{ "add_skip_keys": true, "dequantize_fp32": true, "dynamic_loss_threshold": -1, "group_size": 0, "hadamard_group_size": 256, "is_integer": true, "is_training": false, "modules_dtype_dict": { "int5": [ "transformer_blocks.13.norm2.linear_2.weight", "transformer_blocks.7.norm2.linear_2.weight", "transformer_blocks.8.norm2.linear_2.weight" ], "uint4": [ "transformer_blocks.0.attn1.to_k.weight", "transformer_blocks.0.attn1.to_out.0.weight", "transformer_blocks.0.attn1.to_q.weight", "transformer_blocks.0.attn1.to_v.weight", "transformer_blocks.0.attn2.to_k.weight", "transformer_blocks.0.attn2.to_out.0.weight", "transformer_blocks.0.attn2.to_q.weight", "transformer_blocks.0.attn2.to_v.weight", "transformer_blocks.0.ff.net.0.proj.weight", "transformer_blocks.0.ff.net.2.weight", "transformer_blocks.1.attn1.to_k.weight", "transformer_blocks.1.attn1.to_out.0.weight", "transformer_blocks.1.attn1.to_q.weight", "transformer_blocks.1.attn1.to_v.weight", "transformer_blocks.1.attn2.to_k.weight", "transformer_blocks.1.attn2.to_out.0.weight", "transformer_blocks.1.attn2.to_q.weight", "transformer_blocks.1.attn2.to_v.weight", "transformer_blocks.1.ff.net.0.proj.weight", "transformer_blocks.1.ff.net.2.weight", "transformer_blocks.1.norm1.linear_1.weight", "transformer_blocks.1.norm1.linear_2.weight", "transformer_blocks.1.norm2.linear_1.weight", "transformer_blocks.1.norm2.linear_2.weight", "transformer_blocks.1.norm3.linear_1.weight", "transformer_blocks.1.norm3.linear_2.weight", "transformer_blocks.10.attn1.to_k.weight", "transformer_blocks.10.attn1.to_out.0.weight", "transformer_blocks.10.attn1.to_q.weight", "transformer_blocks.10.attn1.to_v.weight", "transformer_blocks.10.attn2.to_k.weight", "transformer_blocks.10.attn2.to_out.0.weight", "transformer_blocks.10.attn2.to_q.weight", "transformer_blocks.10.attn2.to_v.weight", "transformer_blocks.10.ff.net.0.proj.weight", "transformer_blocks.10.ff.net.2.weight", "transformer_blocks.10.norm1.linear_1.weight", "transformer_blocks.10.norm1.linear_2.weight", "transformer_blocks.10.norm2.linear_1.weight", "transformer_blocks.10.norm2.linear_2.weight", "transformer_blocks.10.norm3.linear_1.weight", "transformer_blocks.10.norm3.linear_2.weight", "transformer_blocks.11.attn1.to_k.weight", "transformer_blocks.11.attn1.to_out.0.weight", "transformer_blocks.11.attn1.to_q.weight", "transformer_blocks.11.attn1.to_v.weight", "transformer_blocks.11.attn2.to_k.weight", "transformer_blocks.11.attn2.to_out.0.weight", "transformer_blocks.11.attn2.to_q.weight", "transformer_blocks.11.attn2.to_v.weight", "transformer_blocks.11.ff.net.0.proj.weight", "transformer_blocks.11.ff.net.2.weight", "transformer_blocks.11.norm1.linear_1.weight", "transformer_blocks.11.norm1.linear_2.weight", "transformer_blocks.11.norm2.linear_1.weight", "transformer_blocks.11.norm2.linear_2.weight", "transformer_blocks.11.norm3.linear_1.weight", "transformer_blocks.11.norm3.linear_2.weight", "transformer_blocks.12.attn1.to_k.weight", "transformer_blocks.12.attn1.to_out.0.weight", "transformer_blocks.12.attn1.to_q.weight", "transformer_blocks.12.attn1.to_v.weight", "transformer_blocks.12.attn2.to_k.weight", "transformer_blocks.12.attn2.to_out.0.weight", "transformer_blocks.12.attn2.to_q.weight", "transformer_blocks.12.attn2.to_v.weight", "transformer_blocks.12.ff.net.0.proj.weight", "transformer_blocks.12.ff.net.2.weight", "transformer_blocks.12.norm1.linear_1.weight