.down_proj", "model.language_model.layers.38.mlp.experts.86.down_proj", "model.language_model.layers.38.mlp.experts.87.down_proj", "model.language_model.layers.38.mlp.experts.88.down_proj", "model.language_model.layers.38.mlp.experts.89.down_proj", "model.language_model.layers.38.mlp.experts.90.down_proj", "model.language_model.layers.38.mlp.experts.91.down_proj", "model.language_model.layers.38.mlp.experts.92.down_proj", "model.language_model.layers.38.mlp.experts.93.down_proj", "model.language_model.layers.38.mlp.experts.94.down_proj", "model.language_model.layers.38.mlp.experts.95.down_proj", "model.language_model.layers.38.mlp.experts.96.down_proj", "model.language_model.layers.38.mlp.experts.97.down_proj", "model.language_model.layers.38.mlp.experts.98.down_proj", "model.language_model.layers.38.mlp.experts.99.down_proj", "model.language_model.layers.38.mlp.experts.100.down_proj", "model.language_model.layers.38.mlp.experts.101.down_proj", "model.language_model.layers.38.mlp.experts.102.down_proj", "model.language_model.layers.38.mlp.experts.103.down_proj", "model.language_model.layers.38.mlp.experts.104.down_proj", "model.language_model.layers.38.mlp.experts.105.down_proj", "model.language_model.layers.38.mlp.experts.106.down_proj", "model.language_model.layers.38.mlp.experts.107.down_proj", "model.language_model.layers.38.mlp.experts.108.down_proj", "model.language_model.layers.38.mlp.experts.109.down_proj", "model.language_model.layers.38.mlp.experts.110.down_proj", "model.language_model.layers.38.mlp.experts.111.down_proj", "model.language_model.layers.38.mlp.experts.112.down_proj", "model.language_model.layers.38.mlp.experts.113.down_proj", "model.language_model.layers.38.mlp.experts.114.down_proj", "model.language_model.layers.38.mlp.experts.115.down_proj", "model.language_model.layers.38.mlp.experts.116.down_proj", "model.language_model.layers.38.mlp.experts.117.down_proj", "model.language_model.layers.38.mlp.experts.118.down_proj", "model.language_model.layers.38.mlp.experts.119.down_proj", "model.language_model.layers.38.mlp.experts.120.down_proj", "model.language_model.layers.38.mlp.experts.121.down_proj", "model.language_model.layers.38.mlp.experts.122.down_proj", "model.language_model.layers.38.mlp.experts.123.down_proj", "model.language_model.layers.38.mlp.experts.124.down_proj", "model.language_model.layers.38.mlp.experts.125.down_proj", "model.language_model.layers.38.mlp.experts.126.down_proj", "model.language_model.layers.38.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.0007278031669557122, "dbits": 603979776 } ] }, { "idx": 78, "layers": [ "model.language_model.layers.39.self_attn.q_proj", "model.language_model.layers.39.self_attn.k_proj", "model.language_model.layers.39.self_attn.v_proj", "model.language_model.layers.39.self_attn.o_proj" ], "candidates": [ { "dkld": 3.409888595343191e-05, "dbits": 18874368 } ] }, { "idx": 79, "layers": [ "model.language_model.layers.39.mlp.experts.0.gate_proj", "model.language_model.layers.39.mlp.experts.1.gate_proj", "model.language_model.layers.39.mlp.experts.2.gate_proj", "model.language_model.layers.39.mlp.experts.3.gate_proj", "model.language_model.layers.39.mlp.experts.4.gate_proj", "model.language_model.layers.39.mlp.experts.5.gate_proj", "model.language_model.layers.39.mlp.experts.6.gate_proj", "model.language_model.layers.39.mlp.experts.7.gate_proj", "model.language_model.layers.39.mlp.experts.8.gate_proj", "model.language_model.layers.39.mlp.experts.9.gate_proj", "model.language_model.layers.39.mlp.experts.10.gate_proj", "model.language_model.layers.39.mlp.experts.11.gate_proj", "model.language_model.layers.39.mlp.experts.12.gate_proj", "model.language_model.layers.39.mlp.experts.13.gate_proj", "model.language_model.layers.39.mlp.experts.14.gate_proj", "model.language_model.layers.39.mlp.experts.15.gate_proj", "model.language_model.layers.39.mlp.experts.16.gate_proj", "model.language_model.layers.39.mlp.experts.17.gate_proj", "model.language_model.layers.39.mlp.experts.18.gate_proj", "model.language_model.layers.39.mlp.experts.19.gate_proj", "model.language_model.layers.39.mlp.experts.20.gate_proj", "model.language_model.layers.39.mlp.experts.21.gate_proj", "model.language_model.layers.39.mlp.experts.22.gate_proj", "model.language_model.layers.39.mlp.experts.23.gate_proj", "model.language_model.layers.39.mlp.experts.24.gate_proj", "model.language_model.layers.39.mlp.experts.25.gate_proj", "model.language_model.layers.39.mlp.experts.26.gate_proj", "model.language_model.layers.39.mlp.experts.27.gate_proj", "model.language_model.layers.39.mlp.experts.28.gate_proj", "model.language_model.layers.39.mlp.experts.29.gate_proj", "model.language_model.layers.39.mlp.experts.30.gate_proj", "model.language_model.layers.39.mlp.experts.31.gate_proj", "model.language_model.layers.39.mlp.experts.32.gate_proj", "model.language_model.layers.39.mlp.experts.33.gate_proj", "model.language_model.layers.39.mlp.experts.34.gate_proj", "model.language_model.layers.39.mlp.experts.35.gate_proj", "model.language_model.layers.39.mlp.experts.36.gate_proj", "model.language_model.layers.39.mlp.experts.37.gate_proj", "model.language_model.layers.39.mlp.experts.38.gate_proj", "model.language_model.layers.39.mlp.experts.39.gate_proj", "model.language_model.layers.39.mlp.experts.40.gate_proj", "model.language_model.layers.39.mlp.experts.41.gate_proj", "model.language_model.layers.39.mlp.experts.42.gate_proj", "model.language_model.layers.39.mlp.experts.43.gate_proj", "model.language_model.layers.39.mlp.experts.44.gate_proj", "model.language_model.layers.39.mlp.experts.45.gate_proj", "model.language_model.layers.39.mlp.experts.46.gate_proj", "model.language_model.layers.39.mlp.experts.47.gate_proj", "model.language_model.layers.39.mlp.experts.48.gate_proj", "model.language_model.layers.39.mlp.experts.49.gate_proj", "model.language_model.layers.39.mlp.experts.50.gate_proj", "model.language_model.layers.39.mlp.experts.51.gate_proj", "model.language_model.layers.39.mlp.experts.52.gate_proj", "model.language_model.layers.39.mlp.experts.53.gate_proj", "model.language_model.layers.39.mlp.experts.54.gate_proj", "model.language_model.layers.39.mlp.experts.55.gate_proj", "model.language_model.layers.39.mlp.experts.56.gate_proj", "model.language_model.layers.39.mlp.experts.57.gate_proj", "model.language_model.layers.39.mlp.experts.58.gate_proj", "model.language_model.layers.39.mlp.experts.59.gate_proj", "model.language_model.layers.39.mlp.experts.60.gate_proj", "model.language_model.layers.39.mlp.experts.61.gate_proj", "model.language_model.layers.39.mlp.experts.62.gate_proj", "model.language_model.layers.39.mlp.experts.63.gate_proj", "model.language_model.layers.39.mlp.experts.64.gate_proj", "model.language_model.layers.39.mlp.experts.65.gate_proj", "model.language_model.layers.39.mlp.experts.66.gate_proj", "model.language_model.layers.39.mlp.experts.67.gate_proj", "model.language_model.layers.39.mlp.experts.68.gate_proj", "model.language_model.layers.39.mlp.experts.69.gate_proj", "model.language_model.layers.39.mlp.experts.70.gate_proj", "model.language_model.layers.39.mlp.experts.71.gate_proj", "model.language_model.layers.39.mlp.experts.72.gate_proj", "model.language_model.layers.39.mlp.experts.73.gate_proj", "model.language_model.layers.39.mlp.experts.74.gate_proj", "model.language_model.layers.39.mlp.experts.75.gate_proj", "model.language_model.layers.39.mlp.experts.76.gate_proj", "model.language_model.layers.39.mlp.experts.77.gate_proj", "model.language_model.layers.39.mlp.experts.78.gate_proj", "model.language_model.layers.39.mlp.experts.79.gate_proj", "model.language_model.layers.39.mlp.experts.80.gate_proj", "model.language_model.layers.39.mlp.experts.81.gate_proj", "model.language_model.layers.39.mlp.experts.82.gate_proj", "model.language_model.layers.39.mlp.experts.83.gate_proj", "model.language_model.layers.39.mlp.experts.84.gate_proj", "model.language_model.layers.39.mlp.experts.85.gate_proj", "model.language_model.layers.39.mlp.experts.86.gate_proj", "model.language_model.layers.39.mlp.experts.87.gate_proj", "model.language_model.layers.39.mlp.experts.88.gate_proj", "model.language_model.layers.39.mlp.experts.89.gate_proj", "model.language_model.layers.39.mlp.experts.90.gate_proj", "model.language_model.layers.39.mlp.experts.91.gate_proj", "model.language_model.layers.39.mlp.experts.92.gate_proj", "model.language_model.layers.39.mlp.experts.93.gate_proj", "model.language_model.layers.39.mlp.experts.94.gate_proj", "model.language_model.layers.39.mlp.experts.95.gate_proj", "model.language_model.layers.39.mlp.experts.96.gate_proj", "model.language_model.layers.39.mlp.experts.97.gate_proj", "model.language_model.layers.39.mlp.experts.98.gate_proj", "model.language_model.layers.39.mlp.experts.99.gate_proj", "model.language_model.layers.39.mlp.experts.100.gate_proj", "model.language_model.layers.39.mlp.experts.101.gate_proj", "model.language_model.layers.39.mlp.experts.102.gate_proj", "model.language_model.layers.39.mlp.experts.103.gate_proj", "model.language_model.layers.39.mlp.experts.104.gate_proj", "model.language_model.layers.39.mlp.experts.105.gate_proj", "model.language_model.layers.39.mlp.experts.106.gate_proj", "model.language_model.layers.39.mlp.experts.107.gate_proj", "model.language_model.layers.39.mlp.experts.108.gate_proj", "model.language_model.layers.39.mlp.experts.109.gate_proj", "model.language_model.layers.39.mlp.experts.110.gate_proj", "model.language_model.layers.39.mlp.experts.111.gate_proj", "model.language_model.layers.39.mlp.experts.112.gate_proj", "model.language_model.layers.39.mlp.experts.113.gate_proj", "model.language_model.layers.39.mlp.experts.114.gate_proj", "model.language_model.layers.39.mlp.experts.115.gate_proj", "model.language_model.layers.39.mlp.experts.116.gate_proj", "model.language_model.layers.39.mlp.experts.117.gate_proj", "model.language_model.layers.39.mlp.experts.118.gate_proj", "model.language_model.layers.39.mlp.experts.119.gate_proj", "model.language_model.layers.39.mlp.experts.120.gate_proj", "model.language_model.layers.39.mlp.experts.121.gate_proj", "model.language_model.layers.39.mlp.experts.122.gate_proj", "model.language_model.layers.39.mlp.experts.123.gate_proj", "model.language_model.layers.39.mlp.experts.124.gate_proj", "model.language_model.layers.39.mlp.experts.125.gate_proj", "model.language_model.layers.39.mlp.experts.126.gate_proj", "model.language_model.layers.39.mlp.experts.127.gate_proj", "model.language_model.layers.39.mlp.experts.0.up_proj", "model.language_model.layers.39.mlp.experts.1.up_proj", "model.language_model.layers.39.mlp.experts.2.up_proj", "model.language_model.layers.39.mlp.experts.3.up_proj", "model.language_model.layers.39.mlp.experts.4.up_proj", "model.language_model.layers.39.mlp.experts.5.up_proj", "model.language_model.layers.39.mlp.experts.6.up_proj", "model.language_model.layers.39.mlp.experts.7.up_proj", "model.language_model.layers.39.mlp.experts.8.up_proj", "model.language_model.layers.39.mlp.experts.9.up_proj", "model.language_model.layers.39.mlp.experts.10.up_proj", "model.language_model.layers.39.mlp.experts.11.up_proj", "model.language_model.layers.39.mlp.experts.12.up_proj", "model.language_model.layers.39.mlp.experts.13.up_proj", "model.language_model.layers.39.mlp.experts.14.up_proj", "model.language_model.layers.39.mlp.experts.15.up_proj", "model.language_model.layers.39.mlp.experts.16.up_proj", "model.language_model.layers.39.mlp.experts.17.up_proj", "model.language_model.layers.39.mlp.experts.18.up_proj", "model.language_model.layers.39.mlp.experts.19.up_proj", "model.language_model.layers.39.mlp.experts.20.up_proj", "model.language_model.layers.39.mlp.experts.21.up_proj", "model.language_model.layers.39.mlp.experts.22.up_proj", "model.language_model.layers.39.mlp.experts.23.up_proj", "model.language_model.layers.39.mlp.experts.24.up_proj", "model.language_model.layers.39.mlp.experts.25.up_proj", "model.language_model.layers.39.mlp.experts.26.up_proj", "model.language_model.layers.39.mlp.experts.27.up_proj", "model.language_model.layers.39.mlp.experts.28.up_proj", "model.language_model.layers.39.mlp.experts.29.up_proj", "model.language_model.layers.39.mlp.experts.30.up_proj", "model.language_model.layers.39.mlp.experts.31.up_proj", "model.language_model.layers.39.mlp.experts.32.up_proj", "model.language_model.layers.39.mlp.experts.33.up_proj", "model.language_model.layers.39.mlp.experts.34.up_proj", "model.language_model.layers.39.mlp.experts.35.up_proj", "model.language_model.layers.39.mlp.experts.36.up_proj", "model.language_model.layers.39.mlp.experts.37.up_proj", "model.language_model.layers.39.mlp.experts.38.up_proj", "model.language_model.layers.39.mlp.experts.39.up_proj", "model.language_model.layers.39.mlp.experts.40.up_proj", "model.language_model.layers.39.mlp.experts.41.up_proj", "model.language_model.layers.39.mlp.experts.42.up_proj", "model.language_model.layers.39.mlp.experts.43.up_proj", "model.language_model.layers.39.mlp.experts.44.up_proj", "model.language_model.layers.39.mlp.experts.45.up_proj", "model.language_model.layers.39.mlp.experts.46.up_proj", "model.language_model.layers.39.mlp.experts.47.up_proj", "model.language_model.layers.39.mlp.experts.48.up_proj", "model.language_model.layers.39.mlp.experts.49.up_proj", "model.language_model.layers.39.mlp.experts.50.up_proj", "model.language_model.layers.39.mlp.experts.51.up_proj", "model.language_model.layers.39.mlp.experts.52.up_proj", "model.language_model.layers.39.mlp.experts.53.up_proj", "model.language_model.layers.39.mlp.experts.54.up_proj", "model.language_model.layers.39.mlp.experts.55.up_proj", "model.language_model.layers.39.mlp.experts.56.up_proj", "model.language_model.layers.39.mlp.experts.57.up_proj", "model.language_model.layers.39.mlp.experts.58.up_proj", "model.language_model.layers.39.mlp.experts.59.up_proj", "model.language_model.layers.39.mlp.experts.60.up_proj", "model.language_model.layers.39.mlp.experts.61.up_proj", "model.language_model.layers.39.mlp.experts.62.up_proj", "model.language_model.layers.39.mlp.experts.63.up_proj", "model.language_model.layers.39.mlp.experts.64.up_proj", "model.language_model.layers.39.mlp.experts.65.up_proj", "model.language_model.layers.39.mlp.experts.66.up_proj", "model.language_model.layers.39.mlp.experts.67.up_proj", "model.language_model.layers.39.mlp.experts.68.up_proj", "model.language_model.layers.39.mlp.experts.69.up_proj", "model.language_model.layers.39.mlp.experts.70.up_proj", "model.language_model.layers.39.mlp.experts.71.up_proj", "model.language_model.layers.39.mlp.experts.72.up_proj", "model.language_model.layers.39.mlp.experts.73.up_proj", "model.language_model.layers.39.mlp.experts.74.up_proj", "model.language_model.layers.39.mlp.experts.75.up_proj", "model.language_model.layers.39.mlp.experts.76.up_proj", "model.language_model.layers.39.mlp.experts.77.up_proj", "model.language_model.layers.39.mlp.experts.78.up_proj", "model.language_model.layers.39.mlp.experts.79.up_proj", "model.language_model.layers.39.mlp.experts.80.up_proj", "model.language_model.layers.39.mlp.experts.81.up_proj", "model.language_model.layers.39.mlp.experts.82.up_proj", "model.language_model.layers.39.mlp.experts.83.up_proj", "model.language_model.layers.39.mlp.experts.84.up_proj", "model.language_model.layers.39.mlp.experts.85.up_proj", "model.language_model.layers.39.mlp.experts.86.up_proj", "model.language_model.layers.39.mlp.experts.87.up_proj", "model.language_model.layers.39.mlp.experts.88.up_proj", "model.language_model.layers.39.mlp.experts.89.up_proj", "model.language_model.layers.39.mlp.experts.90.up_proj", "model.language_model.layers.39.mlp.experts.91.up_proj", "model.language_model.layers.39.mlp.experts.92.up_proj", "model.language_model.layers.39.mlp.experts.93.up_proj", "model.language_model.layers.39.mlp.experts.94.up_proj", "model.language_model.layers.39.mlp.experts.95.up_proj", "model.language_model.layers.39.mlp.experts.96.up_proj", "model.language_model.layers.39.mlp.experts.97.up_proj", "model.language_model.layers.39.mlp.experts.98.up_proj", "model.language_model.layers.39.mlp.experts.99.up_proj", "model.language_model.layers.39.mlp.experts.100.up_proj", "model.language_model.layers.39.mlp.experts.101.up_proj", "model.language_model.layers.39.mlp.experts.102.up_proj", "model.language_model.layers.39.mlp.experts.103.up_proj", "model.language_model.layers.39.mlp.experts.104.up_proj", "model.language_model.layers.39.mlp.experts.105.up_proj", "model.language_model.layers.39.mlp.experts.106.up_proj", "model.language_model.layers.39.mlp.experts.107.up_proj", "model.language_model.layers.39.mlp.experts.108.up_proj", "model.language_model.layers.39.mlp.experts.109.up_proj", "model.language_model.layers.39.mlp.experts.110.up_proj", "model.language_model.layers.39.mlp.experts.111.up_proj", "model.language_model.layers.39.mlp.experts.112.up_proj", "model.language_model.layers.39.mlp.experts.113.up_proj", "model.language_model.layers.39.mlp.experts.114.up_proj", "model.language_model.layers.39.mlp.experts.115.up_proj", "model.language_model.layers.39.mlp.experts.116.up_proj", "model.language_model.layers.39.mlp.experts.117.up_proj", "model.language_model.layers.39.mlp.experts.118.up_proj", "model.language_model.layers.39.mlp.experts.119.up_proj", "model.language_model.layers.39.mlp.experts.120.up_proj", "model.language_model.layers.39.mlp.experts.121.up_proj", "model.language_model.layers.39.mlp.experts.122.up_proj", "model.language_model.layers.39.mlp.experts.123.up_proj", "model.language_model.layers.39.mlp.experts.124.up_proj", "model.language_model.layers.39.mlp.experts.125.up_proj", "model.language_model.layers.39.mlp.experts.126.up_proj", "model.language_model.layers.39.mlp.experts.127.up_proj", "model.language_model.layers.39.mlp.experts.0.down_proj", "model.language_model.layers.39.mlp.experts.1.down_proj", "model.language_model.layers.39.mlp.experts.2.down_proj", "model.language_model.layers.39.mlp.experts.3.down_proj", "model.language_model.layers.39.mlp.experts.4.down_proj", "model.language_model.layers.39.mlp.experts.5.down_proj", "model.language_model.layers.39.mlp.experts.6.down_proj", "model.language_model.layers.39.mlp.experts.7.down_proj", "model.language_model.layers.39.mlp.experts.8.down_proj", "model.language_model.layers.39.mlp.experts.9.down_proj", "model.language_model.layers.39.mlp.experts.10.down_proj", "model.language_model.layers.39.mlp.experts.11.down_proj", "model.language_model.layers.39.mlp.experts.12.down_proj", "model.language_model.layers.39.mlp.experts.13.down_proj", "model.language_model.layers.39.mlp.experts.14.down_proj", "model.language_model.layers.39.mlp.experts.15.down_proj", "model.language_model.layers.39.mlp.experts.16.down_proj", "model.language_model.layers.39.mlp.experts.17.down_proj", "model.language_model.layers.39.mlp.experts.18.down_proj", "model.language_model.layers.39.mlp.experts.19.down_proj", "model.language_model.layers.39.mlp.experts.20.down_proj", "model.language_model.layers.39.mlp.experts.21.down_proj", "model.language_model.layers.39.mlp.experts.22.down_proj", "model.language_model.layers.39.mlp.experts.23.down_proj", "model.language_model.layers.39.mlp.experts.24.down_proj", "model.language_model.layers.39.mlp.experts.25.down_proj", "model.language_model.layers.39.mlp.experts.26.down_proj", "model.language_model.layers.39.mlp.experts.27.down_proj", "model.language_model.layers.39.mlp.experts.28.down_proj", "model.language_model.layers.39.mlp.experts.29.down_proj", "model.language_model.layers.39.mlp.experts.30.down_proj", "model.language_model.layers.39.mlp.experts.31.down_proj", "model.language_model.layers.39.mlp.experts.32.down_proj", "model.language_model.layers.39.mlp.experts.33.down_proj", "model.language_model.layers.39.mlp.experts.34.down_proj", "model.language_model.layers.39.mlp.experts.35.down_proj", "model.language_model.layers.39.mlp.experts.36.down_proj", "model.language_model.layers.39.mlp.experts.37.down_proj", "model.language_model.layers.39.mlp.experts.38.down_proj", "model.language_model.layers.39.mlp.experts.39.down_proj", "model.language_model.layers.39.mlp.experts.40.down_proj", "model.language_model.layers.39.mlp.experts.41.down_proj", "model.language_model.layers.39.mlp.experts.42.down_proj", "model.language_model.layers.39.mlp.experts.43.down_proj", "model.language_model.layers.39.mlp.experts.44.down_proj", "model.language_model.layers.39.mlp.experts.45.down_proj", "model.language_model.layers.39.mlp.experts.46.down_proj", "model.language_model.layers.39.mlp.experts.47.down_proj", "model.language_model.layers.39.mlp.experts.48.down_proj", "model.language_model.layers.39.mlp.experts.49.down_proj", "model.language_model.layers.39.mlp.experts.50.down_proj", "model.language_model.layers.39.mlp.experts.51.down_proj", "model.language_model.layers.39.mlp.experts.52.down_proj", "model.language_model.layers.39.mlp.experts.53.down_proj", "model.language_model.layers.39.mlp.experts.54.down_proj", "model.language_model.layers.39.mlp.experts.55.down_proj", "model.language_model.layers.39.mlp.experts.56.down_proj", "model.language_model.layers.39.mlp.experts.57.down_proj", "model.language_model.layers.39.mlp.experts.58.down_proj", "model.language_model.layers.39.mlp.experts.59.down_proj", "model.language_model.layers.39.mlp.experts.60.down_proj", "model.language_model.layers.39.mlp.experts.61.down_proj", "model.language_model.layers.39.mlp.experts.62.down_proj", "model.language_model.layers.39.mlp.experts.63.down_proj", "model.language_model.layers.39.mlp.experts.64.down_proj", "model.language_model.layers.39.mlp.experts.65.down_proj", "model.language_model.layers.39.mlp.experts.66.down_proj", "model.language_model.layers.39.mlp.experts.67.down_proj", "model.language_model.layers.39.mlp.experts.68.down_proj", "model.language_model.layers.39.mlp.experts.69.down_proj", "model.language_model.layers.39.mlp.experts.70.down_proj", "model.language_model.layers.39.mlp.experts.71.down_proj", "model.language_model.layers.39.mlp.experts.72.down_proj", "model.language_model.layers.39.mlp.experts.73.down_proj", "model.language_model.layers.39.mlp.experts.74.down_proj", "model.language_model.layers.39.mlp.experts.75.down_proj", "model.language_model.layers.39.mlp.experts.76.down_proj", "model.language_model.layers.39.mlp.experts.77.down_proj", "model.language_model.layers.39.mlp.experts.78.down_proj", "model.language_model.layers.39.mlp.experts.79.down_proj", "model.language_model.layers.39.mlp.experts.80.down_proj", "model.language_model.layers.39.mlp.experts.81.down_proj", "model.language_model.layers.39.mlp.experts.82.down_proj", "model.language_model.layers.39.mlp.experts.83.down_proj", "model.language_model.layers.39.mlp.experts.84.down_proj", "model.language_model.layers.39.mlp.experts.85.down_proj", "model.language_model.layers.39.mlp.experts.86.down_proj", "model.language_model.layers.39.mlp.experts.87.down_proj", "model.language_model.layers.39.mlp.experts.88.down_proj", "model.language_model.layers.39.mlp.experts.89.down_proj", "model.language_model.layers.39.mlp.experts.90.down_proj", "model.language_model.layers.39.mlp.experts.91.down_proj", "model.language_model.layers.39.mlp.experts.92.down_proj", "model.language_model.layers.39.mlp.experts.93.down_proj", "model.language_model.layers.39.mlp.experts.94.down_proj", "model.language_model.layers.39.mlp.experts.95.down_proj", "model.language_model.layers.39.mlp.experts.96.down_proj", "model.language_model.layers.39.mlp.experts.97.down_proj", "model.language_model.layers.39.mlp.experts.98.down_proj", "model.language_model.layers.39.mlp.experts.99.down_proj", "model.language_model.layers.39.mlp.experts.100.down_proj", "model.language_model.layers.39.mlp.experts.101.down_proj", "model.language_model.layers.39.mlp.experts.102.down_proj", "model.language_model.layers.39.mlp.experts.103.down_proj", "model.language_model.layers.39.mlp.experts.104.down_proj", "model.language_model.layers.39.mlp.experts.105.down_proj", "model.language_model.layers.39.mlp.experts.106.down_proj", "model.language_model.layers.39.mlp.experts.107.down_proj", "model.language_model.layers.39.mlp.experts.108.down_proj", "model.language_model.layers.39.mlp.experts.109.down_proj", "model.language_model.layers.39.mlp.experts.110.down_proj", "model.language_model.layers.39.mlp.experts.111.down_proj", "model.language_model.layers.39.mlp.experts.112.down_proj", "model.language_model.layers.39.mlp.experts.113.down_proj", "model.language_model.layers.39.mlp.experts.114.down_proj", "model.language_model.layers.39.mlp.experts.115.down_proj", "model.language_model.layers.39.mlp.experts.116.down_proj", "model.language_model.layers.39.mlp.experts.117.down_proj", "model.language_model.layers.39.mlp.experts.118.down_proj", "model.language_model.layers.39.mlp.experts.119.down_proj", "model.language_model.layers.39.mlp.experts.120.down_proj", "model.language_model.layers.39.mlp.experts.121.down_proj", "model.language_model.layers.39.mlp.experts.122.down_proj", "model.language_model.layers.39.mlp.experts.123.down_proj", "model.language_model.layers.39.mlp.experts.124.down_proj", "model.language_model.layers.39.mlp.experts.125.down_proj", "model.language_model.layers.39.mlp.experts.126.down_proj", "model.language_model.layers.39.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.001248657237738371, "dbits": 603979776 } ] }, { "idx": 80, "layers": [ "model.language_model.layers.40.self_attn.q_proj", "model.language_model.layers.40.self_attn.k_proj", "model.language_model.layers.40.self_attn.v_proj", "model.language_model.layers.40.