{ "method": "static_mixed_precision", "base_model": "nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16", "reference": "structural_rules", "target_bpw": 2.5, "achieved_bpw": 2.520806400894448, "n_high_bits": 290, "n_low_bits": 63, "threshold": 0.0, "per_layer": { "lm_head": { "bits": 4, "group_size": 64 }, "backbone.layers.87.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.87.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.87.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.87.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.87.mixer.switch_mlp.fc2": { "bits": 4, "group_size": 64 }, "backbone.layers.87.mixer.switch_mlp.fc1": { "bits": 4, "group_size": 64 }, "backbone.layers.86.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.86.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.85.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.85.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.85.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.85.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.85.mixer.switch_mlp.fc2": { "bits": 4, "group_size": 64 }, "backbone.layers.85.mixer.switch_mlp.fc1": { "bits": 4, "group_size": 64 }, "backbone.layers.84.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.84.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.83.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.83.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.83.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.83.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.83.mixer.switch_mlp.fc2": { "bits": 4, "group_size": 64 }, "backbone.layers.83.mixer.switch_mlp.fc1": { "bits": 4, "group_size": 64 }, "backbone.layers.82.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.82.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.81.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.81.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.81.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.81.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.81.mixer.switch_mlp.fc2": { "bits": 4, "group_size": 64 }, "backbone.layers.81.mixer.switch_mlp.fc1": { "bits": 4, "group_size": 64 }, "backbone.layers.80.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.80.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.79.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.79.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.79.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.79.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.79.mixer.switch_mlp.fc2": { "bits": 4, "group_size": 64 }, "backbone.layers.79.mixer.switch_mlp.fc1": { "bits": 4, "group_size": 64 }, "backbone.layers.78.mixer.o_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.78.mixer.v_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.78.mixer.k_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.78.mixer.q_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.77.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.77.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.76.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.76.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.76.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.76.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.76.mixer.switch_mlp.fc2": { "bits": 2, "group_size": 64 }, "backbone.layers.76.mixer.switch_mlp.fc1": { "bits": 2, "group_size": 64 }, "backbone.layers.75.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.75.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.74.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.74.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.74.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.74.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.74.mixer.switch_mlp.fc2": { "bits": 2, "group_size": 64 }, "backbone.layers.74.mixer.switch_mlp.fc1": { "bits": 2, "group_size": 64 }, "backbone.layers.73.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.73.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.72.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.72.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.72.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.72.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.72.mixer.switch_mlp.fc2": { "bits": 2, "group_size": 64 }, "backbone.layers.72.mixer.switch_mlp.fc1": { "bits": 2, "group_size": 64 }, "backbone.layers.71.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.71.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.70.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.70.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.70.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.70.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.70.mixer.switch_mlp.fc2": { "bits": 2, "group_size": 64 }, "backbone.layers.70.mixer.switch_mlp.fc1": { "bits": 2, "group_size": 64 }, "backbone.layers.69.mixer.o_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.69.mixer.v_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.69.mixer.k_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.69.mixer.q_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.68.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.68.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.67.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.67.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.67.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.67.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.67.mixer.switch_mlp.fc2": { "bits": 2, "group_size": 64 }, "backbone.layers.67.mixer.switch_mlp.fc1": { "bits": 2, "group_size": 64 }, "backbone.layers.66.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.66.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.65.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.65.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.65.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.65.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.65.mixer.switch_mlp.fc2": { "bits": 2, "group_size": 64 }, "backbone.layers.65.mixer.switch_mlp.fc1": { "bits": 2, "group_size": 64 }, "backbone.layers.64.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.64.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.63.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.63.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.63.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.63.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.63.mixer.switch_mlp.fc2": { "bits": 2, "group_size": 64 }, "backbone.layers.63.mixer.switch_mlp.fc1": { "bits": 2, "group_size": 64 }, "backbone.layers.62.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.62.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.61.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.61.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.61.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.61.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.61.mixer.switch_mlp.fc2": { "bits": 2, "group_size": 64 }, "backbone.layers.61.mixer.switch_mlp.fc1": { "bits": 2, "group_size": 64 }, "backbone.layers.60.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.60.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.59.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.59.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.59.