", "transformer_blocks.12.norm1.linear_2.weight", "transformer_blocks.12.norm2.linear_1.weight", "transformer_blocks.12.norm2.linear_2.weight", "transformer_blocks.12.norm3.linear_1.weight", "transformer_blocks.12.norm3.linear_2.weight", "transformer_blocks.13.attn1.to_k.weight", "transformer_blocks.13.attn1.to_out.0.weight", "transformer_blocks.13.attn1.to_q.weight", "transformer_blocks.13.attn1.to_v.weight", "transformer_blocks.13.attn2.to_k.weight", "transformer_blocks.13.attn2.to_out.0.weight", "transformer_blocks.13.attn2.to_q.weight", "transformer_blocks.13.attn2.to_v.weight", "transformer_blocks.13.ff.net.0.proj.weight", "transformer_blocks.13.ff.net.2.weight", "transformer_blocks.13.norm1.linear_1.weight", "transformer_blocks.13.norm1.linear_2.weight", "transformer_blocks.13.norm2.linear_1.weight", "transformer_blocks.13.norm3.linear_1.weight", "transformer_blocks.13.norm3.linear_2.weight", "transformer_blocks.14.attn1.to_k.weight", "transformer_blocks.14.attn1.to_out.0.weight", "transformer_blocks.14.attn1.to_q.weight", "transformer_blocks.14.attn1.to_v.weight", "transformer_blocks.14.attn2.to_k.weight", "transformer_blocks.14.attn2.to_out.0.weight", "transformer_blocks.14.attn2.to_q.weight", "transformer_blocks.14.attn2.to_v.weight", "transformer_blocks.14.ff.net.0.proj.weight", "transformer_blocks.14.ff.net.2.weight", "transformer_blocks.14.norm1.linear_1.weight", "transformer_blocks.14.norm1.linear_2.weight", "transformer_blocks.14.norm2.linear_1.weight", "transformer_blocks.14.norm2.linear_2.weight", "transformer_blocks.14.norm3.linear_1.weight", "transformer_blocks.14.norm3.linear_2.weight", "transformer_blocks.15.attn1.to_k.weight", "transformer_blocks.15.attn1.to_out.0.weight", "transformer_blocks.15.attn1.to_q.weight", "transformer_blocks.15.attn1.to_v.weight", "transformer_blocks.15.attn2.to_k.weight", "transformer_blocks.15.attn2.to_out.0.weight", "transformer_blocks.15.attn2.to_q.weight", "transformer_blocks.15.attn2.to_v.weight", "transformer_blocks.15.ff.net.0.proj.weight", "transformer_blocks.15.ff.net.2.weight", "transformer_blocks.15.norm1.linear_1.weight", "transformer_blocks.15.norm1.linear_2.weight", "transformer_blocks.15.norm2.linear_1.weight", "transformer_blocks.15.norm2.linear_2.weight", "transformer_blocks.15.norm3.linear_1.weight", "transformer_blocks.15.norm3.linear_2.weight", "transformer_blocks.16.attn1.to_k.weight", "transformer_blocks.16.attn1.to_out.0.weight", "transformer_blocks.16.attn1.to_q.weight", "transformer_blocks.16.attn1.to_v.weight", "transformer_blocks.16.attn2.to_k.weight", "transformer_blocks.16.attn2.to_out.0.weight", "transformer_blocks.16.attn2.to_q.weight", "transformer_blocks.16.attn2.to_v.weight", "transformer_blocks.16.ff.net.0.proj.weight", "transformer_blocks.16.ff.net.2.weight", "transformer_blocks.16.norm1.linear_1.weight", "transformer_blocks.16.norm1.linear_2.weight", "transformer_blocks.16.norm2.linear_1.weight", "transformer_blocks.16.norm2.linear_2.weight", "transformer_blocks.16.norm3.linear_1.weight", "transformer_blocks.16.norm3.linear_2.weight", "transformer_blocks.17.attn1.to_k.weight", "transformer_blocks.17.attn1.to_out.0.weight", "transformer_blocks.17.attn1.to_q.weight", "transformer_blocks.17.attn1.to_v.weight", "transformer_blocks.17.attn2.to_k.weight", "transformer_blocks.17.attn2.to_out.0.weight", "transformer_blocks.17.attn2.to_q.weight", "transformer_blocks.17.attn2.