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0005954350344836712, "dbits": 18874368 } ] }, { "idx": 81, "layers": [ "model.language_model.layers.40.mlp.experts.0.gate_proj", "model.language_model.layers.40.mlp.experts.1.gate_proj", "model.language_model.layers.40.mlp.experts.2.gate_proj", "model.language_model.layers.40.mlp.experts.3.gate_proj", "model.language_model.layers.40.mlp.experts.4.gate_proj", "model.language_model.layers.40.mlp.experts.5.gate_proj", "model.language_model.layers.40.mlp.experts.6.gate_proj", "model.language_model.layers.40.mlp.experts.7.gate_proj", "model.language_model.layers.40.mlp.experts.8.gate_proj", "model.language_model.layers.40.mlp.experts.9.gate_proj", "model.language_model.layers.40.mlp.experts.10.gate_proj", "model.language_model.layers.40.mlp.experts.11.gate_proj", "model.language_model.layers.40.mlp.experts.12.gate_proj", "model.language_model.layers.40.mlp.experts.13.gate_proj", "model.language_model.layers.40.mlp.experts.14.gate_proj", "model.language_model.layers.40.mlp.experts.15.gate_proj", "model.language_model.layers.40.mlp.experts.16.gate_proj", "model.language_model.layers.40.mlp.experts.17.gate_proj", "model.language_model.layers.40.mlp.experts.18.gate_proj", "model.language_model.layers.40.mlp.experts.19.gate_proj", "model.language_model.layers.40.mlp.experts.20.gate_proj", "model.language_model.layers.40.mlp.experts.21.gate_proj", "model.language_model.layers.40.mlp.experts.22.gate_proj", "model.language_model.layers.40.mlp.experts.23.gate_proj", "model.language_model.layers.40.mlp.experts.24.gate_proj", "model.language_model.layers.40.mlp.experts.25.gate_proj", "model.language_model.layers.40.mlp.experts.26.gate_proj", "model.language_model.layers.40.mlp.experts.27.gate_proj", "model.language_model.layers.40.mlp.experts.28.gate_proj", "model.language_model.layers.40.mlp.experts.29.gate_proj", "model.language_model.layers.40.mlp.experts.30.gate_proj", "model.language_model.layers.40.mlp.experts.31.gate_proj", "model.language_model.layers.40.mlp.experts.32.gate_proj", "model.language_model.layers.40.mlp.experts.33.gate_proj", "model.language_model.layers.40.mlp.experts.34.gate_proj", "model.language_model.layers.40.mlp.experts.35.gate_proj", "model.language_model.layers.40.mlp.experts.36.gate_proj", "model.language_model.layers.40.mlp.experts.37.gate_proj", "model.language_model.layers.40.mlp.experts.38.gate_proj", "model.language_model.layers.40.mlp.experts.39.gate_proj", "model.language_model.layers.40.mlp.experts.40.gate_proj", "model.language_model.layers.40.mlp.experts.41.gate_proj", "model.language_model.layers.40.mlp.experts.42.gate_proj", "model.language_model.layers.40.mlp.experts.43.gate_proj", "model.language_model.layers.40.mlp.experts.44.gate_proj", "model.language_model.layers.40.mlp.experts.45.gate_proj", "model.language_model.layers.40.mlp.experts.46.gate_proj", "model.language_model.layers.40.mlp.experts.47.gate_proj", "model.language_model.layers.40.mlp.experts.48.gate_proj", "model.language_model.layers.40.mlp.experts.49.gate_proj", "model.language_model.layers.40.mlp.experts.50.gate_proj", "model.language_model.layers.40.mlp.experts.51.gate_proj", "model.language_model.layers.40.mlp.experts.52.gate_proj", "model.language_model.layers.40.mlp.experts.53.gate_proj", "model.language_model.layers.40.mlp.experts.54.gate_proj", "model.language_model.layers.40.mlp.experts.55.gate_proj", "model.language_model.layers.40.mlp.experts.56.gate_proj", "model.language_model.layers.40.mlp.experts.57.gate_proj", "model.language_model.layers.40.mlp.experts.58.gate_proj", "model.language_model.layers.40.mlp.experts.59.gate_proj", "model.language_model.layers.40.mlp.experts.60.gate_proj", "model.language_model.layers.40.mlp.experts.61.gate_proj", "model.language_model.layers.40.mlp.experts.62.gate_proj", "model.language_model.layers.40.mlp.experts.63.gate_proj", "model.language_model.layers.40.mlp.experts.64.gate_proj", "model.language_model.layers.40.mlp.experts.65.gate_proj", "model.language_model.layers.40.mlp.experts.66.gate_proj", "model.language_model.layers.40.mlp.experts.67.gate_proj", "model.language_model.layers.40.mlp.experts.68.gate_proj", "model.language_model.layers.40.mlp.experts.69.gate_proj", "model.language_model.layers.40.mlp.experts.70.gate_proj", "model.language_model.layers.40.mlp.experts.71.gate_proj", "model.language_model.layers.40.mlp.experts.72.gate_proj", "model.language_model.layers.40.mlp.experts.73.gate_proj", "model.language_model.layers.40.mlp.experts.74.gate_proj", "model.language_model.layers.40.mlp.experts.75.gate_proj", "model.language_model.layers.40.mlp.experts.76.gate_proj", "model.language_model.layers.40.mlp.experts.77.gate_proj", "model.language_model.layers.40.mlp.experts.78.gate_proj", "model.language_model.layers.40.mlp.experts.79.gate_proj", "model.language_model.layers.40.mlp.experts.80.gate_proj", "model.language_model.layers.40.mlp.experts.81.gate_proj", "model.language_model.layers.40.mlp.experts.82.gate_proj", "model.language_model.layers.40.mlp.experts.83.gate_proj", "model.language_model.layers.40.mlp.experts.84.gate_proj", "model.language_model.layers.40.mlp.experts.85.gate_proj", "model.language_model.layers.40.mlp.experts.86.gate_proj", "model.language_model.layers.40.mlp.experts.87.gate_proj", "model.language_model.layers.40.mlp.experts.88.gate_proj", "model.language_model.layers.40.mlp.experts.89.gate_proj", "model.language_model.layers.40.mlp.experts.90.gate_proj", "model.language_model.layers.40.mlp.experts.91.gate_proj", "model.language_model.layers.40.mlp.experts.92.gate_proj", "model.language_model.layers.40.mlp.experts.93.gate_proj", "model.language_model.layers.40.mlp.experts.94.gate_proj", "model.language_model.layers.40.mlp.experts.95.gate_proj", "model.language_model.layers.40.mlp.experts.96.gate_proj", "model.language_model.layers.40.mlp.experts.97.gate_proj", "model.language_model.layers.40.mlp.experts.98.gate_proj", "model.language_model.layers.40.mlp.experts.99.gate_proj", "model.language_model.layers.40.mlp.experts.100.gate_proj", "model.language_model.layers.40.mlp.experts.101.gate_proj", "model.language_model.layers.40.mlp.experts.102.gate_proj", "model.language_model.layers.40.mlp.experts.103.gate_proj", "model.language_model.layers.40.mlp.experts.104.gate_proj", "model.language_model.layers.40.mlp.experts.105.gate_proj", "model.language_model.layers.40.mlp.experts.106.gate_proj", "model.language_model.layers.40.mlp.experts.107.gate_proj", "model.language_model.layers.40.mlp.experts.108.gate_proj", "model.language_model.layers.40.mlp.experts.109.gate_proj", "model.language_model.layers.40.mlp.experts.110.gate_proj", "model.language_model.layers.40.mlp.experts.111.gate_proj", "model.language_model.layers.40.mlp.experts.112.gate_proj", "model.language_model.layers.40.mlp.experts.113.gate_proj", "model.language_model.layers.40.mlp.experts.114.gate_proj", "model.language_model.layers.40.mlp.experts.115.gate_proj", "model.language_model.layers.40.mlp.experts.116.gate_proj", "model.language_model.layers.40.mlp.experts.117.gate_proj", "model.language_model.layers.40.mlp.experts.118.gate_proj", "model.language_model.layers.40.mlp.experts.119.gate_proj", "model.language_model.layers.40.mlp.experts.120.gate_proj", "model.language_model.layers.40.mlp.experts.121.gate_proj", "model.language_model.layers.40.mlp.experts.122.gate_proj", "model.language_model.layers.40.mlp.experts.123.gate_proj", "model.language_model.layers.40.mlp.experts.124.gate_proj", "model.language_model.layers.40.mlp.experts.125.gate_proj", "model.language_model.layers.40.mlp.experts.126.gate_proj", "model.language_model.layers.40.mlp.experts.127.gate_proj", "model.language_model.layers.40.mlp.experts.0.up_proj", "model.language_model.layers.40.mlp.experts.1.up_proj", "model.language_model.layers.40.mlp.experts.2.up_proj", "model.language_model.layers.40.mlp.experts.3.up_proj", "model.language_model.layers.40.mlp.experts.4.up_proj", "model.language_model.layers.40.mlp.experts.5.up_proj", "model.language_model.layers.40.mlp.experts.6.up_proj", "model.language_model.layers.40.mlp.experts.7.up_proj", "model.language_model.layers.40.mlp.experts.8.up_proj", "model.language_model.layers.40.mlp.experts.9.up_proj", "model.language_model.layers.40.mlp.experts.10.up_proj", "model.language_model.layers.40.mlp.experts.11.up_proj", "model.language_model.layers.40.mlp.experts.12.up_proj", "model.language_model.layers.40.mlp.experts.13.up_proj", "model.language_model.layers.40.mlp.experts.14.up_proj", "model.language_model.layers.40.mlp.experts.15.up_proj", "model.language_model.layers.40.mlp.experts.16.up_proj", "model.language_model.layers.40.mlp.experts.17.up_proj", "model.language_model.layers.40.mlp.experts.18.up_proj", "model.language_model.layers.40.mlp.experts.19.up_proj", "model.language_model.layers.40.mlp.experts.20.up_proj", "model.language_model.layers.40.mlp.experts.21.up_proj", "model.language_model.layers.40.mlp.experts.22.up_proj", "model.language_model.layers.40.mlp.experts.23.up_proj", "model.language_model.layers.40.mlp.experts.24.up_proj", "model.language_model.layers.40.mlp.experts.25.up_proj", "model.language_model.layers.40.mlp.experts.26.up_proj", "model.language_model.layers.40.mlp.experts.27.up_proj", "model.language_model.layers.40.mlp.experts.28.up_proj", "model.language_model.layers.40.mlp.experts.29.up_proj", "model.language_model.layers.40.mlp.experts.30.up_proj", "model.language_model.layers.40.mlp.experts.31.up_proj", "model.language_model.layers.40.mlp.experts.32.up_proj", "model.language_model.layers.40.mlp.experts.33.up_proj", "model.language_model.layers.40.mlp.experts.34.up_proj", "model.language_model.layers.40.mlp.experts.35.up_proj", "model.language_model.layers.40.mlp.experts.36.up_proj", "model.language_model.layers.40.mlp.experts.37.up_proj", "model.language_model.layers.40.mlp.experts.38.up_proj", "model.language_model.layers.40.mlp.experts.39.up_proj", "model.language_model.layers.40.mlp.experts.40.up_proj", "model.language_model.layers.40.mlp.experts.41.up_proj", "model.language_model.layers.40.mlp.experts.42.up_proj", "model.language_model.layers.40.mlp.experts.43.up_proj", "model.language_model.layers.40.mlp.experts.44.up_proj", "model.language_model.layers.40.mlp.experts.45.up_proj", "model.language_model.layers.40.mlp.experts.46.up_proj", "model.language_model.layers.40.mlp.experts.47.up_proj", "model.language_model.layers.40.mlp.experts.48.up_proj", "model.language_model.layers.40.mlp.experts.49.up_proj", "model.language_model.layers.40.mlp.experts.50.up_proj", "model.language_model.layers.40.mlp.experts.51.up_proj", "model.language_model.layers.40.mlp.experts.52.up_proj", "model.language_model.layers.40.mlp.experts.53.up_proj", "model.language_model.layers.40.mlp.experts.54.up_proj", "model.language_model.layers.40.mlp.experts.55.up_proj", "model.language_model.layers.40.mlp.experts.56.up_proj", "model.language_model.layers.40.mlp.experts.57.up_proj", "model.language_model.layers.40.mlp.experts.58.up_proj", "model.language_model.layers.40.mlp.experts.59.up_proj", "model.language_model.layers.40.mlp.experts.60.up_proj", "model.language_model.layers.40.mlp.experts.61.up_proj", "model.language_model.layers.40.mlp.experts.62.up_proj", "model.language_model.layers.40.mlp.experts.63.up_proj", "model.language_model.layers.40.mlp.experts.64.up_proj", "model.language_model.layers.40.mlp.experts.65.up_proj", "model.language_model.layers.40.mlp.experts.66.up_proj", "model.language_model.layers.40.mlp.experts.67.up_proj", "model.language_model.layers.40.mlp.experts.68.up_proj", "model.language_model.layers.40.mlp.experts.69.up_proj", "model.language_model.layers.40.mlp.experts.70.up_proj", "model.language_model.layers.40.mlp.experts.71.up_proj", "model.language_model.layers.40.mlp.experts.72.up_proj", "model.language_model.layers.40.mlp.experts.73.up_proj", "model.language_model.layers.40.mlp.experts.74.up_proj", "model.language_model.layers.40.mlp.experts.75.up_proj", "model.language_model.layers.40.mlp.experts.76.up_proj", "model.language_model.layers.40.mlp.experts.77.up_proj", "model.language_model.layers.40.mlp.experts.78.up_proj", "model.language_model.layers.40.mlp.experts.79.up_proj", "model.language_model.layers.40.mlp.experts.80.up_proj", "model.language_model.layers.40.mlp.experts.81.up_proj", "model.language_model.layers.40.mlp.experts.82.up_proj", "model.language_model.layers.40.mlp.experts.83.up_proj", "model.language_model.layers.40.mlp.experts.84.up_proj", "model.language_model.layers.40.mlp.experts.85.up_proj", "model.language_model.layers.40.mlp.experts.86.up_proj", "model.language_model.layers.40.mlp.experts.87.up_proj", "model.language_model.layers.40.mlp.experts.88.up_proj", "model.language_model.layers.40.mlp.experts.89.up_proj", "model.language_model.layers.40.mlp.experts.90.up_proj", "model.language_model.layers.40.mlp.experts.91.up_proj", "model.language_model.layers.40.mlp.experts.92.up_proj", "model.language_model.layers.40.mlp.experts.93.up_proj", "model.language_model.layers.40.mlp.experts.94.up_proj", "model.language_model.layers.40.mlp.experts.95.up_proj", "model.language_model.layers.40.mlp.experts.96.up_proj", "model.language_model.layers.40.mlp.experts.97.up_proj", "model.language_model.layers.40.mlp.experts.98.up_proj", "model.language_model.layers.40.mlp.experts.99.up_proj", "model.language_model.layers.40.mlp.experts.100.up_proj", "model.language_model.layers.40.mlp.experts.101.up_proj", "model.language_model.layers.40.mlp.experts.102.up_proj", "model.language_model.layers.40.mlp.experts.103.up_proj", "model.language_model.layers.40.mlp.experts.104.up_proj", "model.language_model.layers.40.mlp.experts.105.up_proj", "model.language_model.layers.40.mlp.experts.106.up_proj", "model.language_model.layers.40.mlp.experts.107.up_proj", "model.language_model.layers.40.mlp.experts.108.up_proj", "model.language_model.layers.40.mlp.experts.109.up_proj", "model.language_model.layers.40.mlp.experts.110.up_proj", "model.language_model.layers.40.mlp.experts.111.up_proj", "model.language_model.layers.40.mlp.experts.112.up_proj", "model.language_model.layers.40.mlp.experts.113.up_proj", "model.language_model.layers.40.mlp.experts.114.up_proj", "model.language_model.layers.40.mlp.experts.115.up_proj", "model.language_model.layers.40.mlp.experts.116.up_proj", "model.language_model.layers.40.mlp.experts.117.up_proj", "model.language_model.layers.40.mlp.experts.118.up_proj", "model.language_model.layers.40.mlp.experts.119.up_proj", "model.language_model.layers.40.mlp.experts.120.up_proj", "model.language_model.layers.40.mlp.experts.121.up_proj", "model.language_model.layers.40.mlp.experts.122.up_proj", "model.language_model.layers.40.mlp.experts.123.up_proj", "model.language_model.layers.40.mlp.experts.124.up_proj", "model.language_model.layers.40.mlp.experts.125.up_proj", "model.language_model.layers.40.mlp.experts.126.up_proj", "model.language_model.layers.40.mlp.experts.127.up_proj", "model.language_model.layers.40.mlp.experts.0.down_proj", "model.language_model.layers.40.mlp.experts.1.down_proj", "model.language_model.layers.40.mlp.experts.2.down_proj", "model.language_model.layers.40.mlp.experts.3.down_proj", "model.language_model.layers.40.mlp.experts.4.down_proj", "model.language_model.layers.40.mlp.experts.5.down_proj", "model.language_model.layers.40.mlp.experts.6.down_proj", "model.language_model.layers.40.mlp.experts.7.down_proj", "model.language_model.layers.40.mlp.experts.8.down_proj", "model.language_model.layers.40.mlp.experts.9.down_proj", "model.language_model.layers.40.mlp.experts.10.down_proj", "model.language_model.layers.40.mlp.experts.11.down_proj", "model.language_model.layers.40.mlp.experts.12.down_proj", "model.language_model.layers.40.mlp.experts.13.down_proj", "model.language_model.layers.40.mlp.experts.14.down_proj", "model.language_model.layers.40.mlp.experts.15.down_proj", "model.language_model.layers.40.mlp.experts.16.down_proj", "model.language_model.layers.40.mlp.experts.17.down_proj", "model.language_model.layers.40.mlp.experts.18.down_proj", "model.language_model.layers.40.mlp.experts.19.down_proj", "model.language_model.layers.40.mlp.experts.20.down_proj", "model.language_model.layers.40.mlp.experts.21.down_proj", "model.language_model.layers.40.mlp.experts.22.down_proj", "model.language_model.layers.40.mlp.experts.23.down_proj", "model.language_model.layers.40.mlp.experts.24.down_proj", "model.language_model.layers.40.mlp.experts.25.down_proj", "model.language_model.layers.40.mlp.experts.26.down_proj", "model.language_model.layers.40.mlp.experts.27.down_proj", "model.language_model.layers.40.mlp.experts.28.down_proj", "model.language_model.layers.40.mlp.experts.29.down_proj", "model.language_model.layers.40.mlp.experts.30.down_proj", "model.language_model.layers.40.mlp.experts.31.down_proj", "model.language_model.layers.40.mlp.experts.32.down_proj", "model.language_model.layers.40.mlp.experts.33.down_proj", "model.language_model.layers.40.mlp.experts.34.down_proj", "model.language_model.layers.40.mlp.experts.35.down_proj", "model.language_model.layers.40.mlp.experts.36.down_proj", "model.language_model.layers.40.mlp.experts.37.down_proj", "model.language_model.layers.40.mlp.experts.38.down_proj", "model.language_model.layers.40.mlp.experts.39.down_proj", "model.language_model.layers.40.mlp.experts.40.down_proj", "model.language_model.layers.40.mlp.experts.41.down_proj", "model.language_model.layers.40.mlp.experts.42.down_proj", "model.language_model.layers.40.mlp.experts.43.down_proj", "model.language_model.layers.40.mlp.experts.44.down_proj", "model.language_model.layers.40.mlp.experts.45.down_proj", "model.language_model.layers.40.mlp.experts.46.down_proj", "model.language_model.layers.40.mlp.experts.47.down_proj", "model.language_model.layers.40.mlp.experts.48.down_proj", "model.language_model.layers.40.mlp.experts.49.down_proj", "model.language_model.layers.40.mlp.experts.50.down_proj", "model.language_model.layers.40.mlp.experts.51.down_proj", "model.language_model.layers.40.mlp.experts.52.down_proj", "model.language_model.layers.40.mlp.experts.53.down_proj", "model.language_model.layers.40.mlp.experts.54.down_proj", "model.language_model.layers.40.mlp.experts.55.down_proj", "model.language_model.layers.40.mlp.experts.56.down_proj", "model.language_model.layers.40.mlp.experts.57.down_proj", "model.language_model.layers.40.mlp.experts.58.down_proj", "model.language_model.layers.40.mlp.experts.59.down_proj", "model.language_model.layers.40.mlp.experts.60.down_proj", "model.language_model.layers.40.mlp.experts.61.down_proj", "model.language_model.layers.40.mlp.experts.62.down_proj", "model.language_model.layers.40.mlp.experts.63.down_proj", "model.language_model.layers.40.mlp.experts.64.down_proj", "model.language_model.layers.40.mlp.experts.65.down_proj", "model.language_model.layers.40.mlp.experts.66.down_proj", "model.language_model.layers.40.mlp.experts.67.down_proj", "model.language_model.layers.40.mlp.experts.68.down_proj", "model.language_model.layers.40.mlp.experts.69.down_proj", "model.language_model.layers.40.mlp.experts.70.down_proj", "model.language_model.layers.40.mlp.experts.71.down_proj", "model.language_model.layers.40.mlp.experts.72.down_proj", "model.language_model.layers.40.mlp.experts.73.down_proj", "model.language_model.layers.40.mlp.experts.74.down_proj", "model.language_model.layers.40.mlp.experts.75.down_proj", "model.language_model.layers.40.mlp.experts.76.down_proj", "model.language_model.layers.40.mlp.experts.77.down_proj", "model.language_model.layers.40.mlp.experts.78.down_proj", "model.language_model.layers.40.mlp.experts.79.down_proj", "model.language_model.layers.40.mlp.experts.80.down_proj", "model.language_model.layers.40.mlp.experts.81.down_proj", "model.language_model.layers.40.mlp.experts.82.down_proj", "model.language_model.layers.40.mlp.experts.83.down_proj", "model.language_model.layers.40.mlp.experts.84.down_proj", "model.language_model.layers.40.mlp.experts.85.down_proj", "model.language_model.layers.40.mlp.experts.86.down_proj", "model.language_model.layers.40.mlp.experts.87.down_proj", "model.language_model.layers.40.mlp.experts.88.down_proj", "model.language_model.layers.40.mlp.experts.89.down_proj", "model.language_model.layers.40.mlp.experts.90.down_proj", "model.language_model.layers.40.mlp.experts.91.down_proj", "model.language_model.layers.40.mlp.experts.92.down_proj", "model.language_model.layers.40.mlp.experts.93.down_proj", "model.language_model.layers.40.mlp.experts.94.down_proj", "model.language_model.layers.40.mlp.experts.95.down_proj", "model.language_model.layers.40.mlp.experts.96.down_proj", "model.language_model.layers.40.mlp.experts.97.down_proj", "model.language_model.layers.40.mlp.experts.98.down_proj", "model.language_model.layers.40.mlp.experts.99.down_proj", "model.language_model.layers.40.mlp.experts.100.down_proj", "model.language_model.layers.40.mlp.experts.101.down_proj", "model.language_model.layers.40.mlp.experts.102.down_proj", "model.language_model.layers.40.mlp.experts.103.down_proj", "model.language_model.layers.40.mlp.experts.104.down_proj", "model.language_model.layers.40.mlp.experts.105.down_proj", "model.language_model.layers.40.mlp.experts.106.down_proj", "model.language_model.layers.40.mlp.experts.107.down_proj", "model.language_model.layers.40.mlp.experts.108.down_proj", "model.language_model.layers.40.mlp.experts.109.down_proj", "model.language_model.layers.40.mlp.experts.110.down_proj", "model.language_model.layers.40.mlp.experts.111.down_proj", "model.language_model.layers.40.mlp.experts.112.down_proj", "model.language_model.layers.40.mlp.experts.113.down_proj", "model.language_model.layers.40.mlp.experts.114.down_proj", "model.language_model.layers.40.mlp.experts.115.down_proj", "model.language_model.layers.40.mlp.experts.116.down_proj", "model.language_model.layers.40.mlp.experts.117.down_proj", "model.language_model.layers.40.mlp.experts.118.down_proj", "model.language_model.layers.40.mlp.experts.119.down_proj", "model.language_model.layers.40.mlp.experts.120.down_proj", "model.language_model.layers.40.mlp.experts.121.down_proj", "model.language_model.layers.40.mlp.experts.122.down_proj", "model.language_model.layers.40.mlp.experts.123.down_proj", "model.language_model.layers.40.mlp.experts.124.down_proj", "model.language_model.layers.40.mlp.experts.125.down_proj", "model.language_model.layers.40.mlp.experts.126.down_proj", "model.language_model.layers.40.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.000845912285149103, "dbits": 603979776 } ] }, { "idx": 82, "layers": [ "model.language_model.layers.41.self_attn.q_proj", "model.language_model.layers.41.self_attn.k_proj", "model.language_model.layers.41.self_attn.v_proj", "model.language_model.layers.41.