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.59.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.59.mixer.switch_mlp.fc2": { "bits": 2, "group_size": 64 }, "backbone.layers.59.mixer.switch_mlp.fc1": { "bits": 2, "group_size": 64 }, "backbone.layers.58.mixer.o_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.58.mixer.v_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.58.mixer.k_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.58.mixer.q_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.57.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.57.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.56.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.56.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.56.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.56.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.56.mixer.switch_mlp.fc2": { "bits": 2, "group_size": 64 }, "backbone.layers.56.mixer.switch_mlp.fc1": { "bits": 2, "group_size": 64 }, "backbone.layers.55.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.55.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.54.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.54.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.54.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.54.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.54.mixer.switch_mlp.fc2": { "bits": 2, "group_size": 64 }, "backbone.layers.54.mixer.switch_mlp.fc1": { "bits": 2, "group_size": 64 }, "backbone.layers.53.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.53.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.52.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.52.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.52.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.52.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.52.mixer.switch_mlp.fc2": { "bits": 2, "group_size": 64 }, "backbone.layers.52.mixer.switch_mlp.fc1": { "bits": 2, "group_size": 64 }, "backbone.layers.51.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.51.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.50.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.50.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.50.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.50.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.50.mixer.switch_mlp.fc2": { "bits": 2, "group_size": 64 }, "backbone.layers.50.mixer.switch_mlp.fc1": { "bits": 2, "group_size": 64 }, "backbone.layers.49.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.49.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.48.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.48.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.48.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.48.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.48.mixer.switch_mlp.fc2": { "bits": 2, "group_size": 64 }, "backbone.layers.48.mixer.switch_mlp.fc1": { "bits": 2, "group_size": 64 }, "backbone.layers.47.mixer.o_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.47.mixer.v_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.47.mixer.k_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.47.mixer.q_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.46.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.46.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.45.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.45.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.45.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.45.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.45.mixer.switch_mlp.fc2": { "bits": 2, "group_size": 64 }, "backbone.layers.45.mixer.switch_mlp.fc1": { "bits": 2, "group_size": 64 }, "backbone.layers.44.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.44.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.43.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.43.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.43.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.43.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.43.mixer.switch_mlp.fc2": { "bits": 2, "group_size": 64 }, "backbone.layers.43.mixer.switch_mlp.fc1": { "bits": 2, "group_size": 64 }, "backbone.layers.42.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.42.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.41.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.41.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.41.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.41.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.41.mixer.switch_mlp.fc2": { "bits": 2, "group_size": 64 }, "backbone.layers.41.mixer.switch_mlp.fc1": { "bits": 2, "group_size": 64 }, "backbone.layers.40.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.40.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.39.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.39.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.39.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.39.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.39.mixer.switch_mlp.fc2": { "bits": 2, "group_size": 64 }, "backbone.layers.39.mixer.switch_mlp.fc1": { "bits": 2, "group_size": 64 }, "backbone.layers.38.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.38.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.37.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.37.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.37.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.37.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.37.mixer.switch_mlp.fc2": { "bits": 2, "group_size": 64 }, "backbone.layers.37.mixer.switch_mlp.fc1": { "bits": 2, "group_size": 64 }, "backbone.layers.36.mixer.o_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.36.mixer.v_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.36.mixer.k_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.36.mixer.q_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.35.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.35.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.34.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.34.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.34.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.34.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.34.mixer.switch_mlp.fc2": { "bits": 2, "group_size": 64 }, "backbone.layers.34.mixer.switch_mlp.fc1": { "bits": 2, "group_size": 64 }, "backbone.layers.33.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.33.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.32.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.32.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.32.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.32.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.32.mixer.switch_mlp.fc2": { "bits": 2, "group_size": 64 }, "backbone.layers.32.mixer.switch_mlp.fc1": { "bits": 2, "group_size": 64 }, "backbone.layers.31.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.31.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.30.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.30.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.30.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.30.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.30.mixer.switch_mlp.fc2": { "bits": 2, "group_size": 64 }, "backbone.layers.30.mixer.switch_mlp.fc1": { "bits": 2, "group_size": 64 }, "backbone.layers.29.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.29.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.28.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.28.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.28.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.28.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.28.mixer.switch_mlp.fc2": { "bits": 2, "group_size": 64 }, "backbone.layers.28.mixer.switch_mlp.fc1": { "bits": 2, "group_size": 64 }, "backbone.layers.27.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.27.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.26.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.26.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.26.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.26.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.26.mixer.switch_mlp.fc2": { "bits": 2, "group_size": 64 }, "backbone.layers.26.mixer.switch_mlp.fc1": { "bits": 2, "group_size": 64 }, "backbone.layers.25.mixer.o_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.25.mixer.v_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.25.mixer.k_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.25.mixer.q_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.24.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.24.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.23.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.23.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.23.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.23.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.23.mixer.switch_mlp.fc2": { "bits": 2, "group_size": 64 }, "backbone.layers.23.mixer.switch_mlp.fc1": { "bits": 2, "group_size": 64 }, "backbone.layers.22.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.22.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.21.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.21.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.21.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.21.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.21.mixer.switch_mlp.fc2": { "bits": 2, "group_size": 64 }, "backbone.layers.21.mixer.switch_mlp.fc1": { "bits": 2, "group_size": 64 }, "backbone.layers.20.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.20.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.19.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.19.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.19.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.19.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.19.mixer.switch_mlp.fc2": { "bits": 2, "group_size": 64 }, "backbone.layers.19.mixer.switch_mlp.fc1": { "bits": 2, "group_size": 64 }, "backbone.layers.18.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.18.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.17.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.17.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.17.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.17.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.17.mixer.switch_mlp.fc2": { "bits": 2, "group_size": 64 }, "backbone.layers.17.mixer.switch_mlp.fc1": { "bits": 2, "group_size": 64 }, "backbone.layers.16.mixer.o_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.16.mixer.v_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.16.mixer.k_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.16.mixer.q_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.15.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.15.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.14.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.14.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.14.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.14.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.14.mixer.switch_mlp.fc2": { "bits": 2, "group_size": 64 }, "backbone.layers.14.mixer.switch_mlp.fc1": { "bits": 2, "group_size": 64 }, "backbone.layers.13.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.13.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.12.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.12.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.12.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.12.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.12.mixer.switch_mlp.fc2": { "bits": 2, "group_size": 64 }, "backbone.layers.12.mixer.switch_mlp.fc1": { "bits": 2, "group_size": 64 }, "backbone.layers.11.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.11.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.10.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.10.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.10.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.10.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.10.mixer.switch_mlp.fc2": { "bits": 2, "group_size": 64 }, "backbone.layers.10.mixer.switch_mlp.fc1": { "bits": 2, "group_size": 64 }, "backbone.layers.9.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.9.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.8.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.8.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.8.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.8.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.8.mixer.switch_mlp.fc2": { "bits": 2, "group_size": 64 }, "backbone.layers.8.mixer.switch_mlp.fc1": { "bits": 4, "group_size": 64 }, "backbone.layers.7.mixer.o_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.7.mixer.v_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.7.mixer.k_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.7.mixer.q_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.6.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.6.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.5.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.5.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.5.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.5.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.5.mixer.switch_mlp.fc2": { "bits": 4, "group_size": 64 }, "backbone.layers.5.mixer.switch_mlp.fc1": { "bits": 4, "group_size": 64 }, "backbone.layers.4.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.4.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.3.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.3.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.3.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.3.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.3.mixer.switch_mlp.fc2": { "bits": 4, "group_size": 64 }, "backbone.layers.3.mixer.switch_mlp.fc1": { "bits": 4, "group_size": 64 }, "backbone.layers.2.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.2.mixer.in_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.1.mixer.fc2_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.1.mixer.fc1_latent_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.1.mixer.shared_experts.down_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.1.mixer.shared_experts.up_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.1.mixer.switch_mlp.fc2": { "bits": 4, "group_size": 64 }, "backbone.layers.1.mixer.switch_mlp.fc1": { "bits": 4, "group_size": 64 }, "backbone.layers.0.mixer.out_proj": { "bits": 4, "group_size": 64 }, "backbone.layers.0.mixer.in_proj": { "bits": 4, "group_size": 64 } } }