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0010907790623605196, "dbits": 18874368 } ] }, { "idx": 83, "layers": [ "model.language_model.layers.41.mlp.experts.0.gate_proj", "model.language_model.layers.41.mlp.experts.1.gate_proj", "model.language_model.layers.41.mlp.experts.2.gate_proj", "model.language_model.layers.41.mlp.experts.3.gate_proj", "model.language_model.layers.41.mlp.experts.4.gate_proj", "model.language_model.layers.41.mlp.experts.5.gate_proj", "model.language_model.layers.41.mlp.experts.6.gate_proj", "model.language_model.layers.41.mlp.experts.7.gate_proj", "model.language_model.layers.41.mlp.experts.8.gate_proj", "model.language_model.layers.41.mlp.experts.9.gate_proj", "model.language_model.layers.41.mlp.experts.10.gate_proj", "model.language_model.layers.41.mlp.experts.11.gate_proj", "model.language_model.layers.41.mlp.experts.12.gate_proj", "model.language_model.layers.41.mlp.experts.13.gate_proj", "model.language_model.layers.41.mlp.experts.14.gate_proj", "model.language_model.layers.41.mlp.experts.15.gate_proj", "model.language_model.layers.41.mlp.experts.16.gate_proj", "model.language_model.layers.41.mlp.experts.17.gate_proj", "model.language_model.layers.41.mlp.experts.18.gate_proj", "model.language_model.layers.41.mlp.experts.19.gate_proj", "model.language_model.layers.41.mlp.experts.20.gate_proj", "model.language_model.layers.41.mlp.experts.21.gate_proj", "model.language_model.layers.41.mlp.experts.22.gate_proj", "model.language_model.layers.41.mlp.experts.23.gate_proj", "model.language_model.layers.41.mlp.experts.24.gate_proj", "model.language_model.layers.41.mlp.experts.25.gate_proj", "model.language_model.layers.41.mlp.experts.26.gate_proj", "model.language_model.layers.41.mlp.experts.27.gate_proj", "model.language_model.layers.41.mlp.experts.28.gate_proj", "model.language_model.layers.41.mlp.experts.29.gate_proj", "model.language_model.layers.41.mlp.experts.30.gate_proj", "model.language_model.layers.41.mlp.experts.31.gate_proj", "model.language_model.layers.41.mlp.experts.32.gate_proj", "model.language_model.layers.41.mlp.experts.33.gate_proj", "model.language_model.layers.41.mlp.experts.34.gate_proj", "model.language_model.layers.41.mlp.experts.35.gate_proj", "model.language_model.layers.41.mlp.experts.36.gate_proj", "model.language_model.layers.41.mlp.experts.37.gate_proj", "model.language_model.layers.41.mlp.experts.38.gate_proj", "model.language_model.layers.41.mlp.experts.39.gate_proj", "model.language_model.layers.41.mlp.experts.40.gate_proj", "model.language_model.layers.41.mlp.experts.41.gate_proj", "model.language_model.layers.41.mlp.experts.42.gate_proj", "model.language_model.layers.41.mlp.experts.43.gate_proj", "model.language_model.layers.41.mlp.experts.44.gate_proj", "model.language_model.layers.41.mlp.experts.45.gate_proj", "model.language_model.layers.41.mlp.experts.46.gate_proj", "model.language_model.layers.41.mlp.experts.47.gate_proj", "model.language_model.layers.41.mlp.experts.48.gate_proj", "model.language_model.layers.41.mlp.experts.49.gate_proj", "model.language_model.layers.41.mlp.experts.50.gate_proj", "model.language_model.layers.41.mlp.experts.51.gate_proj", "model.language_model.layers.41.mlp.experts.52.gate_proj", "model.language_model.layers.41.mlp.experts.53.gate_proj", "model.language_model.layers.41.mlp.experts.54.gate_proj", "model.language_model.layers.41.mlp.experts.55.gate_proj", "model.language_model.layers.41.mlp.experts.56.gate_proj", "model.language_model.layers.41.mlp.experts.57.gate_proj", "model.language_model.layers.41.mlp.experts.58.gate_proj", "model.language_model.layers.41.mlp.experts.59.gate_proj", "model.language_model.layers.41.mlp.experts.60.gate_proj", "model.language_model.layers.41.mlp.experts.61.gate_proj", "model.language_model.layers.41.mlp.experts.62.gate_proj", "model.language_model.layers.41.mlp.experts.63.gate_proj", "model.language_model.layers.41.mlp.experts.64.gate_proj", "model.language_model.layers.41.mlp.experts.65.gate_proj", "model.language_model.layers.41.mlp.experts.66.gate_proj", "model.language_model.layers.41.mlp.experts.67.gate_proj", "model.language_model.layers.41.mlp.experts.68.gate_proj", "model.language_model.layers.41.mlp.experts.69.gate_proj", "model.language_model.layers.41.mlp.experts.70.gate_proj", "model.language_model.layers.41.mlp.experts.71.gate_proj", "model.language_model.layers.41.mlp.experts.72.gate_proj", "model.language_model.layers.41.mlp.experts.73.gate_proj", "model.language_model.layers.41.mlp.experts.74.gate_proj", "model.language_model.layers.41.mlp.experts.75.gate_proj", "model.language_model.layers.41.mlp.experts.76.gate_proj", "model.language_model.layers.41.mlp.experts.77.gate_proj", "model.language_model.layers.41.mlp.experts.78.gate_proj", "model.language_model.layers.41.mlp.experts.79.gate_proj", "model.language_model.layers.41.mlp.experts.80.gate_proj", "model.language_model.layers.41.mlp.experts.81.gate_proj", "model.language_model.layers.41.mlp.experts.82.gate_proj", "model.language_model.layers.41.mlp.experts.83.gate_proj", "model.language_model.layers.41.mlp.experts.84.gate_proj", "model.language_model.layers.41.mlp.experts.85.gate_proj", "model.language_model.layers.41.mlp.experts.86.gate_proj", "model.language_model.layers.41.mlp.experts.87.gate_proj", "model.language_model.layers.41.mlp.experts.88.gate_proj", "model.language_model.layers.41.mlp.experts.89.gate_proj", "model.language_model.layers.41.mlp.experts.90.gate_proj", "model.language_model.layers.41.mlp.experts.91.gate_proj", "model.language_model.layers.41.mlp.experts.92.gate_proj", "model.language_model.layers.41.mlp.experts.93.gate_proj", "model.language_model.layers.41.mlp.experts.94.gate_proj", "model.language_model.layers.41.mlp.experts.95.gate_proj", "model.language_model.layers.41.mlp.experts.96.gate_proj", "model.language_model.layers.41.mlp.experts.97.gate_proj", "model.language_model.layers.41.mlp.experts.98.gate_proj", "model.language_model.layers.41.mlp.experts.99.gate_proj", "model.language_model.layers.41.mlp.experts.100.gate_proj", "model.language_model.layers.41.mlp.experts.101.gate_proj", "model.language_model.layers.41.mlp.experts.102.gate_proj", "model.language_model.layers.41.mlp.experts.103.gate_proj", "model.language_model.layers.41.mlp.experts.104.gate_proj", "model.language_model.layers.41.mlp.experts.105.gate_proj", "model.language_model.layers.41.mlp.experts.106.gate_proj", "model.language_model.layers.41.mlp.experts.107.gate_proj", "model.language_model.layers.41.mlp.experts.108.gate_proj", "model.language_model.layers.41.mlp.experts.109.gate_proj", "model.language_model.layers.41.mlp.experts.110.gate_proj", "model.language_model.layers.41.mlp.experts.111.gate_proj", "model.language_model.layers.41.mlp.experts.112.gate_proj", "model.language_model.layers.41.mlp.experts.113.gate_proj", "model.language_model.layers.41.mlp.experts.114.gate_proj", "model.language_model.layers.41.mlp.experts.115.gate_proj", "model.language_model.layers.41.mlp.experts.116.gate_proj", "model.language_model.layers.41.mlp.experts.117.gate_proj", "model.language_model.layers.41.mlp.experts.118.gate_proj", "model.language_model.layers.41.mlp.experts.119.gate_proj", "model.language_model.layers.41.mlp.experts.120.gate_proj", "model.language_model.layers.41.mlp.experts.121.gate_proj", "model.language_model.layers.41.mlp.experts.122.gate_proj", "model.language_model.layers.41.mlp.experts.123.gate_proj", "model.language_model.layers.41.mlp.experts.124.gate_proj", "model.language_model.layers.41.mlp.experts.125.gate_proj", "model.language_model.layers.41.mlp.experts.126.gate_proj", "model.language_model.layers.41.mlp.experts.127.gate_proj", "model.language_model.layers.41.mlp.experts.0.up_proj", "model.language_model.layers.41.mlp.experts.1.up_proj", "model.language_model.layers.41.mlp.experts.2.up_proj", "model.language_model.layers.41.mlp.experts.3.up_proj", "model.language_model.layers.41.mlp.experts.4.up_proj", "model.language_model.layers.41.mlp.experts.5.up_proj", "model.language_model.layers.41.mlp.experts.6.up_proj", "model.language_model.layers.41.mlp.experts.7.up_proj", "model.language_model.layers.41.mlp.experts.8.up_proj", "model.language_model.layers.41.mlp.experts.9.up_proj", "model.language_model.layers.41.mlp.experts.10.up_proj", "model.language_model.layers.41.mlp.experts.11.up_proj", "model.language_model.layers.41.mlp.experts.12.up_proj", "model.language_model.layers.41.mlp.experts.13.up_proj", "model.language_model.layers.41.mlp.experts.14.up_proj", "model.language_model.layers.41.mlp.experts.15.up_proj", "model.language_model.layers.41.mlp.experts.16.up_proj", "model.language_model.layers.41.mlp.experts.17.up_proj", "model.language_model.layers.41.mlp.experts.18.up_proj", "model.language_model.layers.41.mlp.experts.19.up_proj", "model.language_model.layers.41.mlp.experts.20.up_proj", "model.language_model.layers.41.mlp.experts.21.up_proj", "model.language_model.layers.41.mlp.experts.22.up_proj", "model.language_model.layers.41.mlp.experts.23.up_proj", "model.language_model.layers.41.mlp.experts.24.up_proj", "model.language_model.layers.41.mlp.experts.25.up_proj", "model.language_model.layers.41.mlp.experts.26.up_proj", "model.language_model.layers.41.mlp.experts.27.up_proj", "model.language_model.layers.41.mlp.experts.28.up_proj", "model.language_model.layers.41.mlp.experts.29.up_proj", "model.language_model.layers.41.mlp.experts.30.up_proj", "model.language_model.layers.41.mlp.experts.31.up_proj", "model.language_model.layers.41.mlp.experts.32.up_proj", "model.language_model.layers.41.mlp.experts.33.up_proj", "model.language_model.layers.41.mlp.experts.34.up_proj", "model.language_model.layers.41.mlp.experts.35.up_proj", "model.language_model.layers.41.mlp.experts.36.up_proj", "model.language_model.layers.41.mlp.experts.37.up_proj", "model.language_model.layers.41.mlp.experts.38.up_proj", "model.language_model.layers.41.mlp.experts.39.up_proj", "model.language_model.layers.41.mlp.experts.40.up_proj", "model.language_model.layers.41.mlp.experts.41.up_proj", "model.language_model.layers.41.mlp.experts.42.up_proj", "model.language_model.layers.41.mlp.experts.43.up_proj", "model.language_model.layers.41.mlp.experts.44.up_proj", "model.language_model.layers.41.mlp.experts.45.up_proj", "model.language_model.layers.41.mlp.experts.46.up_proj", "model.language_model.layers.41.mlp.experts.47.up_proj", "model.language_model.layers.41.mlp.experts.48.up_proj", "model.language_model.layers.41.mlp.experts.49.up_proj", "model.language_model.layers.41.mlp.experts.50.up_proj", "model.language_model.layers.41.mlp.experts.51.up_proj", "model.language_model.layers.41.mlp.experts.52.up_proj", "model.language_model.layers.41.mlp.experts.53.up_proj", "model.language_model.layers.41.mlp.experts.54.up_proj", "model.language_model.layers.41.mlp.experts.55.up_proj", "model.language_model.layers.41.mlp.experts.56.up_proj", "model.language_model.layers.41.mlp.experts.57.up_proj", "model.language_model.layers.41.mlp.experts.58.up_proj", "model.language_model.layers.41.mlp.experts.59.up_proj", "model.language_model.layers.41.mlp.experts.60.up_proj", "model.language_model.layers.41.mlp.experts.61.up_proj", "model.language_model.layers.41.mlp.experts.62.up_proj", "model.language_model.layers.41.mlp.experts.63.up_proj", "model.language_model.layers.41.mlp.experts.64.up_proj", "model.language_model.layers.41.mlp.experts.65.up_proj", "model.language_model.layers.41.mlp.experts.66.up_proj", "model.language_model.layers.41.mlp.experts.67.up_proj", "model.language_model.layers.41.mlp.experts.68.up_proj", "model.language_model.layers.41.mlp.experts.69.up_proj", "model.language_model.layers.41.mlp.experts.70.up_proj", "model.language_model.layers.41.mlp.experts.71.up_proj", "model.language_model.layers.41.mlp.experts.72.up_proj", "model.language_model.layers.41.mlp.experts.73.up_proj", "model.language_model.layers.41.mlp.experts.74.up_proj", "model.language_model.layers.41.mlp.experts.75.up_proj", "model.language_model.layers.41.mlp.experts.76.up_proj", "model.language_model.layers.41.mlp.experts.77.up_proj", "model.language_model.layers.41.mlp.experts.78.up_proj", "model.language_model.layers.41.mlp.experts.79.up_proj", "model.language_model.layers.41.mlp.experts.80.up_proj", "model.language_model.layers.41.mlp.experts.81.up_proj", "model.language_model.layers.41.mlp.experts.82.up_proj", "model.language_model.layers.41.mlp.experts.83.up_proj", "model.language_model.layers.41.mlp.experts.84.up_proj", "model.language_model.layers.41.mlp.experts.85.up_proj", "model.language_model.layers.41.mlp.experts.86.up_proj", "model.language_model.layers.41.mlp.experts.87.up_proj", "model.language_model.layers.41.mlp.experts.88.up_proj", "model.language_model.layers.41.mlp.experts.89.up_proj", "model.language_model.layers.41.mlp.experts.90.up_proj", "model.language_model.layers.41.mlp.experts.91.up_proj", "model.language_model.layers.41.mlp.experts.92.up_proj", "model.language_model.layers.41.mlp.experts.93.up_proj", "model.language_model.layers.41.mlp.experts.94.up_proj", "model.language_model.layers.41.mlp.experts.95.up_proj", "model.language_model.layers.41.mlp.experts.96.up_proj", "model.language_model.layers.41.mlp.experts.97.up_proj", "model.language_model.layers.41.mlp.experts.98.up_proj", "model.language_model.layers.41.mlp.experts.99.up_proj", "model.language_model.layers.41.mlp.experts.100.up_proj", "model.language_model.layers.41.mlp.experts.101.up_proj", "model.language_model.layers.41.mlp.experts.102.up_proj", "model.language_model.layers.41.mlp.experts.103.up_proj", "model.language_model.layers.41.mlp.experts.104.up_proj", "model.language_model.layers.41.mlp.experts.105.up_proj", "model.language_model.layers.41.mlp.experts.106.up_proj", "model.language_model.layers.41.mlp.experts.107.up_proj", "model.language_model.layers.41.mlp.experts.108.up_proj", "model.language_model.layers.41.mlp.experts.109.up_proj", "model.language_model.layers.41.mlp.experts.110.up_proj", "model.language_model.layers.41.mlp.experts.111.up_proj", "model.language_model.layers.41.mlp.experts.112.up_proj", "model.language_model.layers.41.mlp.experts.113.up_proj", "model.language_model.layers.41.mlp.experts.114.up_proj", "model.language_model.layers.41.mlp.experts.115.up_proj", "model.language_model.layers.41.mlp.experts.116.up_proj", "model.language_model.layers.41.mlp.experts.117.up_proj", "model.language_model.layers.41.mlp.experts.118.up_proj", "model.language_model.layers.41.mlp.experts.119.up_proj", "model.language_model.layers.41.mlp.experts.120.up_proj", "model.language_model.layers.41.mlp.experts.121.up_proj", "model.language_model.layers.41.mlp.experts.122.up_proj", "model.language_model.layers.41.mlp.experts.123.up_proj", "model.language_model.layers.41.mlp.experts.124.up_proj", "model.language_model.layers.41.mlp.experts.125.up_proj", "model.language_model.layers.41.mlp.experts.126.up_proj", "model.language_model.layers.41.mlp.experts.127.up_proj", "model.language_model.layers.41.mlp.experts.0.down_proj", "model.language_model.layers.41.mlp.experts.1.down_proj", "model.language_model.layers.41.mlp.experts.2.down_proj", "model.language_model.layers.41.mlp.experts.3.down_proj", "model.language_model.layers.41.mlp.experts.4.down_proj", "model.language_model.layers.41.mlp.experts.5.down_proj", "model.language_model.layers.41.mlp.experts.6.down_proj", "model.language_model.layers.41.mlp.experts.7.down_proj", "model.language_model.layers.41.mlp.experts.8.down_proj", "model.language_model.layers.41.mlp.experts.9.down_proj", "model.language_model.layers.41.mlp.experts.10.down_proj", "model.language_model.layers.41.mlp.experts.11.down_proj", "model.language_model.layers.41.mlp.experts.12.down_proj", "model.language_model.layers.41.mlp.experts.13.down_proj", "model.language_model.layers.41.mlp.experts.14.down_proj", "model.language_model.layers.41.mlp.experts.15.down_proj", "model.language_model.layers.41.mlp.experts.16.down_proj", "model.language_model.layers.41.mlp.experts.17.down_proj", "model.language_model.layers.41.mlp.experts.18.down_proj", "model.language_model.layers.41.mlp.experts.19.down_proj", "model.language_model.layers.41.mlp.experts.20.down_proj", "model.language_model.layers.41.mlp.experts.21.down_proj", "model.language_model.layers.41.mlp.experts.22.down_proj", "model.language_model.layers.41.mlp.experts.23.down_proj", "model.language_model.layers.41.mlp.experts.24.down_proj", "model.language_model.layers.41.mlp.experts.25.down_proj", "model.language_model.layers.41.mlp.experts.26.down_proj", "model.language_model.layers.41.mlp.experts.27.down_proj", "model.language_model.layers.41.mlp.experts.28.down_proj", "model.language_model.layers.41.mlp.experts.29.down_proj", "model.language_model.layers.41.mlp.experts.30.down_proj", "model.language_model.layers.41.mlp.experts.31.down_proj", "model.language_model.layers.41.mlp.experts.32.down_proj", "model.language_model.layers.41.mlp.experts.33.down_proj", "model.language_model.layers.41.mlp.experts.34.down_proj", "model.language_model.layers.41.mlp.experts.35.down_proj", "model.language_model.layers.41.mlp.experts.36.down_proj", "model.language_model.layers.41.mlp.experts.37.down_proj", "model.language_model.layers.41.mlp.experts.38.down_proj", "model.language_model.layers.41.mlp.experts.39.down_proj", "model.language_model.layers.41.mlp.experts.40.down_proj", "model.language_model.layers.41.mlp.experts.41.down_proj", "model.language_model.layers.41.mlp.experts.42.down_proj", "model.language_model.layers.41.mlp.experts.43.down_proj", "model.language_model.layers.41.mlp.experts.44.down_proj", "model.language_model.layers.41.mlp.experts.45.down_proj", "model.language_model.layers.41.mlp.experts.46.down_proj", "model.language_model.layers.41.mlp.experts.47.down_proj", "model.language_model.layers.41.mlp.experts.48.down_proj", "model.language_model.layers.41.mlp.experts.49.down_proj", "model.language_model.layers.41.mlp.experts.50.down_proj", "model.language_model.layers.41.mlp.experts.51.down_proj", "model.language_model.layers.41.mlp.experts.52.down_proj", "model.language_model.layers.41.mlp.experts.53.down_proj", "model.language_model.layers.41.mlp.experts.54.down_proj", "model.language_model.layers.41.mlp.experts.55.down_proj", "model.language_model.layers.41.mlp.experts.56.down_proj", "model.language_model.layers.41.mlp.experts.57.down_proj", "model.language_model.layers.41.mlp.experts.58.down_proj", "model.language_model.layers.41.mlp.experts.59.down_proj", "model.language_model.layers.41.mlp.experts.60.down_proj", "model.language_model.layers.41.mlp.experts.61.down_proj", "model.language_model.layers.41.mlp.experts.62.down_proj", "model.language_model.layers.41.mlp.experts.63.down_proj", "model.language_model.layers.41.mlp.experts.64.down_proj", "model.language_model.layers.41.mlp.experts.65.down_proj", "model.language_model.layers.41.mlp.experts.66.down_proj", "model.language_model.layers.41.mlp.experts.67.down_proj", "model.language_model.layers.41.mlp.experts.68.down_proj", "model.language_model.layers.41.mlp.experts.69.down_proj", "model.language_model.layers.41.mlp.experts.70.down_proj", "model.language_model.layers.41.mlp.experts.71.down_proj", "model.language_model.layers.41.mlp.experts.72.down_proj", "model.language_model.layers.41.mlp.experts.73.down_proj", "model.language_model.layers.41.mlp.experts.74.down_proj", "model.language_model.layers.41.mlp.experts.75.down_proj", "model.language_model.layers.41.mlp.experts.76.down_proj", "model.language_model.layers.41.mlp.experts.77.down_proj", "model.language_model.layers.41.mlp.experts.78.down_proj", "model.language_model.layers.41.mlp.experts.79.down_proj", "model.language_model.layers.41.mlp.experts.80.down_proj", "model.language_model.layers.41.mlp.experts.81.down_proj", "model.language_model.layers.41.mlp.experts.82.down_proj", "model.language_model.layers.41.mlp.experts.83.down_proj", "model.language_model.layers.41.mlp.experts.84.down_proj", "model.language_model.layers.41.mlp.experts.85.down_proj", "model.language_model.layers.41.mlp.experts.86.down_proj", "model.language_model.layers.41.mlp.experts.87.down_proj", "model.language_model.layers.41.mlp.experts.88.down_proj", "model.language_model.layers.41.mlp.experts.89.down_proj", "model.language_model.layers.41.mlp.experts.90.down_proj", "model.language_model.layers.41.mlp.experts.91.down_proj", "model.language_model.layers.41.mlp.experts.92.down_proj", "model.language_model.layers.41.mlp.experts.93.down_proj", "model.language_model.layers.41.mlp.experts.94.down_proj", "model.language_model.layers.41.mlp.experts.95.down_proj", "model.language_model.layers.41.mlp.experts.96.down_proj", "model.language_model.layers.41.mlp.experts.97.down_proj", "model.language_model.layers.41.mlp.experts.98.down_proj", "model.language_model.layers.41.mlp.experts.99.down_proj", "model.language_model.layers.41.mlp.experts.100.down_proj", "model.language_model.layers.41.mlp.experts.101.down_proj", "model.language_model.layers.41.mlp.experts.102.down_proj", "model.language_model.layers.41.mlp.experts.103.down_proj", "model.language_model.layers.41.mlp.experts.104.down_proj", "model.language_model.layers.41.mlp.experts.105.down_proj", "model.language_model.layers.41.mlp.experts.106.down_proj", "model.language_model.layers.41.mlp.experts.107.down_proj", "model.language_model.layers.41.mlp.experts.108.down_proj", "model.language_model.layers.41.mlp.experts.109.down_proj", "model.language_model.layers.41.mlp.experts.110.down_proj", "model.language_model.layers.41.mlp.experts.111.down_proj", "model.language_model.layers.41.mlp.experts.112.down_proj", "model.language_model.layers.41.mlp.experts.113.down_proj", "model.language_model.layers.41.mlp.experts.114.down_proj", "model.language_model.layers.41.mlp.experts.115.down_proj", "model.language_model.layers.41.mlp.experts.116.down_proj", "model.language_model.layers.41.mlp.experts.117.down_proj", "model.language_model.layers.41.mlp.experts.118.down_proj", "model.language_model.layers.41.mlp.experts.119.down_proj", "model.language_model.layers.41.mlp.experts.120.down_proj", "model.language_model.layers.41.mlp.experts.121.down_proj", "model.language_model.layers.41.mlp.experts.122.down_proj", "model.language_model.layers.41.mlp.experts.123.down_proj", "model.language_model.layers.41.mlp.experts.124.down_proj", "model.language_model.layers.41.mlp.experts.125.down_proj", "model.language_model.layers.41.mlp.experts.126.down_proj", "model.language_model.layers.41.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.0009274060837924508, "dbits": 603979776 } ] }, { "idx": 84, "layers": [ "model.language_model.layers.42.self_attn.q_proj", "model.language_model.layers.42.self_attn.k_proj", "model.language_model.layers.42.self_attn.v_proj", "model.language_model.layers.42.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0009553127922117682, "dbits": 18874368 } ] }, { "idx": 85, "layers": [ "model.language_model.layers.42.mlp.experts.0.gate_proj", "model.language_model.layers.42.mlp.experts.1.gate_proj", "model.language_model.layers.42.mlp.experts.2.gate_proj", "model.language_model.layers.42.mlp.experts.3.gate_proj", "model.language_model.layers.42.mlp.experts.4.gate_proj", "model.language_model.layers.42.mlp.experts.5.gate_proj", "model.language_model.layers.42.mlp.experts.6.gate_proj", "model.language_model.layers.42.mlp.experts.7.gate_proj", "model.language_model.layers.42.mlp.experts.8.gate_proj", "model.language_model.layers.42.mlp.experts.9.gate_proj", "model.language_model.layers.42.mlp.experts.10.gate_proj", "model.language_model.layers.42.mlp.experts.11.gate_proj", "model.language_model.layers.42.mlp.experts.12.gate_proj", "model.language_model.layers.42.mlp.experts.13.gate_proj", "model.language_model.layers.42.mlp.experts.14.gate_proj", "model.language_model.layers.42.mlp.experts.15.gate_proj", "model.language_model.layers.42.mlp.experts.16.gate_proj", "model.language_model.layers.42.mlp.experts.17.gate_proj", "model.language_model.layers.42.mlp.experts.18.gate_proj", "model.language_model.layers.42.mlp.experts.19.gate_proj", "model.language_model.layers.42.mlp.experts.20.gate_proj", "model.language_model.layers.42.mlp.experts.21.gate_proj", "model.language_model.layers.42.mlp.experts.22.gate_proj", "model.language_model.layers.42.mlp.experts.23.gate_proj", "model.language_model.layers.42.mlp.experts.24.gate_proj", "model.language_model.layers.42.mlp.experts.25.gate_proj", "model.language_model.layers.42.mlp.experts.26.gate_proj", "model.language_model.layers.42.mlp.experts.27.gate_proj", "model.language_model.layers.42.mlp.experts.28.gate_proj", "model.language_model.layers.42.mlp.experts.29.gate_proj", "model.language_model.layers.42.mlp.experts.30.gate_proj", "model.language_model.layers.42.mlp.experts.31.gate_proj", "model.language_model.layers.42.mlp.experts.32.gate_proj", "model.language_model.layers.42.mlp.experts.33.gate_proj", "model.language_model.layers.42.mlp.experts.34.gate_proj", "model.language_model.layers.42.mlp.experts.35.gate_proj", "model.language_model.layers.42.mlp.experts.36.gate_proj", "model.language_model.layers.42.mlp.experts.37.gate_proj", "model.language_model.layers.42.mlp.experts.38.gate_proj", "model.language_model.layers.42.mlp.experts.39.gate_proj", "model.language_model.layers.42.mlp.experts.40.gate_proj", "model.language_model.layers.42.mlp.experts.41.gate_proj", "model.language_model.layers.42.mlp.experts.42.gate_proj", "model.language_model.layers.42.mlp.experts.43.gate_proj", "model.language_model.layers.42.mlp.experts.44.gate_proj", "model.language_model.layers.42.mlp.experts.45.gate_proj", "model.language_model.layers.42.mlp.experts.46.gate_proj", "model.language_model.layers.42.mlp.experts.47.gate_proj", "model.language_model.layers.42.mlp.experts.48.gate_proj", "model.language_model.layers.42.mlp.experts.49.gate_proj", "model.language_model.layers.42.mlp.experts.50.gate_proj", "model.language_model.layers.42.mlp.experts.51.gate_proj", "model.language_model.layers.42.mlp.experts.52.gate_proj", "model.language_model.layers.42.mlp.experts.53.gate_proj", "model.language_model.layers.42.mlp.experts.54.gate_proj", "model.language_model.layers.42.mlp.experts.55.gate_proj", "model.language_model.layers.42.mlp.experts.56.gate_proj", "model.language_model.layers.42.mlp.experts.57.gate_proj", "model.language_model.layers.42.mlp.experts.58.gate_proj", "model.language_model.layers.42.mlp.experts.59.gate_proj", "model.language_model.layers.42.mlp.experts.60.gate_proj", "model.language_model.layers.42.mlp.experts.61.gate_proj", "model.language_model.layers.42.mlp.experts.62.gate_proj", "model.language_model.layers.42.mlp.experts.63.gate_proj", "model.language_model.layers.42.mlp.experts.64.gate_proj", "model.language_model.layers.42.mlp.experts.65.gate_proj", "model.language_model.layers.42.mlp.experts.66.gate_proj", "model.language_model.layers.42.mlp.experts.67.gate_proj", "model.language_model.layers.42.mlp.experts.68.gate_proj", "model.language_model.layers.42.mlp.experts.69.gate_proj", "model.language_model.layers.42.mlp.experts.70.gate_proj", "model.language_model.layers.42.mlp.experts.71.gate_proj", "model.language_model.layers.42.mlp.experts.72.gate_proj", "model.language_model.layers.42.mlp.experts.73.gate_proj", "model.language_model.layers.42.mlp.experts.74.gate_proj", "model.language_model.layers.42.mlp.experts.75.gate_proj", "model.language_model.layers.42.mlp.experts.76.gate_proj", "model.language_model.layers.42.mlp.experts.77.gate_proj", "model.language_model.layers.42.mlp.experts.78.gate_proj", "model.language_model.layers.42.mlp.experts.79.gate_proj", "model.language_model.layers.42.mlp.experts.80.gate_proj", "model.language_model.layers.42.mlp.experts.81.gate_proj", "model.language_model.layers.42.mlp.experts.82.gate_proj", "model.language_model.layers.42.mlp.experts.83.gate_proj", "model.language_model.layers.42.mlp.experts.84.gate_proj", "model.language_model.layers.42.mlp.experts.85.gate_proj", "model.language_model.layers.42.mlp.experts.86.gate_proj", "model.language_model.layers.42.mlp.experts.87.gate_proj", "model.language_model.layers.42.mlp.experts.88.gate_proj", "model.language_model.layers.42.mlp.experts.89.gate_proj", "model.language_model.layers.42.mlp.experts.90.gate_proj", "model.language_model.layers.42.mlp.experts.91.gate_proj", "model.language_model.layers.42.mlp.experts.92.gate_proj", "model.language_model.layers.42.mlp.experts.93.gate_proj", "model.language_model.layers.42.mlp.experts.94.gate_proj", "model.language_model.layers.42.mlp.experts.95.gate_proj", "model.language_model.layers.42.mlp.experts.96.gate_proj", "model.language_model.layers.42.mlp.experts.97.gate_proj", "model.language_model.layers.42.mlp.experts.98.gate_proj", "model.language_model.layers.42.mlp.experts.99.gate_proj", "model.language_model.layers.42.mlp.experts.100.gate_proj", "model.language_model.layers.42.mlp.experts.101.gate_proj", "model.language_model.layers.42.mlp.experts.102.gate_proj", "model.language_model.layers.42.mlp.experts.103.gate_proj", "model.language_model.layers.42.mlp.experts.104.gate_proj", "model.language_model.layers.42.mlp.experts.105.gate_proj", "model.language_model.layers.42.mlp.experts.106.gate_proj", "model.language_model.layers.42.mlp.experts.107.gate_proj", "model.language_model.layers.42.mlp.experts.108.gate_proj", "model.language_model.layers.42.mlp.experts.109.gate_proj", "model.language_model.layers.42.mlp.experts.110.gate_proj", "model.language_model.layers.42.mlp.experts.111.gate_proj", "model.language_model.layers.42.mlp.experts.112.gate_proj", "model.language_model.layers.42.mlp.experts.113.gate_proj", "model.language_model.layers.42.mlp.experts.114.gate_proj", "model.language_model.layers.42.mlp.experts.115.gate_proj", "model.language_model.layers.42.mlp.experts.116.gate_proj", "model.language_model.layers.42.mlp.experts.117.gate_proj", "model.language_model.layers.42.mlp.experts.118.gate_proj", "model.language_model.layers.42.mlp.experts.119.gate_proj", "model.language_model.layers.42.mlp.experts.120.gate_proj", "model.language_model.layers.42.mlp.experts.121.gate_proj", "model.language_model.layers.42.mlp.experts.122.gate_proj", "model.language_model.layers.42.mlp.experts.123.gate_proj", "model.language_model.layers.42.mlp.experts.124.gate_proj", "model.language_model.layers.42.mlp.experts.125.gate_proj", "model.language_model.layers.42.mlp.experts.126.gate_proj", "model.language_model.layers.42.mlp.experts.127.gate_proj", "model.language_model.layers.42.mlp.experts.0.up_proj", "model.language_model.layers.42.mlp.experts.1.up_proj", "model.language_model.layers.42.mlp.experts.2.up_proj", "model.language_model.layers.42.mlp.experts.3.up_proj", "model.language_model.layers.42.mlp.experts.4.up_proj", "model.language_model.layers.42.mlp.experts.5.up_proj", "model.language_model.layers.42.mlp.experts.6.up_proj", "model.language_model.layers.42.mlp.experts.7.up_proj", "model.language_model.layers.42.mlp.experts.8.up_proj", "model.language_model.layers.42.mlp.experts.9.up_proj", "model.language_model.layers.42.mlp.experts.10.up_proj", "model.language_model.layers.42.mlp.experts.11.up_proj", "model.language_model.layers.42.mlp.experts.12.up_proj", "model.language_model.layers.42.mlp.experts.13.up_proj", "model.language_model.layers.42.mlp.experts.14.up_proj", "model.language_model.layers.42.mlp.experts.15.up_proj", "model.language_model.layers.42.mlp.experts.16.up_proj", "model.language_model.layers.42.mlp.experts.17.up_proj", "model.language_model.layers.42.mlp.experts.18.up_proj", "model.language_model.layers.42.mlp.experts.19.up_proj", "model.language_model.layers.42.mlp.experts.20.up_proj", "model.language_model.layers.42.mlp.experts.21.up_proj", "model.language_model.layers.42.mlp.experts.22.up_proj", "model.language_model.layers.42.mlp.experts.23.up_proj", "model.language_model.layers.42.mlp.experts.24.up_proj", "model.language_model.layers.42.mlp.experts.25.up_proj", "model.language_model.layers.42.mlp.experts.26.up_proj", "model.language_model.layers.42.mlp.experts.27.up_proj", "model.language_model.layers.42.mlp.experts.28.up_proj", "model.language_model.layers.42.mlp.experts.29.up_proj", "model.language_model.layers.42.mlp.experts.30.up_proj", "model.language_model.layers.42.mlp.experts.31.up_proj", "model.language_model.layers.42.mlp.experts.32.up_proj", "model.language_model.layers.42.mlp.experts.33.up_proj", "model.language_model.layers.42.mlp.experts.34.up_proj", "model.language_model.layers.42.mlp.experts.35.up_proj", "model.language_model.layers.42.mlp.experts.36.up_proj", "model.language_model.layers.42.mlp.experts.37.up_proj", "model.language_model.layers.42.mlp.experts.38.up_proj", "model.language_model.layers.42.mlp.experts.39.up_proj", "model.language_model.layers.42.mlp.experts.40.up_proj", "model.language_model.layers.42.mlp.experts.41.up_proj", "model.language_model.layers.42.mlp.experts.42.up_proj", "model.language_model.layers.42.mlp.experts.43.up_proj", "model.language_model.layers.42.mlp.experts.44.up_proj", "model.language_model.layers.42.mlp.experts.45.up_proj", "model.language_model.layers.42.mlp.experts.46.up_proj", "model.language_model.layers.42.mlp.experts.47.up_proj", "model.language_model.layers.42.mlp.experts.48.up_proj", "model.language_model.layers.42.mlp.experts.49.up_proj", "model.language_model.layers.42.mlp.experts.50.up_proj", "model.language_model.layers.42.mlp.experts.51.up_proj", "model.language_model.layers.42.mlp.experts.52.up_proj", "model.language_model.layers.42.mlp.experts.53.up_proj", "model.language_model.layers.42.mlp.experts.54.up_proj", "model.language_model.layers.42.mlp.experts.55.up_proj", "model.language_model.layers.42.mlp.experts.56.up_proj", "model.language_model.layers.42.mlp.experts.57.up_proj", "model.language_model.layers.42.mlp.experts.58.up_proj", "model.language_model.layers.42.mlp.experts.59.up_proj", "model.language_model.layers.42.mlp.experts.60.up_proj", "model.language_model.layers.42.mlp.experts.61.up_proj", "model.language_model.layers.42.mlp.experts.62.up_proj", "model.language_model.layers.42.mlp.experts.63.up_proj", "model.language_model.layers.42.mlp.experts.64.up_proj", "model.language_model.layers.42.mlp.experts.65.up_proj", "model.language_model.layers.42.mlp.experts.66.up_proj", "model.language_model.layers.42.mlp.experts.67.up_proj", "model.language_model.layers.42.mlp.experts.68.up_proj", "model.language_model.layers.42.mlp.experts.69.up_proj", "model.language_model.layers.42.mlp.experts.70.up_proj", "model.language_model.layers.42.mlp.experts.71.up_proj", "model.language_model.layers.42.mlp.experts.72.up_proj", "model.language_model.layers.42.mlp.experts.73.up_proj", "model.language_model.layers.42.mlp.experts.74.up_proj", "model.language_model.layers.42.mlp.experts.75.up_proj", "model.language_model.layers.42.mlp.experts.76.up_proj", "model.language_model.layers.42.mlp.experts.77.up_proj", "model.language_model.layers.42.mlp.experts.78.up_proj", "model.language_model.layers.42.mlp.experts.79.up_proj", "model.language_model.layers.42.mlp.experts.80.up_proj", "model.language_model.layers.42.mlp.experts.81.up_proj", "model.language_model.layers.42.mlp.experts.82.up_proj", "model.language_model.layers.42.mlp.experts.83.up_proj", "model.language_model.layers.42.mlp.experts.84.up_proj", "model.language_model.layers.42.mlp.experts.85.up_proj", "model.language_model.layers.42.mlp.experts.86.up_proj", "model.language_model.layers.42.mlp.experts.87.up_proj", "model.language_model.layers.42.mlp.experts.88.up_proj", "model.language_model.layers.42.mlp.experts.89.up_proj", "model.language_model.layers.42.mlp.experts.90.up_proj", "model.language_model.layers.42.mlp.experts.91.up_proj", "model.language_model.layers.42.mlp.experts.92.up_proj", "model.language_model.layers.42.mlp.experts.93.up_proj", "model.language_model.layers.42.mlp.experts.94.up_proj", "model.language_model.layers.42.mlp.experts.95.up_proj", "model.language_model.layers.42.mlp.experts.96.up_proj", "model.language_model.layers.42.mlp.experts.97.up_proj", "model.language_model.layers.42.mlp.experts.98.up_proj", "model.language_model.layers.42.mlp.experts.99.up_proj", "model.language_model.layers.42.mlp.experts.100.up_proj", "model.language_model.layers.42.mlp.experts.101.up_proj", "model.language_model.layers.42.mlp.experts.102.up_proj", "model.language_model.layers.42.mlp.experts.103.up_proj", "model.language_model.layers.42.mlp.experts.104.up_proj", "model.language_model.layers.42.mlp.experts.105.up_proj", "model.language_model.layers.42.mlp.experts.106.up_proj", "model.language_model.layers.42.mlp.experts.107.up_proj", "model.language_model.layers.42.mlp.experts.108.up_proj", "model.language_model.layers.42.mlp.experts.109.up_proj", "model.language_model.layers.42.mlp.experts.110.up_proj", "model.language_model.layers.42.mlp.experts.111.up_proj", "model.language_model.layers.42.mlp.experts.112.up_proj", "model.language_model.layers.42.mlp.experts.113.up_proj", "model.language_model.layers.42.mlp.experts.114.up_proj", "model.language_model.layers.42.mlp.experts.115.up_proj", "model.language_model.layers.42.mlp.experts.116.up_proj", "model.language_model.layers.42.mlp.experts.117.up_proj", "model.language_model.layers.42.mlp.experts.118.up_proj", "model.language_model.layers.42.mlp.experts.119.up_proj", "model.language_model.layers.42.mlp.experts.120.up_proj", "model.language_model.layers.42.mlp.experts.121.up_proj", "model.language_model.layers.42.mlp.experts.122.up_proj", "model.language_model.layers.42.mlp.experts.123.up_proj", "model.language_model.layers.42.mlp.experts.124.up_proj", "model.language_model.layers.42.mlp.experts.125.up_proj", "model.language_model.layers.42.mlp.experts.126.up_proj", "model.language_model.layers.42.mlp.experts.127.up_proj", "model.language_model.layers.42.mlp.experts.0.down_proj", "model.language_model.layers.42.mlp.experts.1.down_proj", "model.language_model.layers.42.mlp.experts.2.down_proj", "model.language_model.layers.42.mlp.experts.3.down_proj", "model.language_model.layers.42.mlp.experts.4.down_proj", "model.language_model.layers.42.mlp.experts.5.down_proj", "model.language_model.layers.42.mlp.experts.6.down_proj", "model.language_model.layers.42.mlp.experts.7.down_proj", "model.language_model.layers.42.mlp.experts.8.down_proj", "model.language_model.layers.42.mlp.experts.9.down_proj", "model.language_model.layers.42.mlp.experts.10.down_proj", "model.language_model.layers.42.mlp.experts.11.down_proj", "model.language_model.layers.42.mlp.experts.12.down_proj", "model.language_model.layers.42.mlp.experts.13.down_proj", "model.language_model.layers.42.mlp.experts.14.down_proj", "model.language_model.layers.42.mlp.experts.15.down_proj", "model.language_model.layers.42.mlp.experts.16.down_proj", "model.language_model.layers.42.mlp.experts.17.down_proj", "model.language_model.layers.42.mlp.experts.18.down_proj", "model.language_model.layers.42.mlp.experts.19.down_proj", "model.language_model.layers.42.mlp.experts.20.down_proj", "model.language_model.layers.42.mlp.experts.21.down_proj", "model.language_model.layers.42.mlp.experts.22.down_proj", "model.language_model.layers.42.mlp.experts.23.down_proj", "model.language_model.layers.42.mlp.experts.24.down_proj", "model.language_model.layers.42.mlp.experts.25.down_proj", "model.language_model.layers.42.mlp.experts.26.down_proj", "model.language_model.layers.42.mlp.experts.27.down_proj", "model.language_model.layers.42.mlp.experts.28.down_proj", "model.language_model.layers.42.mlp.experts.29.down_proj", "model.language_model.layers.42.mlp.experts.30.down_proj", "model.language_model.layers.42.mlp.experts.31.down_proj", "model.language_model.layers.42.mlp.experts.32.down_proj", "model.language_model.layers.42.mlp.experts.33.down_proj", "model.language_model.layers.42.mlp.experts.34.down_proj", "model.language_model.layers.42.mlp.experts.35.down_proj", "model.language_model.layers.42.mlp.experts.36.down_proj", "model.language_model.layers.42.mlp.experts.37.down_proj", "model.language_model.layers.42.mlp.experts.38.down_proj", "model.language_model.layers.42.mlp.experts.39.down_proj", "model.language_model.layers.42.mlp.experts.40.down_proj", "model.language_model.layers.42.mlp.experts.41.down_proj", "model.language_model.layers.42.mlp.experts.42.down_proj", "model.language_model.layers.42.mlp.experts.43.down_proj", "model.language_model.layers.42.mlp.experts.44.down_proj", "model.language_model.layers.42.mlp.experts.45.down_proj", "model.language_model.layers.42.mlp.experts.46.down_proj", "model.language_model.layers.42.mlp.experts.47.down_proj", "model.language_model.layers.42.mlp.experts.48.down_proj", "model.language_model.layers.42.mlp.experts.49.down_proj", "model.language_model.layers.42.mlp.experts.50.down_proj", "model.language_model.layers.42.mlp.experts.51.down_proj", "model.language_model.layers.42.mlp.experts.52.down_proj", "model.language_model.layers.42.mlp.experts.53.down_proj", "model.language_model.layers.42.mlp.experts.54.down_proj", "model.language_model.layers.42.mlp.experts.55.down_proj", "model.language_model.layers.42.mlp.experts.56.down_proj", "model.language_model.layers.42.mlp.experts.57.down_proj", "model.language_model.layers.42.mlp.experts.58.down_proj", "model.language_model.layers.42.mlp.experts.59.down_proj", "model.language_model.layers.42.mlp.experts.60.down_proj", "model.language_model.layers.42.mlp.experts.61.down_proj", "model.language_model.layers.42.mlp.experts.62.down_proj", "model.language_model.layers.42.mlp.experts.63.down_proj", "model.language_model.layers.42.mlp.experts.64.down_proj", "model.language_model.layers.42.mlp.experts.65.down_proj", "model.language_model.layers.42.mlp.experts.66.down_proj", "model.language_model.layers.42.mlp.experts.67.down_proj", "model.language_model.layers.42.mlp.experts.68.down_proj", "model.language_model.layers.42.mlp.experts.69.down_proj", "model.language_model.layers.42.mlp.experts.70.down_proj", "model.language_model.layers.42.mlp.experts.71.down_proj", "model.language_model.layers.42.mlp.experts.72.down_proj", "model.language_model.layers.42.mlp.experts.73.down_proj", "model.language_model.layers.42.mlp.experts.74.down_proj", "model.language_model.layers.42.mlp.experts.75.down_proj", "model.language_model.layers.42.mlp.experts.76.down_proj", "model.language_model.layers.42.mlp.experts.77.down_proj", "model.language_model.layers.42.mlp.experts.78.down_proj", "model.language_model.layers.42.mlp.experts.79.down_proj", "model.language_model.layers.42.mlp.experts.80.down_proj", "model.language_model.layers.42.mlp.experts.81.down_proj", "model.language_model.layers.42.mlp.experts.82.down_proj", "model.language_model.layers.42.mlp.experts.83.down_proj", "model.language_model.layers.42.mlp.experts.84.down_proj", "model.language_model.layers.42.mlp.experts.85.down_proj", "model.language_model.layers.42.mlp.experts.86.down_proj", "model.language_model.layers.42.mlp.experts.87.down_proj", "model.language_model.layers.42.mlp.experts.88.down_proj", "model.language_model.layers.42.mlp.experts.89.down_proj", "model.language_model.layers.42.mlp.experts.90.down_proj", "model.language_model.layers.42.mlp.experts.91.down_proj", "model.language_model.layers.42.mlp.experts.92.down_proj", "model.language_model.layers.42.mlp.experts.93.down_proj", "model.language_model.layers.42.mlp.experts.94.down_proj", "model.language_model.layers.42.mlp.experts.95.down_proj", "model.language_model.layers.42.mlp.experts.96.down_proj", "model.language_model.layers.42.mlp.experts.97.down_proj", "model.language_model.layers.42.mlp.experts.98.down_proj", "model.language_model.layers.42.mlp.experts.99.down_proj", "model.language_model.layers.42.mlp.experts.100.down_proj", "model.language_model.layers.42.mlp.experts.101.down_proj", "model.language_model.layers.42.mlp.experts.102.down_proj", "model.language_model.layers.42.mlp.experts.103.down_proj", "model.language_model.layers.42.mlp.experts.104.down_proj", "model.language_model.layers.42.mlp.experts.105.down_proj", "model.language_model.layers.42.mlp.experts.106.down_proj", "model.language_model.layers.42.mlp.experts.107.down_proj", "model.language_model.layers.42.mlp.experts.108.down_proj", "model.language_model.layers.42.mlp.experts.109.down_proj", "model.language_model.layers.42.mlp.experts.110.down_proj", "model.language_model.layers.42.mlp.experts.111.down_proj", "model.language_model.layers.42.mlp.experts.112.down_proj", "model.language_model.layers.42.mlp.experts.113.down_proj", "model.language_model.layers.42.mlp.experts.114.down_proj", "model.language_model.layers.42.mlp.experts.115.down_proj", "model.language_model.layers.42.mlp.experts.116.down_proj", "model.language_model.layers.42.mlp.experts.117.down_proj", "model.language_model.layers.42.mlp.experts.118.down_proj", "model.language_model.layers.42.mlp.experts.119.down_proj", "model.language_model.layers.42.mlp.experts.120.down_proj", "model.language_model.layers.42.mlp.experts.121.down_proj", "model.language_model.layers.42.mlp.experts.122.down_proj", "model.language_model.layers.42.mlp.experts.123.down_proj", "model.language_model.layers.42.mlp.experts.124.down_proj", "model.language_model.layers.42.mlp.experts.125.down_proj", "model.language_model.layers.42.mlp.experts.126.down_proj", "model.language_model.layers.42.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.0012773756869137315, "dbits": 603979776 } ] }, { "idx": 86, "layers": [ "model.language_model.layers.43.self_attn.q_proj", "model.language_model.layers.43.self_attn.k_proj", "model.language_model.layers.43.self_attn.v_proj", "model.language_model.layers.43.self_attn.o_proj" ], "candidates": [ { "dkld": -0.001451586838811636, "dbits": 18874368 } ] }, { "idx": 87, "layers": [ "model.language_model.layers.43.mlp.experts.0.gate_proj", "model.language_model.layers.43.mlp.experts.1.gate_proj", "model.language_model.layers.43.mlp.experts.2.gate_proj", "model.language_model.layers.43.mlp.experts.3.gate_proj", "model.language_model.layers.43.mlp.experts.4.gate_proj", "model.language_model.layers.43.mlp.experts.5.gate_proj", "model.language_model.layers.43.mlp.experts.6.gate_proj", "model.language_model.layers.43.mlp.experts.7.gate_proj", "model.language_model.layers.43.mlp.experts.8.gate_proj", "model.language_model.layers.43.mlp.experts.9.gate_proj", "model.language_model.layers.43.mlp.experts.10.gate_proj", "model.language_model.layers.43.mlp.experts.11.gate_proj", "model.language_model.layers.43.mlp.experts.12.gate_proj", "model.language_model.layers.43.mlp.experts.13.gate_proj", "model.language_model.layers.43.mlp.experts.14.gate_proj", "model.language_model.layers.43.mlp.experts.15.gate_proj", "model.language_model.layers.43.mlp.experts.16.gate_proj", "model.language_model.layers.43.mlp.experts.17.gate_proj", "model.language_model.layers.43.mlp.experts.18.gate_proj", "model.language_model.layers.43.mlp.experts.19.gate_proj", "model.language_model.layers.43.mlp.experts.20.gate_proj", "model.language_model.layers.43.mlp.experts.21.gate_proj", "model.language_model.layers.43.mlp.experts.22.gate_proj", "model.language_model.layers.43.mlp.experts.23.gate_proj", "model.language_model.layers.43.mlp.experts.24.gate_proj", "model.language_model.layers.43.mlp.experts.25.gate_proj", "model.language_model.layers.43.mlp.experts.26.gate_proj", "model.language_model.layers.43.mlp.experts.27.gate_proj", "model.language_model.layers.43.mlp.experts.28.gate_proj", "model.language_model.layers.43.mlp.experts.29.gate_proj", "model.language_model.layers.43.mlp.experts.30.gate_proj", "model.language_model.layers.43.mlp.experts.31.gate_proj", "model.language_model.layers.43.mlp.experts.32.gate_proj", "model.language_model.layers.43.mlp.experts.33.gate_proj", "model.language_model.layers.43.mlp.experts.34.gate_proj", "model.language_model.layers.43.mlp.experts.35.gate_proj", "model.language_model.layers.43.mlp.experts.36.gate_proj", "model.language_model.layers.43.mlp.experts.37.gate_proj", "model.language_model.layers.43.mlp.experts.38.gate_proj", "model.language_model.layers.43.mlp.experts.39.gate_proj", "model.language_model.layers.43.mlp.experts.40.gate_proj", "model.language_model.layers.43.mlp.experts.41.gate_proj", "model.language_model.layers.43.mlp.experts.42.gate_proj", "model.language_model.layers.43.mlp.experts.43.gate_proj", "model.language_model.layers.43.mlp.experts.44.gate_proj", "model.language_model.layers.43.mlp.experts.45.gate_proj", "model.language_model.layers.43.mlp.experts.46.gate_proj", "model.language_model.layers.43.mlp.experts.47.gate_proj", "model.language_model.layers.43.mlp.experts.48.gate_proj", "model.language_model.layers.43.mlp.experts.49.gate_proj", "model.language_model.layers.43.mlp.experts.50.gate_proj", "model.language_model.layers.43.mlp.experts.51.gate_proj", "model.language_model.layers.43.mlp.experts.52.gate_proj", "model.language_model.layers.43.mlp.experts.53.gate_proj", "model.language_model.layers.43.mlp.experts.54.gate_proj", "model.language_model.layers.43.mlp.experts.55.gate_proj", "model.language_model.layers.43.mlp.experts.56.gate_proj", "model.language_model.layers.43.mlp.experts.57.gate_proj", "model.language_model.layers.43.mlp.experts.58.gate_proj", "model.language_model.layers.43.mlp.experts.59.gate_proj", "model.language_model.layers.43.mlp.experts.60.gate_proj", "model.language_model.layers.43.mlp.experts.61.gate_proj", "model.language_model.layers.43.mlp.experts.62.gate_proj", "model.language_model.layers.43.mlp.experts.63.gate_proj", "model.language_model.layers.43.mlp.experts.64.gate_proj", "model.language_model.layers.43.mlp.experts.65.gate_proj", "model.language_model.layers.43.mlp.experts.66.gate_proj", "model.language_model.layers.43.mlp.experts.67.gate_proj", "model.language_model.layers.43.mlp.experts.68.gate_proj", "model.language_model.layers.43.mlp.experts.69.gate_proj", "model.language_model.layers.43.mlp.experts.70.gate_proj", "model.language_model.layers.43.mlp.experts.71.gate_proj", "model.language_model.layers.43.mlp.experts.72.gate_proj", "model.language_model.layers.43.mlp.experts.73.gate_proj", "model.language_model.layers.43.mlp.experts.74.gate_proj", "model.language_model.layers.43.mlp.experts.75.gate_proj", "model.language_model.layers.43.mlp.experts.76.gate_proj", "model.language_model.layers.43.mlp.experts.77.gate_proj", "model.language_model.layers.43.mlp.experts.78.gate_proj", "model.language_model.layers.43.mlp.experts.79.gate_proj", "model.language_model.layers.43.mlp.experts.80.gate_proj", "model.language_model.layers.43.mlp.experts.81.gate_proj", "model.language_model.layers.43.mlp.experts.82.gate_proj", "model.language_model.layers.43.mlp.experts.83.gate_proj", "model.language_model.layers.43.mlp.experts.84.gate_proj", "model.language_model.layers.43.mlp.experts.85.gate_proj", "model.language_model.layers.43.mlp.experts.86.gate_proj", "model.language_model.layers.43.mlp.experts.87.gate_proj", "model.language_model.layers.43.mlp.experts.88.gate_proj", "model.language_model.layers.43.mlp.experts.89.gate_proj", "model.language_model.layers.43.mlp.experts.90.gate_proj", "model.language_model.layers.43.mlp.experts.91.gate_proj", "model.language_model.layers.43.mlp.experts.92.gate_proj", "model.language_model.layers.43.mlp.experts.93.gate_proj", "model.language_model.layers.43.mlp.experts.94.gate_proj", "model.language_model.layers.43.mlp.experts.95.gate_proj", "model.language_model.layers.43.mlp.experts.96.gate_proj", "model.language_model.layers.43.mlp.experts.97.gate_proj", "model.language_model.layers.43.mlp.experts.98.gate_proj", "model.language_model.layers.43.mlp.experts.99.gate_proj", "model.language_model.layers.43.mlp.experts.100.gate_proj", "model.language_model.layers.43.mlp.experts.101.gate_proj", "model.language_model.layers.43.mlp.experts.102.gate_proj", "model.language_model.layers.43.mlp.experts.103.gate_proj", "model.language_model.layers.43.mlp.experts.104.gate_proj", "model.language_model.layers.43.mlp.experts.105.gate_proj", "model.language_model.layers.43.mlp.experts.106.gate_proj", "model.language_model.layers.43.mlp.experts.107.gate_proj", "model.language_model.layers.43.mlp.experts.108.gate_proj", "model.language_model.layers.43.mlp.experts.109.gate_proj", "model.language_model.layers.43.mlp.experts.110.gate_proj", "model.language_model.layers.43.mlp.experts.111.gate_proj", "model.language_model.layers.43.mlp.experts.112.gate_proj", "model.language_model.layers.43.mlp.experts.113.gate_proj", "model.language_model.layers.43.mlp.experts.114.gate_proj", "model.language_model.layers.43.mlp.experts.115.gate_proj", "model.language_model.layers.43.mlp.experts.116.gate_proj", "model.language_model.layers.43.mlp.experts.117.gate_proj", "model.language_model.layers.43.mlp.experts.118.gate_proj", "model.language_model.layers.43.mlp.experts.119.gate_proj", "model.language_model.layers.43.mlp.experts.120.gate_proj", "model.language_model.layers.43.mlp.experts.121.gate_proj", "model.language_model.layers.43.mlp.experts.122.gate_proj", "model.language_model.layers.43.mlp.experts.123.gate_proj", "model.language_model.layers.43.mlp.experts.124.gate_proj", "model.language_model.layers.43.mlp.experts.125.gate_proj", "model.language_model.layers.43.mlp.experts.126.gate_proj", "model.language_model.layers.43.mlp.experts.127.gate_proj", "model.language_model.layers.43.mlp.experts.0.up_proj", "model.language_model.layers.43.mlp.experts.1.up_proj", "model.language_model.layers.43.mlp.experts.2.up_proj", "model.language_model.layers.43.mlp.experts.3.up_proj", "model.language_model.layers.43.mlp.experts.4.up_proj", "model.language_model.layers.43.mlp.experts.5.up_proj", "model.language_model.layers.43.mlp.experts.6.up_proj", "model.language_model.layers.43.mlp.experts.7.up_proj", "model.language_model.layers.43.mlp.experts.8.up_proj", "model.language_model.layers.43.mlp.experts.9.up_proj", "model.language_model.layers.43.mlp.experts.10.up_proj", "model.language_model.layers.43.mlp.experts.11.up_proj", "model.language_model.layers.43.mlp.experts.12.up_proj", "model.language_model.layers.43.mlp.experts.13.up_proj", "model.language_model.layers.43.mlp.experts.14.up_proj", "model.language_model.layers.43.mlp.experts.15.up_proj", "model.language_model.layers.43.mlp.experts.16.up_proj", "model.language_model.layers.43.mlp.experts.17.up_proj", "model.language_model.layers.43.mlp.experts.18.up_proj", "model.language_model.layers.43.mlp.experts.19.up_proj", "model.language_model.layers.43.mlp.experts.20.up_proj", "model.language_model.layers.43.mlp.experts.21.up_proj", "model.language_model.layers.43.mlp.experts.22.up_proj", "model.language_model.layers.43.mlp.experts.23.up_proj", "model.language_model.layers.43.mlp.experts.24.up_proj", "model.language_model.layers.43.mlp.experts.25.up_proj", "model.language_model.layers.43.mlp.experts.26.up_proj", "model.language_model.layers.43.mlp.experts.27.up_proj", "model.language_model.layers.43.mlp.experts.28.up_proj", "model.language_model.layers.43.mlp.experts.29.up_proj", "model.language_model.layers.43.mlp.experts.30.up_proj", "model.language_model.layers.43.mlp.experts.31.up_proj", "model.language_model.layers.43.mlp.experts.32.up_proj", "model.language_model.layers.43.mlp.experts.33.up_proj", "model.language_model.layers.43.mlp.experts.34.up_proj", "model.language_model.layers.43.mlp.experts.35.up_proj", "model.language_model.layers.43.mlp.experts.36.up_proj", "model.language_model.layers.43.mlp.experts.37.up_proj", "model.language_model.layers.43.mlp.experts.38.up_proj", "model.language_model.layers.43.mlp.experts.39.up_proj", "model.language_model.layers.43.mlp.experts.40.up_proj", "model.language_model.layers.43.mlp.experts.41.up_proj", "model.language_model.layers.43.mlp.experts.42.up_proj", "model.language_model.layers.43.mlp.experts.43.up_proj", "model.language_model.layers.43.mlp.experts.44.up_proj", "model.language_model.layers.43.mlp.experts.45.up_proj", "model.language_model.layers.43.mlp.experts.46.up_proj", "model.language_model.layers.43.mlp.experts.47.up_proj", "model.language_model.layers.43.mlp.experts.48.up_proj", "model.language_model.layers.43.mlp.experts.49.up_proj", "model.language_model.layers.43.mlp.experts.50.up_proj", "model.language_model.layers.43.mlp.experts.51.up_proj", "model.language_model.layers.43.mlp.experts.52.up_proj", "model.language_model.layers.43.mlp.experts.53.up_proj", "model.language_model.layers.43.mlp.experts.54.up_proj", "model.language_model.layers.43.mlp.experts.55.up_proj", "model.language_model.layers.43.mlp.experts.56.up_proj", "model.language_model.layers.43.mlp.experts.57.up_proj", "model.language_model.layers.43.mlp.experts.58.up_proj", "model.language_model.layers.43.mlp.experts.59.up_proj", "model.language_model.layers.43.mlp.experts.60.up_proj", "model.language_model.layers.43.mlp.experts.61.up_proj", "model.language_model.layers.43.mlp.experts.62.up_proj", "model.language_model.layers.43.mlp.experts.63.up_proj", "model.language_model.layers.43.mlp.experts.64.up_proj", "model.language_model.layers.43.mlp.experts.65.up_proj", "model.language_model.layers.43.mlp.experts.66.up_proj", "model.language_model.layers.43.mlp.experts.67.up_proj", "model.language_model.layers.43.mlp.experts.68.up_proj", "model.language_model.layers.43.mlp.experts.69.up_proj", "model.language_model.layers.43.mlp.experts.70.up_proj", "model.language_model.layers.43.mlp.experts.71.up_proj", "model.language_model.layers.43.mlp.experts.72.up_proj", "model.language_model.layers.43.mlp.experts.73.up_proj", "model.language_model.layers.43.mlp.experts.74.up_proj", "model.language_model.layers.43.mlp.experts.75.up_proj", "model.language_model.layers.43.mlp.experts.76.up_proj", "model.language_model.layers.43.mlp.experts.77.up_proj", "model.language_model.layers.43.mlp.experts.78.up_proj", "model.language_model.layers.43.mlp.experts.79.up_proj", "model.language_model.layers.43.mlp.experts.80.up_proj", "model.language_model.layers.43.mlp.experts.81.up_proj", "model.language_model.layers.43.mlp.experts.82.up_proj", "model.language_model.layers.43.mlp.experts.83.up_proj", "model.language_model.layers.43.mlp.experts.84.up_proj", "model.language_model.layers.43.mlp.experts.85.up_proj", "model.language_model.layers.43.mlp.experts.86.up_proj", "model.language_model.layers.43.mlp.experts.87.up_proj", "model.language_model.layers.43.mlp.experts.88.up_proj", "model.language_model.layers.43.mlp.experts.89.up_proj", "model.language_model.layers.43.mlp.experts.90.up_proj", "model.language_model.layers.43.mlp.experts.91.up_proj", "model.language_model.layers.43.mlp.experts.92.up_proj", "model.language_model.layers.43.mlp.experts.93.up_proj", "model.language_model.layers.43.mlp.experts.94.up_proj", "model.language_model.layers.43.mlp.experts.95.up_proj", "model.language_model.layers.43.mlp.experts.96.up_proj", "model.language_model.layers.43.mlp.experts.97.up_proj", "model.language_model.layers.43.mlp.experts.98.up_proj", "model.language_model.layers.43.mlp.experts.99.up_proj", "model.language_model.layers.43.mlp.experts.100.up_proj", "model.language_model.layers.43.mlp.experts.101.up_proj", "model.language_model.layers.43.mlp.experts.102.up_proj", "model.language_model.layers.43.mlp.experts.103.up_proj", "model.language_model.layers.43.mlp.experts.104.up_proj", "model.language_model.layers.43.mlp.experts.105.up_proj", "model.language_model.layers.43.mlp.experts.106.up_proj", "model.language_model.layers.43.mlp.experts.107.up_proj", "model.language_model.layers.43.mlp.experts.108.up_proj", "model.language_model.layers.43.mlp.experts.109.up_proj", "model.language_model.layers.43.mlp.experts.110.up_proj", "model.language_model.layers.43.mlp.experts.111.up_proj", "model.language_model.layers.43.mlp.experts.112.up_proj", "model.language_model.layers.43.mlp.experts.113.up_proj", "model.language_model.layers.43.mlp.experts.114.up_proj", "model.language_model.layers.43.mlp.experts.115.up_proj", "model.language_model.layers.43.mlp.experts.116.up_proj", "model.language_model.layers.43.mlp.experts.117.up_proj", "model.language_model.layers.43.mlp.experts.118.up_proj", "model.language_model.layers.43.mlp.experts.119.up_proj", "model.language_model.layers.43.mlp.experts.120.up_proj", "model.language_model.layers.43.mlp.experts.121.up_proj", "model.language_model.layers.43.mlp.experts.122.up_proj", "model.language_model.layers.43.mlp.experts.123.up_proj", "model.language_model.layers.43.mlp.experts.124.up_proj", "model.language_model.layers.43.mlp.experts.125.up_proj", "model.language_model.layers.43.mlp.experts.126.up_proj", "model.language_model.layers.43.mlp.experts.127.up_proj", "model.language_model.layers.43.mlp.experts.0.down_proj", "model.language_model.layers.43.mlp.experts.1.down_proj", "model.language_model.layers.43.mlp.experts.2.down_proj", "model.language_model.layers.43.mlp.experts.3.down_proj", "model.language_model.layers.43.mlp.experts.4.down_proj", "model.language_model.layers.43.mlp.experts.5.down_proj", "model.language_model.layers.43.mlp.experts.6.down_proj", "model.language_model.layers.43.mlp.experts.7.down_proj", "model.language_model.layers.43.mlp.experts.8.down_proj", "model.language_model.layers.43.mlp.experts.9.down_proj", "model.language_model.layers.43.mlp.experts.10.down_proj", "model.language_model.layers.43.mlp.experts.11.down_proj", "model.language_model.layers.43.mlp.experts.12.down_proj", "model.language_model.layers.43.mlp.experts.13.down_proj", "model.language_model.layers.43.mlp.experts.14.down_proj", "model.language_model.layers.43.mlp.experts.15.down_proj", "model.language_model.layers.43.mlp.experts.16.down_proj", "model.language_model.layers.43.mlp.experts.17.down_proj", "model.language_model.layers.43.mlp.experts.18.down_proj", "model.language_model.layers.43.mlp.experts.19.down_proj", "model.language_model.layers.43.mlp.experts.20.down_proj", "model.language_model.layers.43.mlp.experts.21.down_proj", "model.language_model.layers.43.mlp.experts.22.down_proj", "model.language_model.layers.43.mlp.experts.23.down_proj", "model.language_model.layers.43.mlp.experts.24.down_proj", "model.language_model.layers.43.mlp.experts.25.down_proj", "model.language_model.layers.43.mlp.experts.26.down_proj", "model.language_model.layers.43.mlp.experts.27.down_proj", "model.language_model.layers.43.mlp.experts.28.down_proj", "model.language_model.layers.43.mlp.experts.29.down_proj", "model.language_model.layers.43.mlp.experts.30.down_proj", "model.language_model.layers.43.mlp.experts.31.down_proj", "model.language_model.layers.43.mlp.experts.32.down_proj", "model.language_model.layers.43.mlp.experts.33.down_proj", "model.language_model.layers.43.mlp.experts.34.down_proj", "model.language_model.layers.43.mlp.experts.35.down_proj", "model.language_model.layers.43.mlp.experts.36.down_proj", "model.language_model.layers.43.mlp.experts.37.down_proj", "model.language_model.layers.43.mlp.experts.38.down_proj", "model.language_model.layers.43.mlp.experts.39.down_proj", "model.language_model.layers.43.mlp.experts.40.down_proj", "model.language_model.layers.43.mlp.experts.41.down_proj", "model.language_model.layers.43.mlp.experts.42.down_proj", "model.language_model.layers.43.mlp.experts.43.down_proj", "model.language_model.layers.43.mlp.experts.44.down_proj", "model.language_model.layers.43.mlp.experts.45.down_proj", "model.language_model.layers.43.mlp.experts.46.down_proj", "model.language_model.layers.43.mlp.experts.47.down_proj", "model.language_model.layers.43.mlp.experts.48.down_proj", "model.language_model.layers.43.mlp.experts.49.down_proj", "model.language_model.layers.43.mlp.experts.50.down_proj", "model.language_model.layers.43.mlp.experts.51.down_proj", "model.language_model.layers.43.mlp.experts.52.down_proj", "model.language_model.layers.43.mlp.experts.53.down_proj", "model.language_model.layers.43.mlp.experts.54.down_proj", "model.language_model.layers.43.mlp.experts.55.down_proj", "model.language_model.layers.43.mlp.experts.56.down_proj", "model.language_model.layers.43.mlp.experts.57.down_proj", "model.language_model.layers.43.mlp.experts.58.down_proj", "model.language_model.layers.43.mlp.experts.59.down_proj", "model.language_model.layers.43.mlp.experts.60.down_proj", "model.language_model.layers.43.mlp.experts.61.down_proj", "model.language_model.layers.43.mlp.experts.62.down_proj", "model.language_model.layers.43.mlp.experts.63.down_proj", "model.language_model.layers.43.mlp.experts.64.down_proj", "model.language_model.layers.43.mlp.experts.65.down_proj", "model.language_model.layers.43.mlp.experts.66.down_proj", "model.language_model.layers.43.mlp.experts.67.down_proj", "model.language_model.layers.43.mlp.experts.68.down_proj", "model.language_model.layers.43.mlp.experts.69.down_proj", "model.language_model.layers.43.mlp.experts.70.down_proj", "model.language_model.layers.43.mlp.experts.71.down_proj", "model.language_model.layers.43.mlp.experts.72.down_proj", "model.language_model.layers.43.mlp.experts.73.down_proj", "model.language_model.layers.43.mlp.experts.74.down_proj", "model.language_model.layers.43.mlp.experts.75.down_proj", "model.language_model.layers.43.mlp.experts.76.down_proj", "model.language_model.layers.43.mlp.experts.77.down_proj", "model.language_model.layers.43.mlp.experts.78.down_proj", "model.language_model.layers.43.mlp.experts.79.down_proj", "model.language_model.layers.43.mlp.experts.80.down_proj", "model.language_model.layers.43.mlp.experts.81.down_proj", "model.language_model.layers.43.mlp.experts.82.down_proj", "model.language_model.layers.43.mlp.experts.83.down_proj", "model.language_model.layers.43.mlp.experts.84.down_proj", "model.language_model.layers.43.mlp.experts.85.down_proj", "model.language_model.layers.43.mlp.experts.86.down_proj", "model.language_model.layers.43.mlp.experts.87.down_proj", "model.language_model.layers.43.mlp.experts.88.down_proj", "model.language_model.layers.43.mlp.experts.89.down_proj", "model.language_model.layers.43.mlp.experts.90.down_proj", "model.language_model.layers.43.mlp.experts.91.down_proj", "model.language_model.layers.43.mlp.experts.92.down_proj", "model.language_model.layers.43.mlp.experts.93.down_proj", "model.language_model.layers.43.mlp.experts.94.down_proj", "model.language_model.layers.43.mlp.experts.95.down_proj", "model.language_model.layers.43.mlp.experts.96.down_proj", "model.language_model.layers.43.mlp.experts.97.down_proj", "model.language_model.layers.43.mlp.experts.98.down_proj", "model.language_model.layers.43.mlp.experts.99.down_proj", "model.language_model.layers.43.mlp.experts.100.down_proj", "model.language_model.layers.43.mlp.experts.101.down_proj", "model.language_model.layers.43.mlp.experts.102.down_proj", "model.language_model.layers.43.mlp.experts.103.down_proj", "model.language_model.layers.43.mlp.experts.104.down_proj", "model.language_model.layers.43.mlp.experts.105.down_proj", "model.language_model.layers.43.mlp.experts.106.down_proj", "model.language_model.layers.43.mlp.experts.107.down_proj", "model.language_model.layers.43.mlp.experts.108.down_proj", "model.language_model.layers.43.mlp.experts.109.down_proj", "model.language_model.layers.43.mlp.experts.110.down_proj", "model.language_model.layers.43.mlp.experts.111.down_proj", "model.language_model.layers.43.mlp.experts.112.down_proj", "model.language_model.layers.43.mlp.experts.113.down_proj", "model.language_model.layers.43.mlp.experts.114.down_proj", "model.language_model.layers.43.mlp.experts.115.down_proj", "model.language_model.layers.43.mlp.experts.116.down_proj", "model.language_model.layers.43.mlp.experts.117.down_proj", "model.language_model.layers.43.mlp.experts.118.down_proj", "model.language_model.layers.43.mlp.experts.119.down_proj", "model.language_model.layers.43.mlp.experts.120.down_proj", "model.language_model.layers.43.mlp.experts.121.down_proj", "model.language_model.layers.43.mlp.experts.122.down_proj", "model.language_model.layers.43.mlp.experts.123.down_proj", "model.language_model.layers.43.mlp.experts.124.down_proj", "model.language_model.layers.43.mlp.experts.125.down_proj", "model.language_model.layers.43.mlp.experts.126.down_proj", "model.language_model.layers.43.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.0007468592375516891, "dbits": 603979776 } ] }, { "idx": 88, "layers": [ "model.language_model.layers.44.self_attn.q_proj", "model.language_model.layers.44.self_attn.k_proj", "model.language_model.layers.44.self_attn.v_proj", "model.language_model.layers.44.self_attn.o_proj" ], "candidates": [ { "dkld": -9.326683357357979e-05, "dbits": 18874368 } ] }, { "idx": 89, "layers": [ "model.language_model.layers.44.mlp.experts.0.gate_proj", "model.language_model.layers.44.mlp.experts.1.gate_proj", "model.language_model.layers.44.mlp.experts.2.gate_proj", "model.language_model.layers.44.mlp.experts.3.gate_proj", "model.language_model.layers.44.mlp.experts.4.gate_proj", "model.language_model.layers.44.mlp.experts.5.gate_proj", "model.language_model.layers.44.mlp.experts.6.gate_proj", "model.language_model.layers.44.mlp.experts.7.gate_proj", "model.language_model.layers.44.mlp.experts.8.gate_proj", "model.language_model.layers.44.mlp.experts.9.gate_proj", "model.language_model.layers.44.mlp.experts.10.gate_proj", "model.language_model.layers.44.mlp.experts.11.gate_proj", "model.language_model.layers.44.mlp.experts.12.gate_proj", "model.language_model.layers.44.mlp.experts.13.gate_proj", "model.language_model.layers.44.mlp.experts.14.gate_proj", "model.language_model.layers.44.mlp.experts.15.gate_proj", "model.language_model.layers.44.mlp.experts.16.gate_proj", "model.language_model.layers.44.mlp.experts.17.gate_proj", "model.language_model.layers.44.mlp.experts.18.gate_proj", "model.language_model.layers.44.mlp.experts.19.gate_proj", "model.language_model.layers.44.mlp.experts.20.gate_proj", "model.language_model.layers.44.mlp.experts.21.gate_proj", "model.language_model.layers.44.mlp.experts.22.gate_proj", "model.language_model.layers.44.mlp.experts.23.gate_proj", "model.language_model.layers.44.mlp.experts.24.gate_proj", "model.language_model.layers.44.mlp.experts.25.gate_proj", "model.language_model.layers.44.mlp.experts.26.gate_proj", "model.language_model.layers.44.mlp.experts.27.gate_proj", "model.language_model.layers.44.mlp.experts.28.gate_proj", "model.language_model.layers.44.mlp.experts.29.gate_proj", "model.language_model.layers.44.mlp.experts.30.gate_proj", "model.language_model.layers.44.mlp.experts.31.gate_proj", "model.language_model.layers.44.mlp.experts.32.gate_proj", "model.language_model.layers.44.mlp.experts.33.gate_proj", "model.language_model.layers.44.mlp.experts.34.gate_proj", "model.language_model.layers.44.mlp.experts.35.gate_proj", "model.language_model.layers.44.mlp.experts.36.gate_proj", "model.language_model.layers.44.mlp.experts.37.gate_proj", "model.language_model.layers.44.mlp.experts.38.gate_proj", "model.language_model.layers.44.mlp.experts.39.gate_proj", "model.language_model.layers.44.mlp.experts.40.gate_proj", "model.language_model.layers.44.mlp.experts.41.gate_proj", "model.language_model.layers.44.mlp.experts.42.gate_proj", "model.language_model.layers.44.mlp.experts.43.gate_proj", "model.language_model.layers.44.mlp.experts.44.gate_proj", "model.language_model.layers.44.mlp.experts.45.gate_proj", "model.language_model.layers.44.mlp.experts.46.gate_proj", "model.language_model.layers.44.mlp.experts.47.gate_proj", "model.language_model.layers.44.mlp.experts.48.gate_proj", "model.language_model.layers.44.mlp.experts.49.gate_proj", "model.language_model.layers.44.mlp.experts.50.gate_proj", "model.language_model.layers.44.mlp.experts.51.gate_proj", "model.language_model.layers.44.mlp.experts.52.gate_proj", "model.language_model.layers.44.mlp.experts.53.gate_proj", "model.language_model.layers.44.mlp.experts.54.gate_proj", "model.language_model.layers.44.mlp.experts.55.gate_proj", "model.language_model.layers.44.mlp.experts.56.gate_proj", "model.language_model.layers.44.mlp.experts.57.gate_proj", "model.language_model.layers.44.mlp.experts.58.gate_proj", "model.language_model.layers.44.mlp.experts.59.gate_proj", "model.language_model.layers.44.mlp.experts.60.gate_proj", "model.language_model.layers.44.mlp.experts.61.gate_proj", "model.language_model.layers.44.mlp.experts.62.gate_proj", "model.language_model.layers.44.mlp.experts.63.gate_proj", "model.language_model.layers.44.mlp.experts.64.gate_proj", "model.language_model.layers.44.mlp.experts.65.gate_proj", "model.language_model.layers.44.mlp.experts.66.gate_proj", "model.language_model.layers.44.mlp.experts.67.gate_proj", "model.language_model.layers.44.mlp.experts.68.gate_proj", "model.language_model.layers.44.mlp.experts.69.gate_proj", "model.language_model.layers.44.mlp.experts.70.gate_proj", "model.language_model.layers.44.mlp.experts.71.gate_proj", "model.language_model.layers.44.mlp.experts.72.gate_proj", "model.language_model.layers.44.mlp.experts.73.gate_proj", "model.language_model.layers.44.mlp.experts.74.gate_proj", "model.language_model.layers.44.mlp.experts.75.gate_proj", "model.language_model.layers.44.mlp.experts.76.gate_proj", "model.language_model.layers.44.mlp.experts.77.gate_proj", "model.language_model.layers.44.mlp.experts.78.gate_proj", "model.language_model.layers.44.mlp.experts.79.gate_proj", "model.language_model.layers.44.mlp.experts.80.gate_proj", "model.language_model.layers.44.mlp.experts.81.gate_proj", "model.language_model.layers.44.mlp.experts.82.gate_proj", "model.language_model.layers.44.mlp.experts.83.gate_proj", "model.language_model.layers.44.mlp.experts.84.gate_proj", "model.language_model.layers.44.mlp.experts.85.gate_proj", "model.language_model.layers.44.mlp.experts.86.gate_proj", "model.language_model.layers.44.mlp.experts.87.gate_proj", "model.language_model.layers.44.mlp.experts.88.gate_proj", "model.language_model.layers.44.mlp.experts.89.gate_proj", "model.language_model.layers.44.mlp.experts.90.gate_proj", "model.language_model.layers.44.mlp.experts.91.gate_proj", "model.language_model.layers.44.mlp.experts.92.gate_proj", "model.language_model.layers.44.mlp.experts.93.gate_proj", "model.language_model.layers.44.mlp.experts.94.gate_proj", "model.language_model.layers.44.mlp.experts.95.gate_proj", "model.language_model.layers.44.mlp.experts.96.gate_proj", "model.language_model.layers.44.mlp.experts.97.gate_proj", "model.language_model.layers.44.mlp.experts.98.gate_proj", "model.language_model.layers.44.mlp.experts.99.gate_proj", "model.language_model.layers.44.mlp.experts.100.gate_proj", "model.language_model.layers.44.mlp.experts.101.gate_proj", "model.language_model.layers.44.mlp.experts.102.gate_proj", "model.language_model.layers.44.mlp.experts.103.gate_proj", "model.language_model.layers.44.mlp.experts.104.gate_proj", "model.language_model.layers.44.mlp.experts.105.gate_proj", "model.language_model.layers.44.mlp.experts.106.gate_proj", "model.language_model.layers.44.mlp.experts.107.gate_proj", "model.language_model.layers.44.mlp.experts.108.gate_proj", "model.language_model.layers.44.mlp.experts.109.gate_proj", "model.language_model.layers.44.mlp.experts.110.gate_proj", "model.language_model.layers.44.mlp.experts.111.gate_proj", "model.language_model.layers.44.mlp.experts.112.gate_proj", "model.language_model.layers.44.mlp.experts.113.gate_proj", "model.language_model.layers.44.mlp.experts.114.gate_proj", "model.language_model.layers.44.mlp.experts.115.gate_proj", "model.language_model.layers.44.mlp.experts.116.gate_proj", "model.language_model.layers.44.mlp.experts.117.gate_proj", "model.language_model.layers.44.mlp.experts.118.gate_proj", "model.language_model.layers.44.mlp.experts.119.gate_proj", "model.language_model.layers.44.mlp.experts.120.gate_proj", "model.language_model.layers.44.mlp.experts.121.gate_proj", "model.language_model.layers.44.mlp.experts.122.gate_proj", "model.language_model.layers.44.mlp.experts.123.gate_proj", "model.language_model.layers.44.mlp.experts.124.gate_proj", "model.language_model.layers.44.mlp.experts.125.gate_proj", "model.language_model.layers.44.mlp.experts.126.gate_proj", "model.language_model.layers.44.mlp.experts.127.gate_proj", "model.language_model.layers.44.mlp.experts.0.up_proj", "model.language_model.layers.44.mlp.experts.1.up_proj", "model.language_model.layers.44.mlp.experts.2.up_proj", "model.language_model.layers.44.mlp.experts.3.up_proj", "model.language_model.layers.44.mlp.experts.4.up_proj", "model.language_model.layers.44.mlp.experts.5.up_proj", "model.language_model.layers.44.mlp.experts.6.up_proj", "model.language_model.layers.44.mlp.experts.7.up_proj", "model.language_model.layers.44.mlp.experts.8.up_proj", "model.language_model.layers.44.mlp.experts.9.up_proj", "model.language_model.layers.44.mlp.experts.10.up_proj", "model.language_model.layers.44.mlp.experts.11.up_proj", "model.language_model.layers.44.mlp.experts.12.up_proj", "model.language_model.layers.44.mlp.experts.13.up_proj", "model.language_model.layers.44.mlp.experts.14.up_proj", "model.language_model.layers.44.mlp.experts.15.up_proj", "model.language_model.layers.44.mlp.experts.16.up_proj", "model.language_model.layers.44.mlp.experts.17.up_proj", "model.language_model.layers.44.mlp.experts.18.up_proj", "model.language_model.layers.44.mlp.experts.19.up_proj", "model.language_model.layers.44.mlp.experts.20.up_proj", "model.language_model.layers.44.mlp.experts.21.up_proj", "model.language_model.layers.44.mlp.experts.22.up_proj", "model.language_model.layers.44.mlp.experts.23.up_proj", "model.language_model.layers.44.mlp.experts.24.up_proj", "model.language_model.layers.44.mlp.experts.25.up_proj", "model.language_model.layers.44.mlp.experts.26.up_proj", "model.language_model.layers.44.mlp.experts.27.up_proj", "model.language_model.layers.44.mlp.experts.28.up_proj", "model.language_model.layers.44.mlp.experts.29.up_proj", "model.language_model.layers.44.mlp.experts.30.up_proj", "model.language_model.layers.44.mlp.experts.31.up_proj", "model.language_model.layers.44.mlp.experts.32.up_proj", "model.language_model.layers.44.mlp.experts.33.up_proj", "model.language_model.layers.44.mlp.experts.34.up_proj", "model.language_model.layers.44.mlp.experts.35.up_proj", "model.language_model.layers.44.mlp.experts.36.up_proj", "model.language_model.layers.44.mlp.experts.37.up_proj", "model.language_model.layers.44.mlp.experts.38.up_proj", "model.language_model.layers.44.mlp.experts.39.up_proj", "model.language_model.layers.44.mlp.experts.40.up_proj", "model.language_model.layers.44.mlp.experts.41.up_proj", "model.language_model.layers.44.mlp.experts.42.up_proj", "model.language_model.layers.44.mlp.experts.43.up_proj", "model.language_model.layers.44.mlp.experts.44.up_proj", "model.language_model.layers.44.mlp.experts.45.up_proj", "model.language_model.layers.44.mlp.experts.46.up_proj", "model.language_model.layers.44.mlp.experts.47.up_proj", "model.language_model.layers.44.mlp.experts.48.up_proj", "model.language_model.layers.44.mlp.experts.49.up_proj", "model.language_model.layers.44.mlp.experts.50.up_proj", "model.language_model.layers.44.mlp.experts.51.up_proj", "model.language_model.layers.44.mlp.experts.52.up_proj", "model.language_model.layers.44.mlp.experts.53.up_proj", "model.language_model.layers.44.mlp.experts.54.up_proj", "model.language_model.layers.44.mlp.experts.55.up_proj", "model.language_model.layers.44.mlp.experts.56.up_proj", "model.language_model.layers.44.mlp.experts.57.up_proj", "model.language_model.layers.44.mlp.experts.58.up_proj", "model.language_model.layers.44.mlp.experts.59.up_proj", "model.language_model.layers.44.mlp.experts.60.up_proj", "model.language_model.layers.44.mlp.experts.61.up_proj", "model.language_model.layers.44.mlp.experts.62.up_proj", "model.language_model.layers.44.mlp.experts.63.up_proj", "model.language_model.layers.44.mlp.experts.64.up_proj", "model.language_model.layers.44.mlp.experts.65.up_proj", "model.language_model.layers.44.mlp.experts.66.up_proj", "model.language_model.layers.44.mlp.experts.67.up_proj", "model.language_model.layers.44.mlp.experts.68.up_proj", "model.language_model.layers.44.mlp.experts.69.up_proj", "model.language_model.layers.44.mlp.experts.70.up_proj", "model.language_model.layers.44.mlp.experts.71.up_proj", "model.language_model.layers.44.mlp.experts.72.up_proj", "model.language_model.layers.44.mlp.experts.73.up_proj", "model.language_model.layers.44.mlp.experts.74.up_proj", "model.language_model.layers.44.mlp.experts.75.up_proj", "model.language_model.layers.44.mlp.experts.76.up_proj", "model.language_model.layers.44.mlp.experts.77.up_proj", "model.language_model.layers.44.mlp.experts.78.up_proj", "model.language_model.layers.44.mlp.experts.79.up_proj", "model.language_model.layers.44.mlp.experts.80.up_proj", "model.language_model.layers.44.mlp.experts.81.up_proj", "model.language_model.layers.44.mlp.experts.82.up_proj", "model.language_model.layers.44.mlp.experts.83.up_proj", "model.language_model.layers.44.mlp.experts.84.up_proj", "model.language_model.layers.44.mlp.experts.85.up_proj", "model.language_model.layers.44.mlp.experts.86.up_proj", "model.language_model.layers.44.mlp.experts.87.up_proj", "model.language_model.layers.44.mlp.experts.88.up_proj", "model.language_model.layers.44.mlp.experts.89.up_proj", "model.language_model.layers.44.mlp.experts.90.up_proj", "model.language_model.layers.44.mlp.experts.91.up_proj", "model.language_model.layers.44.mlp.experts.92.up_proj", "model.language_model.layers.44.mlp.experts.93.up_proj", "model.language_model.layers.44.mlp.experts.94.up_proj", "model.language_model.layers.44.mlp.experts.95.up_proj", "model.language_model.layers.44.mlp.experts.96.up_proj", "model.language_model.layers.44.mlp.experts.97.up_proj", "model.language_model.layers.44.mlp.experts.98.up_proj", "model.language_model.layers.44.mlp.experts.99.up_proj", "model.language_model.layers.44.mlp.experts.100.up_proj", "model.language_model.layers.44.mlp.experts.101.up_proj", "model.language_model.layers.44.mlp.experts.102.up_proj", "model.language_model.layers.44.mlp.experts.103.up_proj", "model.language_model.layers.44.mlp.experts.104.up_proj", "model.language_model.layers.44.mlp.experts.105.up_proj", "model.language_model.layers.44.mlp.experts.106.up_proj", "model.language_model.layers.44.mlp.experts.107.up_proj", "model.language_model.layers.44.mlp.experts.108.up_proj", "model.language_model.layers.44.mlp.experts.109.up_proj", "model.language_model.layers.44.mlp.experts.110.up_proj", "model.language_model.layers.44.mlp.experts.111.up_proj", "model.language_model.layers.44.mlp.experts.112.up_proj", "model.language_model.layers.44.mlp.experts.113.up_proj", "model.language_model.layers.44.mlp.experts.114.up_proj", "model.language_model.layers.44.mlp.experts.115.up_proj", "model.language_model.layers.44.mlp.experts.116.up_proj", "model.language_model.layers.44.mlp.experts.117.up_proj", "model.language_model.layers.44.mlp.experts.118.up_proj", "model.language_model.layers.44.mlp.experts.119.up_proj", "model.language_model.layers.44.mlp.experts.120.up_proj", "model.language_model.layers.44.mlp.experts.121.up_proj", "model.language_model.layers.44.mlp.experts.122.up_proj", "model.language_model.layers.44.mlp.experts.123.up_proj", "model.language_model.layers.44.mlp.experts.124.up_proj", "model.language_model.layers.44.mlp.experts.125.up_proj", "model.language_model.layers.44.mlp.experts.126.up_proj", "model.language_model.layers.44.mlp.experts.127.up_proj", "model.language_model.layers.44.mlp.experts.0.down_proj", "model.language_model.layers.44.mlp.experts.1.down_proj", "model.language_model.layers.44.mlp.experts.2.down_proj", "model.language_model.layers.44.mlp.experts.3.down_proj", "model.language_model.layers.44.mlp.experts.4.down_proj", "model.language_model.layers.44.mlp.experts.5.down_proj", "model.language_model.layers.44.mlp.experts.6.down_proj", "model.language_model.layers.44.mlp.experts.7.down_proj", "model.language_model.layers.44.mlp.experts.8.down_proj", "model.language_model.layers.44.mlp.experts.9.down_proj", "model.language_model.layers.44.mlp.experts.10.down_proj", "model.language_model.layers.44.mlp.experts.11.down_proj", "model.language_model.layers.44.mlp.experts.12.down_proj", "model.language_model.layers.44.mlp.experts.13.down_proj", "model.language_model.layers.44.mlp.experts.14.down_proj", "model.language_model.layers.44.mlp.experts.15.down_proj", "model.language_model.layers.44.mlp.experts.16.down_proj", "model.language_model.layers.44.mlp.experts.17.down_proj", "model.language_model.layers.44.mlp.experts.18.down_proj", "model.language_model.layers.44.mlp.experts.19.down_proj", "model.language_model.layers.44.mlp.experts.20.down_proj", "model.language_model.layers.44.mlp.experts.21.down_proj", "model.language_model.layers.44.mlp.experts.22.down_proj", "model.language_model.layers.44.mlp.experts.23.down_proj", "model.language_model.layers.44.mlp.experts.24.down_proj", "model.language_model.layers.44.mlp.experts.25.down_proj", "model.language_model.layers.44.mlp.experts.26.down_proj", "model.language_model.layers.44.mlp.experts.27.down_proj", "model.language_model.layers.44.mlp.experts.28.down_proj", "model.language_model.layers.44.mlp.experts.29.down_proj", "model.language_model.layers.44.mlp.experts.30.down_proj", "model.language_model.layers.44.mlp.experts.31.down_proj", "model.language_model.layers.44.mlp.experts.32.down_proj", "model.language_model.layers.44.mlp.experts.33.down_proj", "model.language_model.layers.44.mlp.experts.34.down_proj", "model.language_model.layers.44.mlp.experts.35.down_proj", "model.language_model.layers.44.mlp.experts.36.down_proj", "model.language_model.layers.44.mlp.experts.37.down_proj", "model.language_model.layers.44.mlp.experts.38.down_proj", "model.language_model.layers.44.mlp.experts.39.down_proj", "model.language_model.layers.44.mlp.experts.40.down_proj", "model.language_model.layers.44.mlp.experts.41.down_proj", "model.language_model.layers.44.mlp.experts.42.down_proj", "model.language_model.layers.44.mlp.experts.43.down_proj", "model.language_model.layers.44.mlp.experts.44.down_proj", "model.language_model.layers.44.mlp.experts.45.down_proj", "model.language_model.layers.44.mlp.experts.46.down_proj", "model.language_model.layers.44.mlp.experts.47.down_proj", "model.language_model.layers.44.mlp.experts.48.down_proj", "model.language_model.layers.44.mlp.experts.49.down_proj", "model.language_model.layers.44.mlp.experts.50.down_proj", "model.language_model.layers.44.mlp.experts.51.down_proj", "model.language_model.layers.44.mlp.experts.52.down_proj", "model.language_model.layers.44.mlp.experts.53.down_proj", "model.language_model.layers.44.mlp.experts.54.down_proj", "model.language_model.layers.44.mlp.experts.55.down_proj", "model.language_model.layers.44.mlp.experts.56.down_proj", "model.language_model.layers.44.mlp.experts.57.down_proj", "model.language_model.layers.44.mlp.experts.58.down_proj", "model.language_model.layers.44.mlp.experts.59.down_proj", "model.language_model.layers.44.mlp.experts.60.down_proj", "model.language_model.layers.44.mlp.experts.61.down_proj", "model.language_model.layers.44.mlp.experts.62.down_proj", "model.language_model.layers.44.mlp.experts.63.down_proj", "model.language_model.layers.44.mlp.experts.64.down_proj", "model.language_model.layers.44.mlp.experts.65.down_proj", "model.language_model.layers.44.mlp.experts.66.down_proj", "model.language_model.layers.44.mlp.experts.67.down_proj", "model.language_model.layers.44.mlp.experts.68.down_proj", "model.language_model.layers.44.mlp.experts.69.down_proj", "model.language_model.layers.44.mlp.experts.70.down_proj", "model.language_model.layers.44.mlp.experts.71.down_proj", "model.language_model.layers.44.mlp.experts.72.down_proj", "model.language_model.layers.44.mlp.experts.73.down_proj", "model.language_model.layers.44.mlp.experts.74.down_proj", "model.language_model.layers.44.mlp.experts.75.down_proj", "model.language_model.layers.44.mlp.experts.76.down_proj", "model.language_model.layers.44.mlp.experts.77.down_proj", "model.language_model.layers.44.mlp.experts.78.down_proj", "model.language_model.layers.44.mlp.experts.79.down_proj", "model.language_model.layers.44.mlp.experts.80.down_proj", "model.language_model.layers.44.mlp.experts.81.down_proj", "model.language_model.layers.44.mlp.experts.82.down_proj", "model.language_model.layers.44.mlp.experts.83.down_proj", "model.language_model.layers.44.mlp.experts.84.down_proj", "model.language_model.layers.44.mlp.experts.85.down_proj", "model.language_model.layers.44.mlp.experts.86.down_proj", "model.language_model.layers.44.mlp.experts.87.down_proj", "model.language_model.layers.44.mlp.experts.88.down_proj", "model.language_model.layers.44.mlp.experts.89.down_proj", "model.language_model.layers.44.mlp.experts.90.down_proj", "model.language_model.layers.44.mlp.experts.91.down_proj", "model.language_model.layers.44.mlp.experts.92.down_proj", "model.language_model.layers.44.mlp.experts.93.down_proj", "model.language_model.layers.44.mlp.experts.94.down_proj", "model.language_model.layers.44.mlp.experts.95.down_proj", "model.language_model.layers.44.mlp.experts.96.down_proj", "model.language_model.layers.44.mlp.experts.97.down_proj", "model.language_model.layers.44.mlp.experts.98.down_proj", "model.language_model.layers.44.mlp.experts.99.down_proj", "model.language_model.layers.44.mlp.experts.100.down_proj", "model.language_model.layers.44.mlp.experts.101.down_proj", "model.language_model.layers.44.mlp.experts.102.down_proj", "model.language_model.layers.44.mlp.experts.103.down_proj", "model.language_model.layers.44.mlp.experts.104.down_proj", "model.language_model.layers.44.mlp.experts.105.down_proj", "model.language_model.layers.44.mlp.experts.106.down_proj", "model.language_model.layers.44.mlp.experts.107.down_proj", "model.language_model.layers.44.mlp.experts.108.down_proj", "model.language_model.layers.44.mlp.experts.109.down_proj", "model.language_model.layers.44.mlp.experts.110.down_proj", "model.language_model.layers.44.mlp.experts.111.down_proj", "model.language_model.layers.44.mlp.experts.112.down_proj", "model.language_model.layers.44.mlp.experts.113.down_proj", "model.language_model.layers.44.mlp.experts.114.down_proj", "model.language_model.layers.44.mlp.experts.115.down_proj", "model.language_model.layers.44.mlp.experts.116.down_proj", "model.language_model.layers.44.mlp.experts.117.down_proj", "model.language_model.layers.44.mlp.experts.118.down_proj", "model.language_model.layers.44.mlp.experts.119.down_proj", "model.language_model.layers.44.mlp.experts.120.down_proj", "model.language_model.layers.44.mlp.experts.121.down_proj", "model.language_model.layers.44.mlp.experts.122.down_proj", "model.language_model.layers.44.mlp.experts.123.down_proj", "model.language_model.layers.44.mlp.experts.124.down_proj", "model.language_model.layers.44.mlp.experts.125.down_proj", "model.language_model.layers.44.mlp.experts.126.down_proj", "model.language_model.layers.44.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.0007986719720065594, "dbits": 603979776 } ] }, { "idx": 90, "layers": [ "model.language_model.layers.45.self_attn.q_proj", "model.language_model.layers.45.self_attn.k_proj", "model.language_model.layers.45.self_attn.v_proj", "model.language_model.layers.45.self_attn.o_proj" ], "candidates": [ { "dkld": 3.2537337392571364e-05, "dbits": 18874368 } ] }, { "idx": 91, "layers": [ "model.language_model.layers.45.mlp.experts.0.gate_proj", "model.language_model.layers.45.mlp.experts.1.gate_proj", "model.language_model.layers.45.mlp.experts.2.gate_proj", "model.language_model.layers.45.mlp.experts.3.gate_proj", "model.language_model.layers.45.mlp.experts.4.gate_proj", "model.language_model.layers.45.mlp.experts.5.gate_proj", "model.language_model.layers.45.mlp.experts.6.gate_proj", "model.language_model.layers.45.mlp.experts.7.gate_proj", "model.language_model.layers.45.mlp.experts.8.gate_proj", "model.language_model.layers.45.mlp.experts.9.gate_proj", "model.language_model.layers.45.mlp.experts.10.gate_proj", "model.language_model.layers.45.mlp.experts.11.gate_proj", "model.language_model.layers.45.mlp.experts.12.gate_proj", "model.language_model.layers.45.mlp.experts.13.gate_proj", "model.language_model.layers.45.mlp.experts.14.gate_proj", "model.language_model.layers.45.mlp.experts.15.gate_proj", "model.language_model.layers.45.mlp.experts.16.gate_proj", "model.language_model.layers.45.mlp.experts.17.gate_proj", "model.language_model.layers.45.mlp.experts.18.gate_proj", "model.language_model.layers.45.mlp.experts.19.gate_proj", "model.language_model.layers.45.mlp.experts.20.gate_proj", "model.language_model.layers.45.mlp.experts.21.gate_proj", "model.language_model.layers.45.mlp.experts.22.gate_proj", "model.language_model.layers.45.mlp.experts.23.gate_proj", "model.language_model.layers.45.mlp.experts.24.gate_proj", "model.language_model.layers.45.mlp.experts.25.gate_proj", "model.language_model.layers.45.mlp.experts.26.gate_proj", "model.language_model.layers.45.mlp.experts.27.gate_proj", "model.language_model.layers.45.mlp.experts.28.gate_proj", "model.language_model.layers.45.mlp.experts.29.gate_proj", "model.language_model.layers.45.mlp.experts.30.gate_proj", "model.language_model.layers.45.mlp.experts.31.gate_proj", "model.language_model.layers.45.mlp.experts.32.gate_proj", "model.language_model.layers.45.mlp.experts.33.gate_proj", "model.language_model.layers.45.mlp.experts.34.gate_proj", "model.language_model.layers.45.mlp.experts.35.gate_proj", "model.language_model.layers.45.mlp.experts.36.gate_proj", "model.language_model.layers.45.mlp.experts.37.gate_proj", "model.language_model.layers.45.mlp.experts.38.gate_proj", "model.language_model.layers.45.mlp.experts.39.gate_proj", "model.language_model.layers.45.mlp.experts.40.gate_proj", "model.language_model.layers.45.mlp.experts.41.gate_proj", "model.language_model.layers.45.mlp.experts.42.gate_proj", "model.language_model.layers.45.mlp.experts.43.gate_proj", "model.language_model.layers.45.mlp.experts.44.gate_proj", "model.language_model.layers.45.mlp.experts.45.gate_proj", "model.language_model.layers.45.mlp.experts.46.gate_proj", "model.language_model.layers.45.mlp.experts.47.gate_proj", "model.language_model.layers.45.mlp.experts.48.gate_proj", "model.language_model.layers.45.mlp.experts.49.gate_proj", "model.language_model.layers.45.mlp.experts.50.gate_proj", "model.language_model.layers.45.mlp.experts.51.gate_proj", "model.language_model.layers.45.mlp.experts.52.gate_proj", "model.language_model.layers.45.mlp.experts.53.gate_proj", "model.language_model.layers.45.mlp.experts.54.gate_proj", "model.language_model.layers.45.mlp.experts.55.gate_proj", "model.language_model.layers.45.mlp.experts.56.gate_proj", "model.language_model.layers.45.mlp.experts.57.gate_proj", "model.language_model.layers.45.mlp.experts.58.gate_proj", "model.language_model.layers.45.mlp.experts.59.gate_proj", "model.language_model.layers.45.mlp.experts.60.gate_proj", "model.language_model.layers.45.mlp.experts.61.gate_proj", "model.language_model.layers.45.mlp.experts.62.gate_proj", "model.language_model.layers.45.mlp.experts.63.gate_proj", "model.language_model.layers.45.mlp.experts.64.gate_proj", "model.language_model.layers.45.mlp.experts.65.gate_proj", "model.language_model.layers.45.mlp.experts.66.gate_proj", "model.language_model.layers.45.mlp.experts.67.gate_proj", "model.language_model.layers.45.mlp.experts.68.gate_proj", "model.language_model.layers.45.mlp.experts.69.gate_proj", "model.language_model.layers.45.mlp.experts.70.gate_proj", "model.language_model.layers.45.mlp.experts.71.gate_proj", "model.language_model.layers.45.mlp.experts.72.gate_proj", "model.language_model.layers.45.mlp.experts.73.gate_proj", "model.language_model.layers.45.mlp.experts.74.gate_proj", "model.language_model.layers.45.mlp.experts.75.gate_proj", "model.language_model.layers.45.mlp.experts.76.gate_proj", "model.language_model.layers.45.mlp.experts.77.gate_proj", "model.language_model.layers.45.mlp.experts.78.gate_proj", "model.language_model.layers.45.mlp.experts.79.gate_proj", "model.language_model.layers.45.mlp.experts.80.gate_proj", "model.language_model.layers.45.mlp.experts.81.gate_proj", "model.language_model.layers.45.mlp.experts.82.gate_proj", "model.language_model.layers.45.mlp.experts.83.gate_proj", "model.language_model.layers.45.mlp.experts.84.gate_proj", "model.language_model.layers.45.mlp.experts.85.gate_proj", "model.language_model.layers.45.mlp.experts.86.gate_proj", "model.language_model.layers.45.mlp.experts.87.gate_proj", "model.language_model.layers.45.mlp.experts.88.gate_proj", "model.language_model.layers.45.mlp.experts.89.gate_proj", "model.language_model.layers.45.mlp.experts.90.gate_proj", "model.language_model.layers.45.mlp.experts.91.gate_proj", "model.language_model.layers.45.mlp.experts.92.gate_proj", "model.language_model.layers.45.mlp.experts.93.gate_proj", "model.language_model.layers.45.mlp.experts.94.gate_proj", "model.language_model.layers.45.mlp.experts.95.gate_proj", "model.language_model.layers.45.mlp.experts.96.gate_proj", "model.language_model.layers.45.mlp.experts.97.gate_proj", "model.language_model.layers.45.mlp.experts.98.gate_proj", "model.language_model.layers.45.mlp.experts.99.gate_proj", "model.language_model.layers.45.mlp.experts.100.gate_proj", "model.language_model.layers.45.mlp.experts.101.gate_proj", "model.language_model.layers.45.mlp.experts.102.gate_proj", "model.language_model.layers.45.mlp.experts.103.gate_proj", "model.language_model.layers.45.mlp.experts.104.gate_proj", "model.language_model.layers.45.mlp.experts.105.gate_proj", "model.language_model.layers.45.mlp.experts.106.gate_proj", "model.language_model.layers.45.mlp.experts.107.gate_proj", "model.language_model.layers.45.mlp.experts.108.gate_proj", "model.language_model.layers.45.mlp.experts.109.gate_proj", "model.language_model.layers.45.mlp.experts.110.gate_proj", "model.language_model.layers.45.mlp.experts.111.gate_proj", "model.language_model.layers.45.mlp.experts.112.gate_proj", "model.language_model.layers.45.mlp.experts.113.gate_proj", "model.language_model.layers.45.mlp.experts.114.gate_proj", "model.language_model.layers.45.mlp.experts.115.gate_proj", "model.language_model.layers.45.mlp.experts.116.gate_proj", "model.language_model.layers.45.mlp.experts.117.gate_proj", "model.language_model.layers.45.mlp.experts.118.gate_proj", "model.language_model.layers.45.mlp.experts.119.gate_proj", "model.language_model.layers.45.mlp.experts.120.gate_proj", "model.language_model.layers.45.mlp.experts.121.gate_proj", "model.language_model.layers.45.mlp.experts.122.gate_proj", "model.language_model.layers.45.mlp.experts.123.gate_proj", "model.language_model.layers.45.mlp.experts.124.gate_proj", "model.language_model.layers.45.mlp.experts.125.gate_proj", "model.language_model.layers.45.mlp.experts.126.gate_proj", "model.language_model.layers.45.mlp.experts.127.gate_proj", "model.language_model.layers.45.mlp.experts.0.up_proj", "model.language_model.layers.45.mlp.experts.1.up_proj", "model.language_model.layers.45.mlp.experts.2.up_proj", "model.language_model.layers.45.mlp.experts.3.up_proj", "model.language_model.layers.45.mlp.experts.4.up_proj", "model.language_model.layers.45.mlp.experts.5.up_proj", "model.language_model.layers.45.mlp.experts.6.up_proj", "model.language_model.layers.45.mlp.experts.7.up_proj", "model.language_model.layers.45.mlp.experts.8.up_proj", "model.language_model.layers.45.mlp.experts.9.up_proj", "model.language_model.layers.45.mlp.experts.10.up_proj", "model.language_model.layers.45.mlp.experts.11.up_proj", "model.language_model.layers.45.mlp.experts.12.up_proj", "model.language_model.layers.45.mlp.experts.13.up_proj", "model.language_model.layers.45.mlp.experts.14.up_proj", "model.language_model.layers.45.mlp.experts.15.up_proj", "model.language_model.layers.45.mlp.experts.16.up_proj", "model.language_model.layers.45.mlp.experts.17.up_proj", "model.language_model.layers.45.mlp.experts.18.up_proj", "model.language_model.layers.45.mlp.experts.19.up_proj", "model.language_model.layers.45.mlp.experts.20.up_proj", "model.language_model.layers.45.mlp.experts.21.up_proj", "model.language_model.layers.45.mlp.experts.22.up_proj", "model.language_model.layers.45.mlp.experts.23.up_proj", "model.language_model.layers.45.mlp.experts.24.up_proj", "model.language_model.layers.45.mlp.experts.25.up_proj", "model.language_model.layers.45.mlp.experts.26.up_proj", "model.language_model.layers.45.mlp.experts.27.up_proj", "model.language_model.layers.45.mlp.experts.28.up_proj", "model.language_model.layers.45.mlp.experts.29.up_proj", "model.language_model.layers.45.mlp.experts.30.up_proj", "model.language_model.layers.45.mlp.experts.31.up_proj", "model.language_model.layers.45.mlp.experts.32.up_proj", "model.language_model.layers.45.mlp.experts.33.up_proj", "model.language_model.layers.45.mlp.experts.34.up_proj", "model.language_model.layers.45.mlp.experts.35.up_proj", "model.language_model.layers.45.mlp.experts.36.up_proj", "model.language_model.layers.45.mlp.experts.37.up_proj", "model.language_model.layers.45.mlp.experts.38.up_proj", "model.language_model.layers.45.mlp.experts.39.up_proj", "model.language_model.layers.45.mlp.experts.40.up_proj", "model.language_model.layers.45.mlp.experts.41.up_proj", "model.language_model.layers.45.mlp.experts.42.up_proj", "model.language_model.layers.45.mlp.experts.43.up_proj", "model.language_model.layers.45.mlp.experts.44.up_proj", "model.language_model.layers.45.mlp.experts.45.up_proj", "model.language_model.layers.45.mlp.experts.46.up_proj", "model.language_model.layers.45.mlp.experts.47.up_proj", "model.language_model.layers.45.mlp.experts.48.up_proj", "model.language_model.layers.45.mlp.experts.49.up_proj", "model.language_model.layers.45.mlp.experts.50.up_proj", "model.language_model.layers.45.mlp.experts.51.up_proj", "model.language_model.layers.45.mlp.experts.52.up_proj", "model.language_model.layers.45.mlp.experts.53.up_proj", "model.language_model.layers.45.mlp.experts.54.up_proj", "model.language_model.layers.45.mlp.experts.55.up_proj", "model.language_model.layers.45.mlp.experts.56.up_proj", "model.language_model.layers.45.mlp.experts.57.up_proj", "model.language_model.layers.45.mlp.experts.58.up_proj", "model.language_model.layers.45.mlp.experts.59.up_proj", "model.language_model.layers.45.mlp.experts.60.up_proj", "model.language_model.layers.45.mlp.experts.61.up_proj", "model.language_model.layers.45.mlp.experts.62.up_proj", "model.language_model.layers.45.mlp.experts.63.up_proj", "model.language_model.layers.45.mlp.experts.64.up_proj", "model.language_model.layers.45.mlp.experts.65.up_proj", "model.language_model.layers.45.mlp.experts.66.up_proj", "model.language_model.layers.45.mlp.experts.67.up_proj", "model.language_model.layers.45.mlp.experts.68.up_proj", "model.language_model.layers.45.mlp.experts.69.up_proj", "model.language_model.layers.45.mlp.experts.70.up_proj", "model.language_model.layers.45.mlp.experts.71.up_proj", "model.language_model.layers.45.mlp.experts.72.up_proj", "model.language_model.layers.45.mlp.experts.73.up_proj", "model.language_model.layers.45.mlp.experts.74.up_proj", "model.language_model.layers.45.mlp.experts.75.up_proj", "model.language_model.layers.45.mlp.experts.76.up_proj", "model.language_model.layers.45.mlp.experts.77.up_proj", "model.language_model.layers.45.mlp.experts.78.up_proj", "model.language_model.layers.45.mlp.experts.79.up_proj", "model.language_model.layers.45.mlp.experts.80.up_proj", "model.language_model.layers.45.mlp.experts.81.up_proj", "model.language_model.layers.45.mlp.experts.82.up_proj", "model.language_model.layers.45.mlp.experts.83.up_proj", "model.language_model.layers.45.mlp.experts.84.up_proj", "model.language_model.layers.45.mlp.experts.85.up_proj", "model.language_model.layers.45.mlp.experts.86.up_proj", "model.language_model.layers.45.mlp.experts.87.up_proj", "model.language_model.layers.45.mlp.experts.88.up_proj", "model.language_model.layers.45.mlp.experts.89.up_proj", "model.language_model.layers.45.mlp.experts.90.up_proj", "model.language_model.layers.45.mlp.experts.91.up_proj", "model.language_model.layers.45.mlp.experts.92.up_proj", "model.language_model.layers.45.mlp.experts.93.up_proj", "model.language_model.layers.45.mlp.experts.94.up_proj", "model.language_model.layers.45.mlp.experts.95.up_proj", "model.language_model.layers.45.mlp.experts.96.up_proj", "model.language_model.layers.45.mlp.experts.97.up_proj", "model.language_model.layers.45.mlp.experts.98.up_proj", "model.language_model.layers.45.mlp.experts.99.up_proj", "model.language_model.layers.45.mlp.experts.100.up_proj", "model.language_model.layers.45.mlp.experts.101.up_proj", "model.language_model.layers.45.mlp.experts.102.up_proj", "model.language_model.layers.45.mlp.experts.103.up_proj", "model.language_model.layers.45.mlp.experts.104.up_proj", "model.language_model.layers.45.mlp.experts.105.up_proj", "model.language_model.layers.45.mlp.experts.106.up_proj", "model.language_model.layers.45.mlp.experts.107.up_proj", "model.language_model.layers.45.mlp.experts.108.up_proj", "model.language_model.layers.45.mlp.experts.109.up_proj", "model.language_model.layers.45.mlp.experts.110.up_proj", "model.language_model.layers.45.mlp.experts.111.up_proj", "model.language_model.layers.45.mlp.experts.112.up_proj", "model.language_model.layers.45.mlp.experts.113.up_proj", "model.language_model.layers.45.mlp.experts.114.up_proj", "model.language_model.layers.45.mlp.experts.115.up_proj", "model.language_model.layers.45.mlp.experts.116.up_proj", "model.language_model.layers.45.mlp.experts.117.up_proj", "model.language_model.layers.45.mlp.experts.118.up_proj", "model.language_model.layers.45.mlp.experts.119.up_proj", "model.language_model.layers.45.mlp.experts.120.up_proj", "model.language_model.layers.45.mlp.experts.121.up_proj", "model.language_model.layers.45.mlp.experts.122.up_proj", "model.language_model.layers.45.mlp.experts.123.up_proj", "model.language_model.layers.45.mlp.experts.124.up_proj", "model.language_model.layers.45.mlp.experts.125.up_proj", "model.language_model.layers.45.mlp.experts.126.up_proj", "model.language_model.layers.45.mlp.experts.127.up_proj", "model.language_model.layers.45.mlp.experts.0.down_proj", "model.language_model.layers.45.mlp.experts.1.down_proj", "model.language_model.layers.45.mlp.experts.2.down_proj", "model.language_model.layers.45.mlp.experts.3.down_proj", "model.language_model.layers.45.mlp.experts.4.down_proj", "model.language_model.layers.45.mlp.experts.5.down_proj", "model.language_model.layers.45.mlp.experts.6.down_proj", "model.language_model.layers.45.mlp.experts.7.down_proj", "model.language_model.layers.45.mlp.experts.8.down_proj", "model.language_model.layers.45.mlp.experts.9.down_proj", "model.language_model.layers.45.mlp.experts.10.down_proj", "model.language_model.layers.45.mlp.experts.11.down_proj", "model.language_model.layers.45.mlp.experts.12.down_proj", "model.language_model.layers.45.mlp.experts.13.down_proj", "model.language_model.layers.45.mlp.experts.14.down_proj", "model.language_model.layers.45.mlp.experts.15.down_proj", "model.language_model.layers.45.mlp.experts.16.down_proj", "model.language_model.layers.45.mlp.experts.17.down_proj", "model.language_model.layers.45.mlp.experts.18.down_proj", "model.language_model.layers.45.mlp.experts.19.down_proj", "model.language_model.layers.45.mlp.experts.20.down_proj", "model.language_model.layers.45.mlp.experts.21.down_proj", "model.language_model.layers.45.mlp.experts.22.down_proj", "model.language_model.layers.45.mlp.experts.23.down_proj", "model.language_model.layers.45.mlp.experts.24.down_proj", "model.language_model.layers.45.mlp.experts.25.down_proj", "model.language_model.layers.45.mlp.experts.26.down_proj", "model.language_model.layers.45.mlp.experts.27.down_proj", "model.language_model.layers.45.mlp.experts.28.down_proj", "model.language_model.layers.45.mlp.experts.29.down_proj", "model.language_model.layers.45.mlp.experts.30.down_proj", "model.language_model.layers.45.mlp.experts.31.down_proj", "model.language_model.layers.45.mlp.experts.32.down_proj", "model.language_model.layers.45.mlp.experts.33.down_proj", "model.language_model.layers.45.mlp.experts.34.down_proj", "model.language_model.layers.45.mlp.experts.35.down_proj", "model.language_model.layers.45.mlp.experts.36.down_proj", "model.language_model.layers.45.mlp.experts.37.down_proj", "model.language_model.layers.45.mlp.experts.38.down_proj", "model.language_model.layers.45.mlp.experts.39.down_proj", "model.language_model.layers.45.mlp.experts.40.down_proj", "model.language_model.layers.45.mlp.experts.41.down_proj", "model.language_model.layers.45.mlp.experts.42.down_proj", "model.language_model.layers.45.mlp.experts.43.down_proj", "model.language_model.layers.45.mlp.experts.44.down_proj", "model.language_model.layers.45.mlp.experts.45.down_proj", "model.language_model.layers.45.mlp.experts.46.down_proj", "model.language_model.layers.45.mlp.experts.47.down_proj", "model.language_model.layers.45.mlp.experts.48.down_proj", "model.language_model.layers.45.mlp.experts.49.down_proj", "model.language_model.layers.45.mlp.experts.50.down_proj", "model.language_model.layers.45.mlp.experts.51.down_proj", "model.language_model.layers.45.mlp.experts.52.down_proj", "model.language_model.layers.45.mlp.experts.53.down_proj", "model.language_model.layers.45.mlp.experts.54.down_proj", "model.language_model.layers.45.mlp.experts.55.down_proj", "model.language_model.layers.45.mlp.experts.56.down_proj", "model.language_model.layers.45.mlp.experts.57.down_proj", "model.language_model.layers.45.mlp.experts.58.down_proj", "model.language_model.layers.45.mlp.experts.59.down_proj", "model.language_model.layers.45.mlp.experts.60.down_proj", "model.language_model.layers.45.mlp.experts.61.down_proj", "model.language_model.layers.45.mlp.experts.62.down_proj", "model.language_model.layers.45.mlp.experts.63.down_proj", "model.language_model.layers.45.mlp.experts.64.down_proj", "model.language_model.layers.45.mlp.experts.65.down_proj", "model.language_model.layers.45.mlp.experts.66.down_proj", "model.language_model.layers.45.mlp.experts.67.down_proj", "model.language_model.layers.45.mlp.experts.68.down_proj", "model.language_model.layers.45.mlp.experts.69.down_proj", "model.language_model.layers.45.mlp.experts.70.down_proj", "model.language_model.layers.45.mlp.experts.71.down_proj", "model.language_model.layers.45.mlp.experts.72.down_proj", "model.language_model.layers.45.mlp.experts.73.down_proj", "model.language_model.layers.45.mlp.experts.74.down_proj", "model.language_model.layers.45.mlp.experts.75.down_proj", "model.language_model.layers.45.mlp.experts.76.down_proj", "model.language_model.layers.45.mlp.experts.77.down_proj", "model.language_model.layers.45.mlp.experts.78.down_proj", "model.language_model.layers.45.mlp.experts.79.down_proj", "model.language_model.layers.45.mlp.experts.80.down_proj", "model.language_model.layers.45.mlp.experts.81.down_proj", "model.language_model.layers.45.mlp.experts.82.down_proj", "model.language_model.layers.45.mlp.experts.83.down_proj", "model.language_model.layers.45.mlp.experts.84.down_proj", "model.language_model.layers.45.mlp.experts.85.down_proj", "model.language_model.layers.45.mlp.experts.86.down_proj", "model.language_model.layers.45.mlp.experts.87.down_proj", "model.language_model.layers.45.mlp.experts.88.down_proj", "model.language_model.layers.45.mlp.experts.89.down_proj", "model.language_model.layers.45.mlp.experts.90.down_proj", "model.language_model.layers.45.mlp.experts.91.down_proj", "model.language_model.layers.45.mlp.experts.92.down_proj", "model.language_model.layers.45.mlp.experts.93.down_proj", "model.language_model.layers.45.mlp.experts.94.down_proj", "model.language_model.layers.45.mlp.experts.95.down_proj", "model.language_model.layers.45.mlp.experts.96.down_proj", "model.language_model.layers.45.mlp.experts.97.down_proj", "model.language_model.layers.45.mlp.experts.98.down_proj", "model.language_model.layers.45.mlp.experts.99.down_proj", "model.language_model.layers.45.mlp.experts.100.down_proj", "model.language_model.layers.45.mlp.experts.101.down_proj", "model.language_model.layers.45.mlp.experts.102.down_proj", "model.language_model.layers.45.mlp.experts.103.down_proj", "model.language_model.layers.45.mlp.experts.104.down_proj", "model.language_model.layers.45.mlp.experts.105.down_proj", "model.language_model.layers.45.mlp.experts.106.down_proj", "model.language_model.layers.45.mlp.experts.107.down_proj", "model.language_model.layers.45.mlp.experts.108.down_proj", "model.language_model.layers.45.mlp.experts.109.down_proj", "model.language_model.layers.45.mlp.experts.110.down_proj", "model.language_model.layers.45.mlp.experts.111.down_proj", "model.language_model.layers.45.mlp.experts.112.down_proj", "model.language_model.layers.45.mlp.experts.113.down_proj", "model.language_model.layers.45.mlp.experts.114.down_proj", "model.language_model.layers.45.mlp.experts.115.down_proj", "model.language_model.layers.45.mlp.experts.116.down_proj", "model.language_model.layers.45.mlp.experts.117.down_proj", "model.language_model.layers.45.mlp.experts.118.down_proj", "model.language_model.layers.45.mlp.experts.119.down_proj", "model.language_model.layers.45.mlp.experts.120.down_proj", "model.language_model.layers.45.mlp.experts.121.down_proj", "model.language_model.layers.45.mlp.experts.122.down_proj", "model.language_model.layers.45.mlp.experts.123.down_proj", "model.language_model.layers.45.mlp.experts.124.down_proj", "model.language_model.layers.45.mlp.experts.125.down_proj", "model.language_model.layers.45.mlp.experts.126.down_proj", "model.language_model.layers.45.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.0007184792309999494, "dbits": 603979776 } ] }, { "idx": 92, "layers": [ "model.language_model.layers.46.self_attn.q_proj", "model.language_model.layers.46.self_attn.k_proj", "model.language_model.layers.46.self_attn.v_proj", "model.language_model.layers.46.self_attn.o_proj" ], "candidates": [ { "dkld": -0.000562804099172351, "dbits": 18874368 } ] }, { "idx": 93, "layers": [ "model.language_model.layers.46.mlp.experts.0.gate_proj", "model.language_model.layers.46.mlp.experts.1.gate_proj", "model.language_model.layers.46.mlp.experts.2.gate_proj", "model.language_model.layers.46.mlp.experts.3.gate_proj", "model.language_model.layers.46.mlp.experts.4.gate_proj", "model.language_model.layers.46.mlp.experts.5.gate_proj", "model.language_model.layers.46.mlp.experts.6.gate_proj", "model.language_model.layers.46.mlp.experts.7.gate_proj", "model.language_model.layers.46.mlp.experts.8.gate_proj", "model.language_model.layers.46.mlp.experts.9.gate_proj", "model.language_model.layers.46.mlp.experts.10.gate_proj", "model.language_model.layers.46.mlp.experts.11.gate_proj", "model.language_model.layers.46.mlp.experts.12.gate_proj", "model.language_model.layers.46.mlp.experts.13.gate_proj", "model.language_model.layers.46.mlp.experts.14.gate_proj", "model.language_model.layers.46.mlp.experts.15.gate_proj", "model.language_model.layers.46.mlp.experts.16.gate_proj", "model.language_model.layers.46.mlp.experts.17.gate_proj", "model.language_model.layers.46.mlp.experts.18.gate_proj", "model.language_model.layers.46.mlp.experts.19.gate_proj", "model.language_model.layers.46.mlp.experts.20.gate_proj", "model.language_model.layers.46.mlp.experts.21.gate_proj", "model.language_model.layers.46.mlp.experts.22.gate_proj", "model.language_model.layers.46.mlp.experts.23.gate_proj", "model.language_model.layers.46.mlp.experts.24.gate_proj", "model.language_model.layers.46.mlp.experts.25.gate_proj", "model.language_model.layers.46.mlp.experts.26.gate_proj", "model.language_model.layers.46.mlp.experts.27.gate_proj", "model.language_model.layers.46.mlp.experts.28.gate_proj", "model.language_model.layers.46.mlp.experts.29.gate_proj", "model.language_model.layers.46.mlp.experts.30.gate_proj", "model.language_model.layers.46.mlp.experts.31.gate_proj", "model.language_model.layers.46.mlp.experts.32.gate_proj", "model.language_model.layers.46.mlp.experts.33.gate_proj", "model.language_model.layers.46.mlp.experts.34.gate_proj", "model.language_model.layers.46.mlp.experts.35.gate_proj", "model.language_model.layers.46.mlp.experts.36.gate_proj", "model.language_model.layers.46.mlp.experts.37.gate_proj", "model.language_model.layers.46.mlp.experts.38.gate_proj", "model.language_model.layers.46.mlp.experts.39.gate_proj", "model.language_model.layers.46.mlp.experts.40.gate_proj", "model.language_model.layers.46.mlp.experts.41.gate_proj", "model.language_model.layers.46.mlp.experts.42.gate_proj", "model.language_model.layers.46.mlp.experts.43.gate_proj", "model.language_model.layers.46.mlp.experts.44.gate_proj", "model.language_model.layers.46.mlp.experts.45.gate_proj", "model.language_model.layers.46.mlp.experts.46.gate_proj", "model.language_model.layers.46.mlp.experts.47.gate_proj", "model.language_model.layers.46.mlp.experts.48.gate_proj", "model.language_model.layers.46.mlp.experts.49.gate_proj", "model.language_model.layers.46.mlp.experts.50.gate_proj", "model.language_model.layers.46.mlp.experts.51.gate_proj", "model.language_model.layers.46.mlp.experts.52.gate_proj", "model.language_model.layers.46.mlp.experts.53.gate_proj", "model.language_model.layers.46.mlp.experts.54.gate_proj", "model.language_model.layers.46.mlp.experts.55.gate_proj", "model.language_model.layers.46.mlp.experts.56.gate_proj", "model.language_model.layers.46.mlp.experts.57.gate_proj", "model.language_model.layers.46.mlp.experts.58.gate_proj", "model.language_model.layers.46.mlp.experts.59.gate_proj", "model.language_model.layers.46.mlp.experts.60.gate_proj", "model.language_model.layers.46.mlp.experts.61.gate_proj", "model.language_model.layers.46.mlp.experts.62.gate_proj", "model.language_model.layers.46.mlp.experts.63.gate_proj", "model.language_model.layers.46.mlp.experts.64.gate_proj", "model.language_model.layers.46.mlp.experts.65.gate_proj", "model.language_model.layers.46.mlp.experts.66.gate_proj", "model.language_model.layers.46.mlp.experts.67.gate_proj", "model.language_model.layers.46.mlp.experts.68.gate_proj", "model.language_model.layers.46.mlp.experts.69.gate_proj", "model.language_model.layers.46.mlp.experts.70.gate_proj", "model.language_model.layers.46.mlp.experts.71.gate_proj", "model.language_model.layers.46.mlp.experts.72.gate_proj", "model.language_model.layers.46.mlp.experts.73.gate_proj", "model.language_model.layers.46.mlp.experts.74.gate_proj", "model.language_model.layers.46.mlp.experts.75.gate_proj", "model.language_model.layers.46.mlp.experts.76.gate_proj", "model.language_model.layers.46.mlp.experts.77.gate_proj", "model.language_model.layers.46.mlp.experts.78.gate_proj", "model.language_model.layers.46.mlp.experts.79.gate_proj", "model.language_model.layers.46.mlp.experts.80.gate_proj", "model.language_model.layers.46.mlp.experts.81.gate_proj", "model.language_model.layers.46.mlp.experts.82.gate_proj", "model.language_model.layers.46.mlp.experts.83.gate_proj", "model.language_model.layers.46.mlp.experts.84.gate_proj", "model.language_model.layers.46.mlp.experts.85.gate_proj", "model.language_model.layers.46.mlp.experts.86.gate_proj", "model.language_model.layers.46.mlp.experts.87.gate_proj", "model.language_model.layers.46.mlp.experts.88.gate_proj", "model.language_model.layers.46.mlp.experts.89.gate_proj", "model.language_model.layers.46.mlp.experts.90.gate_proj", "model.language_model.layers.46.mlp.experts.91.gate_proj", "model.language_model.layers.46.mlp.experts.92.gate_proj", "model.language_model.layers.46.mlp.experts.93.gate_proj", "model.language_model.layers.46.mlp.experts.94.gate_proj", "model.language_model.layers.46.mlp.experts.95.gate_proj", "model.language_model.layers.46.mlp.experts.96.gate_proj", "model.language_model.layers.46.mlp.experts.97.gate_proj", "model.language_model.layers.46.mlp.experts.98.gate_proj", "model.language_model.layers.46.mlp.experts.99.gate_proj", "model.language_model.layers.46.mlp.experts.100.gate_proj", "model.language_model.layers.46.mlp.experts.101.gate_proj", "model.language_model.layers.46.mlp.experts.102.gate_proj", "model.language_model.layers.46.mlp.experts.103.gate_proj", "model.language_model.layers.46.mlp.experts.104.gate_proj", "model.language_model.layers.46.mlp.experts.105.gate_proj", "model.language_model.layers.46.mlp.experts.106.gate_proj", "model.language_model.layers.46.mlp.experts.107.gate_proj", "model.language_model.layers.46.mlp.experts.108.gate_proj", "model.language_model.layers.46.mlp.experts.109.gate_proj", "model.language_model.layers.46.mlp.experts.110.gate_proj", "model.language_model.layers.46.mlp.experts.111.gate_proj", "model.language_model.layers.46.mlp.experts.112.gate_proj", "model.language_model.layers.46.mlp.experts.113.gate_proj", "model.language_model.layers.46.mlp.experts.114.gate_proj", "model.language_model.layers.46.mlp.experts.115.gate_proj", "model.language_model.layers.46.mlp.experts.116.gate_proj", "model.language_model.layers.46.mlp.experts.117.gate_proj", "model.language_model.layers.46.mlp.experts.118.gate_proj", "model.language_model.layers.46.mlp.experts.119.gate_proj", "model.language_model.layers.46.mlp.experts.120.gate_proj", "model.language_model.layers.46.mlp.experts.121.gate_proj", "model.language_model.layers.46.mlp.experts.122.gate_proj", "model.language_model.layers.46.mlp.experts.123.gate_proj", "model.language_model.layers.46.mlp.experts.124.gate_proj", "model.language_model.layers.46.mlp.experts.125.gate_proj", "model.language_model.layers.46.mlp.experts.126.gate_proj", "model.language_model.layers.46.mlp.experts.127.gate_proj", "model.language_model.layers.46.mlp.experts.0.up_proj", "model.language_model.layers.46.mlp.experts.1.up_proj", "model.language_model.layers.46.mlp.experts.2.up_proj", "model.language_model.layers.46.mlp.experts.3.up_proj", "model.language_model.layers.46.mlp.experts.4.up_proj", "model.language_model.layers.46.mlp.experts.5.up_proj", "model.language_model.layers.46.mlp.experts.6.up_proj", "model.language_model.layers.46.mlp.experts.7.up_proj", "model.language_model.layers.46.mlp.experts.8.up_proj", "model.language_model.layers.46.mlp.experts.9.up_proj", "model.language_model.layers.46.mlp.experts.10.up_proj", "model.language_model.layers.46.mlp.experts.11.up_proj", "model.language_model.layers.46.mlp.experts.12.up_proj", "model.language_model.layers.46.mlp.experts.13.up_proj", "model.language_model.layers.46.mlp.experts.14.up_proj", "model.language_model.layers.46.mlp.experts.15.up_proj", "model.language_model.layers.46.mlp.experts.16.up_proj", "model.language_model.layers.46.mlp.experts.17.up_proj", "model.language_model.layers.46.mlp.experts.18.up_proj", "model.language_model.layers.46.mlp.experts.19.up_proj", "model.language_model.layers.46.mlp.experts.20.up_proj", "model.language_model.layers.46.mlp.experts.21.up_proj", "model.language_model.layers.46.mlp.experts.22.up_proj", "model.language_model.layers.46.mlp.experts.23.up_proj", "model.language_model.layers.46.mlp.experts.24.up_proj", "model.language_model.layers.46.mlp.experts.25.up_proj", "model.language_model.layers.46.mlp.experts.26.up_proj", "model.language_model.layers.46.mlp.experts.27.up_proj", "model.language_model.layers.46.mlp.experts.28.up_proj", "model.language_model.layers.46.mlp.experts.29.up_proj", "model.language_model.layers.46.mlp.experts.30.up_proj", "model.language_model.layers.46.mlp.experts.31.up_proj", "model.language_model.layers.46.mlp.experts.32.up_proj", "model.language_model.layers.46.mlp.experts.33.up_proj", "model.language_model.layers.46.mlp.experts.34.up_proj", "model.language_model.layers.46.mlp.experts.35.up_proj", "model.language_model.layers.46.mlp.experts.36.up_proj", "model.language_model.layers.46.mlp.experts.37.up_proj", "model.language_model.layers.46.mlp.experts.38.up_proj", "model.language_model.layers.46.mlp.experts.39.up_proj", "model.language_model.layers.46.mlp.experts.40.up_proj", "model.language_model.layers.46.mlp.experts.41.up_proj", "model.language_model.layers.46.mlp.experts.42.up_proj", "model.language_model.layers.46.mlp.experts.43.up_proj", "model.language_model.layers.46.mlp.experts.44.up_proj", "model.language_model.layers.46.mlp.experts.45.up_proj", "model.language_model.layers.46.mlp.experts.46.up_proj", "model.language_model.layers.46.mlp.experts.47.up_proj", "model.language_model.layers.46.mlp.experts.48.up_proj", "model.language_model.layers.46.mlp.experts.49.up_proj", "model.language_model.layers.46.mlp.experts.50.up_proj", "model.language_model.layers.46.mlp.experts.51.up_proj", "model.language_model.layers.46.mlp.experts.52.up_proj", "model.language_model.layers.46.mlp.experts.53.up_proj", "model.language_model.layers.46.mlp.experts.54.up_proj", "model.language_model.layers.46.mlp.experts.55.up_proj", "model.language_model.layers.46.mlp.experts.56.up_proj", "model.language_model.layers.46.mlp.experts.57.up_proj", "model.language_model.layers.46.mlp.experts.58.up_proj", "model.language_model.layers.46.mlp.experts.59.up_proj", "model.language_model.layers.46.mlp.experts.60.up_proj", "model.language_model.layers.46.mlp.experts.61.up_proj", "model.language_model.layers.46.mlp.experts.62.up_proj", "model.language_model.layers.46.mlp.experts.63.up_proj", "model.language_model.layers.46.mlp.experts.64.up_proj", "model.language_model.layers.46.mlp.experts.65.up_proj", "model.language_model.layers.46.mlp.experts.66.up_proj", "model.language_model.layers.46.mlp.experts.67.up_proj", "model.language_model.layers.46.mlp.experts.68.up_proj", "model.language_model.layers.46.mlp.experts.69.up_proj", "model.language_model.layers.46.mlp.experts.70.up_proj", "model.language_model.layers.46.mlp.experts.71.up_proj", "model.language_model.layers.46.mlp.experts.72.up_proj", "model.language_model.layers.46.mlp.experts.73.up_proj", "model.language_model.layers.46.mlp.experts.74.up_proj", "model.language_model.layers.46.mlp.experts.75.up_proj", "model.language_model.layers.46.mlp.experts.76.up_proj", "model.language_model.layers.46.mlp.experts.77.up_proj", "model.language_model.layers.46.mlp.experts.78.up_proj", "model.language_model.layers.46.mlp.experts.79.up_proj", "model.language_model.layers.46.mlp.experts.80.up_proj", "model.language_model.layers.46.mlp.experts.81.up_proj", "model.language_model.layers.46.mlp.experts.82.up_proj", "model.language_model.layers.46.mlp.experts.83.up_proj", "model.language_model.layers.46.mlp.experts.84.up_proj", "model.language_model.layers.46.mlp.experts.85.up_proj", "model.language_model.layers.46.mlp.experts.86.up_proj", "model.language_model.layers.46.mlp.experts.87.up_proj", "model.language_model.layers.46.mlp.experts.88.up_proj", "model.language_model.layers.46.mlp.experts.89.up_proj", "model.language_model.layers.46.mlp.experts.90.up_proj", "model.language_model.layers.46.mlp.experts.91.up_proj", "model.language_model.layers.46.mlp.experts.92.up_proj", "model.language_model.layers.46.mlp.experts.93.up_proj", "model.language_model.layers.46.mlp.experts.94.up_proj", "model.language_model.layers.46.mlp.experts.95.up_proj", "model.language_model.layers.46.mlp.experts.96.up_proj", "model.language_model.layers.46.mlp.experts.97.up_proj", "model.language_model.layers.46.mlp.experts.98.up_proj", "model.language_model.layers.46.mlp.experts.99.up_proj", "model.language_model.layers.46.mlp.experts.100.up_proj", "model.language_model.layers.46.mlp.experts.101.up_proj", "model.language_model.layers.46.mlp.experts.102.up_proj", "model.language_model.layers.46.mlp.experts.103.up_proj", "model.language_model.layers.46.mlp.experts.104.up_proj", "model.language_model.layers.46.mlp.experts.105.up_proj", "model.language_model.layers.46.mlp.experts.106.up_proj", "model.language_model.layers.46.mlp.experts.107.up_proj", "model.language_model.layers.46.mlp.experts.108.up_proj", "model.language_model.layers.46.mlp.experts.109.up_proj", "model.language_model.layers.46.mlp.experts.110.up_proj", "model.language_model.layers.46.mlp.experts.111.up_proj", "model.language_model.layers.46.mlp.experts.112.up_proj", "model.language_model.layers.46.mlp.experts.113.up_proj", "model.language_model.layers.46.mlp.experts.114.up_proj", "model.language_model.layers.46.mlp.experts.115.up_proj", "model.language_model.layers.46.mlp.experts.116.up_proj", "model.language_model.layers.46.mlp.experts.117.up_proj", "model.language_model.layers.46.mlp.experts.118.up_proj", "model.language_model.layers.46.mlp.experts.119.up_proj", "model.language_model.layers.46.mlp.experts.120.up_proj", "model.language_model.layers.46.mlp.experts.121.up_proj", "model.language_model.layers.46.mlp.experts.122.up_proj", "model.language_model.layers.46.mlp.experts.123.up_proj", "model.language_model.layers.46.mlp.experts.124.up_proj", "model.language_model.layers.46.mlp.experts.125.up_proj", "model.language_model.layers.46.mlp.experts.126.up_proj", "model.language_model.layers.46.mlp.experts.127.up_proj", "model.language_model.layers.46.mlp.experts.0.down_proj", "model.language_model.layers.46.mlp.experts.1.down_proj", "model.language_model.layers.46.mlp.experts.2.down_proj", "model.language_model.layers.46.mlp.experts.3.down_proj", "model.language_model.layers.46.mlp.experts.4.down_proj", "model.language_model.layers.46.mlp.experts.5.down_proj", "model.language_model.layers.46.mlp.experts.6.down_proj", "model.language_model.layers.46.mlp.experts.7.down_proj", "model.language_model.layers.46.mlp.experts.8.down_proj", "model.language_model.layers.46.mlp.experts.9.down_proj", "model.language_model.layers.46.mlp.experts.10.down_proj", "model.language_model.layers.46.mlp.experts.11.down_proj", "model.language_model.layers.46.mlp.experts.12.down_proj", "model.language_model.layers.46.mlp.experts.13.down_proj", "model.language_model.layers.46.mlp.experts.14.down_proj", "model.language_model.layers.46.mlp.experts.15.down_proj", "model.language_model.layers.46.mlp.experts.16.down_proj", "model.language_model.layers.46.mlp.experts.17.down_proj", "model.language_model.layers.46.mlp.experts.18.down_proj", "model.language_model.layers.46.mlp.experts.19.down_proj", "model.language_model.layers.46.mlp.experts.20.down_proj", "model.language_model.layers.46.mlp.experts.21.down_proj", "model.language_model.layers.46.mlp.experts.22.down_proj", "model.language_model.layers.46.mlp.experts.23.down_proj", "model.language_model.layers.46.mlp.experts.24.down_proj", "model.language_model.layers.46.mlp.experts.25.down_proj", "model.language_model.layers.46.mlp.experts.26.down_proj", "model.language_model.layers.46.mlp.experts.27.down_proj", "model.language_model.layers.46.mlp.experts.28.down_proj", "model.language_model.layers.46.mlp.experts.29.down_proj", "model.language_model.layers.46.mlp.experts.30.down_proj", "model.language_model.layers.46.mlp.experts.31.down_proj", "model.language_model.layers.46.mlp.experts.32.down_proj", "model.language_model.layers.46.mlp.experts.33.down_proj", "model.language_model.layers.46.mlp.experts.34.down_proj", "model.language_model.layers.46.mlp.experts.35.down_proj", "model.language_model.layers.46.mlp.experts.36.down_proj", "model.language_model.layers.46.mlp.experts.37.down_proj", "model.language_model.layers.46.mlp.experts.38.down_proj", "model.language_model.layers.46.mlp.experts.39.down_proj", "model.language_model.layers.46.mlp.experts.40.down_proj", "model.language_model.layers.46.mlp.experts.41.down_proj", "model.language_model.layers.46.mlp.experts.42.down_proj", "model.language_model.layers.46.mlp.experts.43.down_proj", "model.language_model.layers.46.mlp.experts.44.down_proj", "model.language_model.layers.46.mlp.experts.45.down_proj", "model.language_model.layers.46.mlp.experts.46.down_proj", "model.language_model.layers.46.mlp.experts.47.down_proj", "model.language_model.layers.46.mlp.experts.48.down_proj", "model.language_model.layers.46.mlp.experts.49.down_proj", "model.language_model.layers.46.mlp.experts.50.down_proj", "model.language_model.layers.46.mlp.experts.51.down_proj", "model.language_model.layers.46.mlp.experts.52.down_proj", "model.language_model.layers.46.mlp.experts.53.down_proj", "model.language_model.layers.46.mlp.experts.54.down_proj", "model.language_model.layers.46.mlp.experts.55.down_proj", "model.language_model.layers.46.mlp.experts.56.down_proj", "model.language_model.layers.46.mlp.experts.57.down_proj", "model.language_model.layers.46.mlp.experts.58.down_proj", "model.language_model.layers.46.mlp.experts.59.down_proj", "model.language_model.layers.46.mlp.experts.60.down_proj", "model.language_model.layers.46.mlp.experts.61.down_proj", "model.language_model.layers.46.mlp.experts.62.down_proj", "model.language_model.layers.46.mlp.experts.63.down_proj", "model.language_model.layers.46.mlp.experts.64.down_proj", "model.language_model.layers.46.mlp.experts.65.down_proj", "model.language_model.layers.46.mlp.experts.66.down_proj", "model.language_model.layers.46.mlp.experts.67.down_proj", "model.language_model.layers.46.mlp.experts.68.down_proj", "model.language_model.layers.46.mlp.experts.69.down_proj", "model.language_model.layers.46.mlp.experts.70.down_proj", "model.language_model.layers.46.mlp.experts.71.down_proj", "model.language_model.layers.46.mlp.experts.72.down_proj", "model.language_model.layers.46.mlp.experts.73.down_proj", "model.language_model.layers.46.mlp.experts.74.down_proj", "model.language_model.layers.46.mlp.experts.75.down_proj", "model.language_model.layers.46.mlp.experts.76.down_proj", "model.language_model.layers.46.mlp.experts.77.down_proj", "model.language_model.layers.46.mlp.experts.78.down_proj", "model.language_model.layers.46.mlp.experts.79.down_proj", "model.language_model.layers.46.mlp.experts.80.down_proj", "model.language_model.layers.46.mlp.experts.81.down_proj", "model.language_model.layers.46.mlp.experts.82.down_proj", "model.language_model.layers.46.mlp.experts.83.down_proj", "model.language_model.layers.46.mlp.experts.84.down_proj", "model.language_model.layers.46.mlp.experts.85.down_proj", "model.language_model.layers.46.mlp.experts.86.down_proj", "model.language_model.layers.46.mlp.experts.87.down_proj", "model.language_model.layers.46.mlp.experts.88.down_proj", "model.language_model.layers.46.mlp.experts.89.down_proj", "model.language_model.layers.46.mlp.experts.90.down_proj", "model.language_model.layers.46.mlp.experts.91.down_proj", "model.language_model.layers.46.mlp.experts.92.down_proj", "model.language_model.layers.46.mlp.experts.93.down_proj", "model.language_model.layers.46.mlp.experts.94.down_proj", "model.language_model.layers.46.mlp.experts.95.down_proj", "model.language_model.layers.46.mlp.experts.96.down_proj", "model.language_model.layers.46.mlp.experts.97.down_proj", "model.language_model.layers.46.mlp.experts.98.down_proj", "model.language_model.layers.46.mlp.experts.99.down_proj", "model.language_model.layers.46.mlp.experts.100.down_proj", "model.language_model.layers.46.mlp.experts.101.down_proj", "model.language_model.layers.46.mlp.experts.102.down_proj", "model.language_model.layers.46.mlp.experts.103.down_proj", "model.language_model.layers.46.mlp.experts.104.down_proj", "model.language_model.layers.46.mlp.experts.105.down_proj", "model.language_model.layers.46.mlp.experts.106.down_proj", "model.language_model.layers.46.mlp.experts.107.down_proj", "model.language_model.layers.46.mlp.experts.108.down_proj", "model.language_model.layers.46.mlp.experts.109.down_proj", "model.language_model.layers.46.mlp.experts.110.down_proj", "model.language_model.layers.46.mlp.experts.111.down_proj", "model.language_model.layers.46.mlp.experts.112.down_proj", "model.language_model.layers.46.mlp.experts.113.down_proj", "model.language_model.layers.46.mlp.experts.114.down_proj", "model.language_model.layers.46.mlp.experts.115.down_proj", "model.language_model.layers.46.mlp.experts.116.down_proj", "model.language_model.layers.46.mlp.experts.117.down_proj", "model.language_model.layers.46.mlp.experts.118.down_proj", "model.language_model.layers.46.mlp.experts.119.down_proj", "model.language_model.layers.46.mlp.experts.120.down_proj", "model.language_model.layers.46.mlp.experts.121.down_proj", "model.language_model.layers.46.mlp.experts.122.down_proj", "model.language_model.layers.46.mlp.experts.123.down_proj", "model.language_model.layers.46.mlp.experts.124.down_proj", "model.language_model.layers.46.mlp.experts.125.down_proj", "model.language_model.layers.46.mlp.experts.126.down_proj", "model.language_model.layers.46.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.004206080269068477, "dbits": 603979776 } ] }, { "idx": 94, "layers": [ "model.language_model.layers.47.self_attn.q_proj", "model.language_model.layers.47.self_attn.k_proj", "model.language_model.layers.47.self_attn.v_proj", "model.language_model.layers.47.self_attn.o_proj" ], "candidates": [ { "dkld": -0.0001106203533709077, "dbits": 18874368 } ] }, { "idx": 95, "layers": [ "model.language_model.layers.47.mlp.experts.0.gate_proj", "model.language_model.layers.47.mlp.experts.1.gate_proj", "model.language_model.layers.47.mlp.experts.2.gate_proj", "model.language_model.layers.47.mlp.experts.3.gate_proj", "model.language_model.layers.47.mlp.experts.4.gate_proj", "model.language_model.layers.47.mlp.experts.5.gate_proj", "model.language_model.layers.47.mlp.experts.6.gate_proj", "model.language_model.layers.47.mlp.experts.7.gate_proj", "model.language_model.layers.47.mlp.experts.8.gate_proj", "model.language_model.layers.47.mlp.experts.9.gate_proj", "model.language_model.layers.47.mlp.experts.10.gate_proj", "model.language_model.layers.47.mlp.experts.11.gate_proj", "model.language_model.layers.47.mlp.experts.12.gate_proj", "model.language_model.layers.47.mlp.experts.13.gate_proj", "model.language_model.layers.47.mlp.experts.14.gate_proj", "model.language_model.layers.47.mlp.experts.15.gate_proj", "model.language_model.layers.47.mlp.experts.16.gate_proj", "model.language_model.layers.47.mlp.experts.17.gate_proj", "model.language_model.layers.47.mlp.experts.18.gate_proj", "model.language_model.layers.47.mlp.experts.19.gate_proj", "model.language_model.layers.47.mlp.experts.20.gate_proj", "model.language_model.layers.47.mlp.experts.21.gate_proj", "model.language_model.layers.47.mlp.experts.22.gate_proj", "model.language_model.layers.47.mlp.experts.23.gate_proj", "model.language_model.layers.47.mlp.experts.24.gate_proj", "model.language_model.layers.47.mlp.experts.25.gate_proj", "model.language_model.layers.47.mlp.experts.26.gate_proj", "model.language_model.layers.47.mlp.experts.27.gate_proj", "model.language_model.layers.47.mlp.experts.28.gate_proj", "model.language_model.layers.47.mlp.experts.29.gate_proj", "model.language_model.layers.47.mlp.experts.30.gate_proj", "model.language_model.layers.47.mlp.experts.31.gate_proj", "model.language_model.layers.47.mlp.experts.32.gate_proj", "model.language_model.layers.47.mlp.experts.33.gate_proj", "model.language_model.layers.47.mlp.experts.34.gate_proj", "model.language_model.layers.47.mlp.experts.35.gate_proj", "model.language_model.layers.47.mlp.experts.36.gate_proj", "model.language_model.layers.47.mlp.experts.37.gate_proj", "model.language_model.layers.47.mlp.experts.38.gate_proj", "model.language_model.layers.47.mlp.experts.39.gate_proj", "model.language_model.layers.47.mlp.experts.40.gate_proj", "model.language_model.layers.47.mlp.experts.41.gate_proj", "model.language_model.layers.47.mlp.experts.42.gate_proj", "model.language_model.layers.47.mlp.experts.43.gate_proj", "model.language_model.layers.47.mlp.experts.44.gate_proj", "model.language_model.layers.47.mlp.experts.45.gate_proj", "model.language_model.layers.47.mlp.experts.46.gate_proj", "model.language_model.layers.47.mlp.experts.47.gate_proj", "model.language_model.layers.47.mlp.experts.48.gate_proj", "model.language_model.layers.47.mlp.experts.49.gate_proj", "model.language_model.layers.47.mlp.experts.50.gate_proj", "model.language_model.layers.47.mlp.experts.51.gate_proj", "model.language_model.layers.47.mlp.experts.52.gate_proj", "model.language_model.layers.47.mlp.experts.53.gate_proj", "model.language_model.layers.47.mlp.experts.54.gate_proj", "model.language_model.layers.47.mlp.experts.55.gate_proj", "model.language_model.layers.47.mlp.experts.56.gate_proj", "model.language_model.layers.47.mlp.experts.57.gate_proj", "model.language_model.layers.47.mlp.experts.58.gate_proj", "model.language_model.layers.47.mlp.experts.59.gate_proj", "model.language_model.layers.47.mlp.experts.60.gate_proj", "model.language_model.layers.47.mlp.experts.61.gate_proj", "model.language_model.layers.47.mlp.experts.62.gate_proj", "model.language_model.layers.47.mlp.experts.63.gate_proj", "model.language_model.layers.47.mlp.experts.64.gate_proj", "model.language_model.layers.47.mlp.experts.65.gate_proj", "model.language_model.layers.47.mlp.experts.66.gate_proj", "model.language_model.layers.47.mlp.experts.67.gate_proj", "model.language_model.layers.47.mlp.experts.68.gate_proj", "model.language_model.layers.47.mlp.experts.69.gate_proj", "model.language_model.layers.47.mlp.experts.70.gate_proj", "model.language_model.layers.47.mlp.experts.71.gate_proj", "model.language_model.layers.47.mlp.experts.72.gate_proj", "model.language_model.layers.47.mlp.experts.73.gate_proj", "model.language_model.layers.47.mlp.experts.74.gate_proj", "model.language_model.layers.47.mlp.experts.75.gate_proj", "model.language_model.layers.47.mlp.experts.76.gate_proj", "model.language_model.layers.47.mlp.experts.77.gate_proj", "model.language_model.layers.47.mlp.experts.78.gate_proj", "model.language_model.layers.47.mlp.experts.79.gate_proj", "model.language_model.layers.47.mlp.experts.80.gate_proj", "model.language_model.layers.47.mlp.experts.81.gate_proj", "model.language_model.layers.47.mlp.experts.82.gate_proj", "model.language_model.layers.47.mlp.experts.83.gate_proj", "model.language_model.layers.47.mlp.experts.84.gate_proj", "model.language_model.layers.47.mlp.experts.85.gate_proj", "model.language_model.layers.47.mlp.experts.86.gate_proj", "model.language_model.layers.47.mlp.experts.87.gate_proj", "model.language_model.layers.47.mlp.experts.88.gate_proj", "model.language_model.layers.47.mlp.experts.89.gate_proj", "model.language_model.layers.47.mlp.experts.90.gate_proj", "model.language_model.layers.47.mlp.experts.91.gate_proj", "model.language_model.layers.47.mlp.experts.92.gate_proj", "model.language_model.layers.47.mlp.experts.93.gate_proj", "model.language_model.layers.47.mlp.experts.94.gate_proj", "model.language_model.layers.47.mlp.experts.95.gate_proj", "model.language_model.layers.47.mlp.experts.96.gate_proj", "model.language_model.layers.47.mlp.experts.97.gate_proj", "model.language_model.layers.47.mlp.experts.98.gate_proj", "model.language_model.layers.47.mlp.experts.99.gate_proj", "model.language_model.layers.47.mlp.experts.100.gate_proj", "model.language_model.layers.47.mlp.experts.101.gate_proj", "model.language_model.layers.47.mlp.experts.102.gate_proj", "model.language_model.layers.47.mlp.experts.103.gate_proj", "model.language_model.layers.47.mlp.experts.104.gate_proj", "model.language_model.layers.47.mlp.experts.105.gate_proj", "model.language_model.layers.47.mlp.experts.106.gate_proj", "model.language_model.layers.47.mlp.experts.107.gate_proj", "model.language_model.layers.47.mlp.experts.108.gate_proj", "model.language_model.layers.47.mlp.experts.109.gate_proj", "model.language_model.layers.47.mlp.experts.110.gate_proj", "model.language_model.layers.47.mlp.experts.111.gate_proj", "model.language_model.layers.47.mlp.experts.112.gate_proj", "model.language_model.layers.47.mlp.experts.113.gate_proj", "model.language_model.layers.47.mlp.experts.114.gate_proj", "model.language_model.layers.47.mlp.experts.115.gate_proj", "model.language_model.layers.47.mlp.experts.116.gate_proj", "model.language_model.layers.47.mlp.experts.117.gate_proj", "model.language_model.layers.47.mlp.experts.118.gate_proj", "model.language_model.layers.47.mlp.experts.119.gate_proj", "model.language_model.layers.47.mlp.experts.120.gate_proj", "model.language_model.layers.47.mlp.experts.121.gate_proj", "model.language_model.layers.47.mlp.experts.122.gate_proj", "model.language_model.layers.47.mlp.experts.123.gate_proj", "model.language_model.layers.47.mlp.experts.124.gate_proj", "model.language_model.layers.47.mlp.experts.125.gate_proj", "model.language_model.layers.47.mlp.experts.126.gate_proj", "model.language_model.layers.47.mlp.experts.127.gate_proj", "model.language_model.layers.47.mlp.experts.0.up_proj", "model.language_model.layers.47.mlp.experts.1.up_proj", "model.language_model.layers.47.mlp.experts.2.up_proj", "model.language_model.layers.47.mlp.experts.3.up_proj", "model.language_model.layers.47.mlp.experts.4.up_proj", "model.language_model.layers.47.mlp.experts.5.up_proj", "model.language_model.layers.47.mlp.experts.6.up_proj", "model.language_model.layers.47.mlp.experts.7.up_proj", "model.language_model.layers.47.mlp.experts.8.up_proj", "model.language_model.layers.47.mlp.experts.9.up_proj", "model.language_model.layers.47.mlp.experts.10.up_proj", "model.language_model.layers.47.mlp.experts.11.up_proj", "model.language_model.layers.47.mlp.experts.12.up_proj", "model.language_model.layers.47.mlp.experts.13.up_proj", "model.language_model.layers.47.mlp.experts.14.up_proj", "model.language_model.layers.47.mlp.experts.15.up_proj", "model.language_model.layers.47.mlp.experts.16.up_proj", "model.language_model.layers.47.mlp.experts.17.up_proj", "model.language_model.layers.47.mlp.experts.18.up_proj", "model.language_model.layers.47.mlp.experts.19.up_proj", "model.language_model.layers.47.mlp.experts.20.up_proj", "model.language_model.layers.47.mlp.experts.21.up_proj", "model.language_model.layers.47.mlp.experts.22.up_proj", "model.language_model.layers.47.mlp.experts.23.up_proj", "model.language_model.layers.47.mlp.experts.24.up_proj", "model.language_model.layers.47.mlp.experts.25.up_proj", "model.language_model.layers.47.mlp.experts.26.up_proj", "model.language_model.layers.47.mlp.experts.27.up_proj", "model.language_model.layers.47.mlp.experts.28.up_proj", "model.language_model.layers.47.mlp.experts.29.up_proj", "model.language_model.layers.47.mlp.experts.30.up_proj", "model.language_model.layers.47.mlp.experts.31.up_proj", "model.language_model.layers.47.mlp.experts.32.up_proj", "model.language_model.layers.47.mlp.experts.33.up_proj", "model.language_model.layers.47.mlp.experts.34.up_proj", "model.language_model.layers.47.mlp.experts.35.up_proj", "model.language_model.layers.47.mlp.experts.36.up_proj", "model.language_model.layers.47.mlp.experts.37.up_proj", "model.language_model.layers.47.mlp.experts.38.up_proj", "model.language_model.layers.47.mlp.experts.39.up_proj", "model.language_model.layers.47.mlp.experts.40.up_proj", "model.language_model.layers.47.mlp.experts.41.up_proj", "model.language_model.layers.47.mlp.experts.42.up_proj", "model.language_model.layers.47.mlp.experts.43.up_proj", "model.language_model.layers.47.mlp.experts.44.up_proj", "model.language_model.layers.47.mlp.experts.45.up_proj", "model.language_model.layers.47.mlp.experts.46.up_proj", "model.language_model.layers.47.mlp.experts.47.up_proj", "model.language_model.layers.47.mlp.experts.48.up_proj", "model.language_model.layers.47.mlp.experts.49.up_proj", "model.language_model.layers.47.mlp.experts.50.up_proj", "model.language_model.layers.47.mlp.experts.51.up_proj", "model.language_model.layers.47.mlp.experts.52.up_proj", "model.language_model.layers.47.mlp.experts.53.up_proj", "model.language_model.layers.47.mlp.experts.54.up_proj", "model.language_model.layers.47.mlp.experts.55.up_proj", "model.language_model.layers.47.mlp.experts.56.up_proj", "model.language_model.layers.47.mlp.experts.57.up_proj", "model.language_model.layers.47.mlp.experts.58.up_proj", "model.language_model.layers.47.mlp.experts.59.up_proj", "model.language_model.layers.47.mlp.experts.60.up_proj", "model.language_model.layers.47.mlp.experts.61.up_proj", "model.language_model.layers.47.mlp.experts.62.up_proj", "model.language_model.layers.47.mlp.experts.63.up_proj", "model.language_model.layers.47.mlp.experts.64.up_proj", "model.language_model.layers.47.mlp.experts.65.up_proj", "model.language_model.layers.47.mlp.experts.66.up_proj", "model.language_model.layers.47.mlp.experts.67.up_proj", "model.language_model.layers.47.mlp.experts.68.up_proj", "model.language_model.layers.47.mlp.experts.69.up_proj", "model.language_model.layers.47.mlp.experts.70.up_proj", "model.language_model.layers.47.mlp.experts.71.up_proj", "model.language_model.layers.47.mlp.experts.72.up_proj", "model.language_model.layers.47.mlp.experts.73.up_proj", "model.language_model.layers.47.mlp.experts.74.up_proj", "model.language_model.layers.47.mlp.experts.75.up_proj", "model.language_model.layers.47.mlp.experts.76.up_proj", "model.language_model.layers.47.mlp.experts.77.up_proj", "model.language_model.layers.47.mlp.experts.78.up_proj", "model.language_model.layers.47.mlp.experts.79.up_proj", "model.language_model.layers.47.mlp.experts.80.up_proj", "model.language_model.layers.47.mlp.experts.81.up_proj", "model.language_model.layers.47.mlp.experts.82.up_proj", "model.language_model.layers.47.mlp.experts.83.up_proj", "model.language_model.layers.47.mlp.experts.84.up_proj", "model.language_model.layers.47.mlp.experts.85.up_proj", "model.language_model.layers.47.mlp.experts.86.up_proj", "model.language_model.layers.47.mlp.experts.87.up_proj", "model.language_model.layers.47.mlp.experts.88.up_proj", "model.language_model.layers.47.mlp.experts.89.up_proj", "model.language_model.layers.47.mlp.experts.90.up_proj", "model.language_model.layers.47.mlp.experts.91.up_proj", "model.language_model.layers.47.mlp.experts.92.up_proj", "model.language_model.layers.47.mlp.experts.93.up_proj", "model.language_model.layers.47.mlp.experts.94.up_proj", "model.language_model.layers.47.mlp.experts.95.up_proj", "model.language_model.layers.47.mlp.experts.96.up_proj", "model.language_model.layers.47.mlp.experts.97.up_proj", "model.language_model.layers.47.mlp.experts.98.up_proj", "model.language_model.layers.47.mlp.experts.99.up_proj", "model.language_model.layers.47.mlp.experts.100.up_proj", "model.language_model.layers.47.mlp.experts.101.up_proj", "model.language_model.layers.47.mlp.experts.102.up_proj", "model.language_model.layers.47.mlp.experts.103.up_proj", "model.language_model.layers.47.mlp.experts.104.up_proj", "model.language_model.layers.47.mlp.experts.105.up_proj", "model.language_model.layers.47.mlp.experts.106.up_proj", "model.language_model.layers.47.mlp.experts.107.up_proj", "model.language_model.layers.47.mlp.experts.108.up_proj", "model.language_model.layers.47.mlp.experts.109.up_proj", "model.language_model.layers.47.mlp.experts.110.up_proj", "model.language_model.layers.47.mlp.experts.111.up_proj", "model.language_model.layers.47.mlp.experts.112.up_proj", "model.language_model.layers.47.mlp.experts.113.up_proj", "model.language_model.layers.47.mlp.experts.114.up_proj", "model.language_model.layers.47.mlp.experts.115.up_proj", "model.language_model.layers.47.mlp.experts.116.up_proj", "model.language_model.layers.47.mlp.experts.117.up_proj", "model.language_model.layers.47.mlp.experts.118.up_proj", "model.language_model.layers.47.mlp.experts.119.up_proj", "model.language_model.layers.47.mlp.experts.120.up_proj", "model.language_model.layers.47.mlp.experts.121.up_proj", "model.language_model.layers.47.mlp.experts.122.up_proj", "model.language_model.layers.47.mlp.experts.123.up_proj", "model.language_model.layers.47.mlp.experts.124.up_proj", "model.language_model.layers.47.mlp.experts.125.up_proj", "model.language_model.layers.47.mlp.experts.126.up_proj", "model.language_model.layers.47.mlp.experts.127.up_proj", "model.language_model.layers.47.mlp.experts.0.down_proj", "model.language_model.layers.47.mlp.experts.1.down_proj", "model.language_model.layers.47.mlp.experts.2.down_proj", "model.language_model.layers.47.mlp.experts.3.down_proj", "model.language_model.layers.47.mlp.experts.4.down_proj", "model.language_model.layers.47.mlp.experts.5.down_proj", "model.language_model.layers.47.mlp.experts.6.down_proj", "model.language_model.layers.47.mlp.experts.7.down_proj", "model.language_model.layers.47.mlp.experts.8.down_proj", "model.language_model.layers.47.mlp.experts.9.down_proj", "model.language_model.layers.47.mlp.experts.10.down_proj", "model.language_model.layers.47.mlp.experts.11.down_proj", "model.language_model.layers.47.mlp.experts.12.down_proj", "model.language_model.layers.47.mlp.experts.13.down_proj", "model.language_model.layers.47.mlp.experts.14.down_proj", "model.language_model.layers.47.mlp.experts.15.down_proj", "model.language_model.layers.47.mlp.experts.16.down_proj", "model.language_model.layers.47.mlp.experts.17.down_proj", "model.language_model.layers.47.mlp.experts.18.down_proj", "model.language_model.layers.47.mlp.experts.19.down_proj", "model.language_model.layers.47.mlp.experts.20.down_proj", "model.language_model.layers.47.mlp.experts.21.down_proj", "model.language_model.layers.47.mlp.experts.22.down_proj", "model.language_model.layers.47.mlp.experts.23.down_proj", "model.language_model.layers.47.mlp.experts.24.down_proj", "model.language_model.layers.47.mlp.experts.25.down_proj", "model.language_model.layers.47.mlp.experts.26.down_proj", "model.language_model.layers.47.mlp.experts.27.down_proj", "model.language_model.layers.47.mlp.experts.28.down_proj", "model.language_model.layers.47.mlp.experts.29.down_proj", "model.language_model.layers.47.mlp.experts.30.down_proj", "model.language_model.layers.47.mlp.experts.31.down_proj", "model.language_model.layers.47.mlp.experts.32.down_proj", "model.language_model.layers.47.mlp.experts.33.down_proj", "model.language_model.layers.47.mlp.experts.34.down_proj", "model.language_model.layers.47.mlp.experts.35.down_proj", "model.language_model.layers.47.mlp.experts.36.down_proj", "model.language_model.layers.47.mlp.experts.37.down_proj", "model.language_model.layers.47.mlp.experts.38.down_proj", "model.language_model.layers.47.mlp.experts.39.down_proj", "model.language_model.layers.47.mlp.experts.40.down_proj", "model.language_model.layers.47.mlp.experts.41.down_proj", "model.language_model.layers.47.mlp.experts.42.down_proj", "model.language_model.layers.47.mlp.experts.43.down_proj", "model.language_model.layers.47.mlp.experts.44.down_proj", "model.language_model.layers.47.mlp.experts.45.down_proj", "model.language_model.layers.47.mlp.experts.46.down_proj", "model.language_model.layers.47.mlp.experts.47.down_proj", "model.language_model.layers.47.mlp.experts.48.down_proj", "model.language_model.layers.47.mlp.experts.49.down_proj", "model.language_model.layers.47.mlp.experts.50.down_proj", "model.language_model.layers.47.mlp.experts.51.down_proj", "model.language_model.layers.47.mlp.experts.52.down_proj", "model.language_model.layers.47.mlp.experts.53.down_proj", "model.language_model.layers.47.mlp.experts.54.down_proj", "model.language_model.layers.47.mlp.experts.55.down_proj", "model.language_model.layers.47.mlp.experts.56.down_proj", "model.language_model.layers.47.mlp.experts.57.down_proj", "model.language_model.layers.47.mlp.experts.58.down_proj", "model.language_model.layers.47.mlp.experts.59.down_proj", "model.language_model.layers.47.mlp.experts.60.down_proj", "model.language_model.layers.47.mlp.experts.61.down_proj", "model.language_model.layers.47.mlp.experts.62.down_proj", "model.language_model.layers.47.mlp.experts.63.down_proj", "model.language_model.layers.47.mlp.experts.64.down_proj", "model.language_model.layers.47.mlp.experts.65.down_proj", "model.language_model.layers.47.mlp.experts.66.down_proj", "model.language_model.layers.47.mlp.experts.67.down_proj", "model.language_model.layers.47.mlp.experts.68.down_proj", "model.language_model.layers.47.mlp.experts.69.down_proj", "model.language_model.layers.47.mlp.experts.70.down_proj", "model.language_model.layers.47.mlp.experts.71.down_proj", "model.language_model.layers.47.mlp.experts.72.down_proj", "model.language_model.layers.47.mlp.experts.73.down_proj", "model.language_model.layers.47.mlp.experts.74.down_proj", "model.language_model.layers.47.mlp.experts.75.down_proj", "model.language_model.layers.47.mlp.experts.76.down_proj", "model.language_model.layers.47.mlp.experts.77.down_proj", "model.language_model.layers.47.mlp.experts.78.down_proj", "model.language_model.layers.47.mlp.experts.79.down_proj", "model.language_model.layers.47.mlp.experts.80.down_proj", "model.language_model.layers.47.mlp.experts.81.down_proj", "model.language_model.layers.47.mlp.experts.82.down_proj", "model.language_model.layers.47.mlp.experts.83.down_proj", "model.language_model.layers.47.mlp.experts.84.down_proj", "model.language_model.layers.47.mlp.experts.85.down_proj", "model.language_model.layers.47.mlp.experts.86.down_proj", "model.language_model.layers.47.mlp.experts.87.down_proj", "model.language_model.layers.47.mlp.experts.88.down_proj", "model.language_model.layers.47.mlp.experts.89.down_proj", "model.language_model.layers.47.mlp.experts.90.down_proj", "model.language_model.layers.47.mlp.experts.91.down_proj", "model.language_model.layers.47.mlp.experts.92.down_proj", "model.language_model.layers.47.mlp.experts.93.down_proj", "model.language_model.layers.47.mlp.experts.94.down_proj", "model.language_model.layers.47.mlp.experts.95.down_proj", "model.language_model.layers.47.mlp.experts.96.down_proj", "model.language_model.layers.47.mlp.experts.97.down_proj", "model.language_model.layers.47.mlp.experts.98.down_proj", "model.language_model.layers.47.mlp.experts.99.down_proj", "model.language_model.layers.47.mlp.experts.100.down_proj", "model.language_model.layers.47.mlp.experts.101.down_proj", "model.language_model.layers.47.mlp.experts.102.down_proj", "model.language_model.layers.47.mlp.experts.103.down_proj", "model.language_model.layers.47.mlp.experts.104.down_proj", "model.language_model.layers.47.mlp.experts.105.down_proj", "model.language_model.layers.47.mlp.experts.106.down_proj", "model.language_model.layers.47.mlp.experts.107.down_proj", "model.language_model.layers.47.mlp.experts.108.down_proj", "model.language_model.layers.47.mlp.experts.109.down_proj", "model.language_model.layers.47.mlp.experts.110.down_proj", "model.language_model.layers.47.mlp.experts.111.down_proj", "model.language_model.layers.47.mlp.experts.112.down_proj", "model.language_model.layers.47.mlp.experts.113.down_proj", "model.language_model.layers.47.mlp.experts.114.down_proj", "model.language_model.layers.47.mlp.experts.115.down_proj", "model.language_model.layers.47.mlp.experts.116.down_proj", "model.language_model.layers.47.mlp.experts.117.down_proj", "model.language_model.layers.47.mlp.experts.118.down_proj", "model.language_model.layers.47.mlp.experts.119.down_proj", "model.language_model.layers.47.mlp.experts.120.down_proj", "model.language_model.layers.47.mlp.experts.121.down_proj", "model.language_model.layers.47.mlp.experts.122.down_proj", "model.language_model.layers.47.mlp.experts.123.down_proj", "model.language_model.layers.47.mlp.experts.124.down_proj", "model.language_model.layers.47.mlp.experts.125.down_proj", "model.language_model.layers.47.mlp.experts.126.down_proj", "model.language_model.layers.47.mlp.experts.127.down_proj" ], "candidates": [ { "dkld": -0.00102646769955754, "dbits": 603979776 } ] } ], "base_kld": 0.07461283262819052, "arch_string": "Qwen3VLMoeForConditionalGeneration" }