to_v.weight", "transformer_blocks.17.ff.net.0.proj.weight", "transformer_blocks.17.ff.net.2.weight", "transformer_blocks.17.norm1.linear_1.weight", "transformer_blocks.17.norm1.linear_2.weight", "transformer_blocks.17.norm2.linear_1.weight", "transformer_blocks.17.norm2.linear_2.weight", "transformer_blocks.17.norm3.linear_1.weight", "transformer_blocks.17.norm3.linear_2.weight", "transformer_blocks.18.attn1.to_k.weight", "transformer_blocks.18.attn1.to_out.0.weight", "transformer_blocks.18.attn1.to_q.weight", "transformer_blocks.18.attn1.to_v.weight", "transformer_blocks.18.attn2.to_k.weight", "transformer_blocks.18.attn2.to_out.0.weight", "transformer_blocks.18.attn2.to_q.weight", "transformer_blocks.18.attn2.to_v.weight", "transformer_blocks.18.ff.net.0.proj.weight", "transformer_blocks.18.ff.net.2.weight", "transformer_blocks.18.norm1.linear_1.weight", "transformer_blocks.18.norm1.linear_2.weight", "transformer_blocks.18.norm2.linear_1.weight", "transformer_blocks.18.norm2.linear_2.weight", "transformer_blocks.18.norm3.linear_1.weight", "transformer_blocks.18.norm3.linear_2.weight", "transformer_blocks.19.attn1.to_k.weight", "transformer_blocks.19.attn1.to_out.0.weight", "transformer_blocks.19.attn1.to_q.weight", "transformer_blocks.19.attn1.to_v.weight", "transformer_blocks.19.attn2.to_k.weight", "transformer_blocks.19.attn2.to_out.0.weight", "transformer_blocks.19.attn2.to_q.weight", "transformer_blocks.19.attn2.to_v.weight", "transformer_blocks.19.ff.net.0.proj.weight", "transformer_blocks.19.ff.net.2.weight", "transformer_blocks.19.norm1.linear_1.weight", "transformer_blocks.19.norm1.linear_2.weight", "transformer_blocks.19.norm2.linear_1.weight", "transformer_blocks.19.norm2.linear_2.weight", "transformer_blocks.19.norm3.linear_1.weight", "transformer_blocks.19.norm3.linear_2.weight", "transformer_blocks.2.attn1.to_k.weight", "transformer_blocks.2.attn1.to_out.0.weight", "transformer_blocks.2.attn1.to_q.weight", "transformer_blocks.2.attn1.to_v.weight", "transformer_blocks.2.attn2.to_k.weight", "transformer_blocks.2.attn2.to_out.0.weight", "transformer_blocks.2.attn2.to_q.weight", "transformer_blocks.2.attn2.to_v.weight", "transformer_blocks.2.ff.net.0.proj.weight", "transformer_blocks.2.ff.net.2.weight", "transformer_blocks.2.norm1.linear_1.weight", "transformer_blocks.2.norm1.linear_2.weight", "transformer_blocks.2.norm2.linear_1.weight", "transformer_blocks.2.norm2.linear_2.weight", "transformer_blocks.2.norm3.linear_1.weight", "transformer_blocks.2.norm3.linear_2.weight", "transformer_blocks.20.attn1.to_k.weight", "transformer_blocks.20.attn1.to_out.0.weight", "transformer_blocks.20.attn1.to_q.weight", "transformer_blocks.20.attn1.to_v.weight", "transformer_blocks.20.attn2.to_k.weight", "transformer_blocks.20.attn2.to_out.0.weight", "transformer_blocks.20.attn2.to_q.weight", "transformer_blocks.20.attn2.to_v.weight", "transformer_blocks.20.ff.net.0.proj.weight", "transformer_blocks.20.ff.net.2.weight", "transformer_blocks.20.norm1.linear_1.weight", "transformer_blocks.20.norm1.linear_2.weight", "transformer_blocks.20.norm2.linear_1.weight", "transformer_blocks.20.norm2.linear_2.weight", "transformer_blocks.20.norm3.linear_1.weight", "transformer_blocks.20.norm3.linear_2.weight", "transformer_blocks.21.attn1.to_k.weight", "transformer_blocks.21.attn1.to_out.0.weight", "transformer_blocks.21.attn1.to_q.weight", "transformer_blocks.21.attn1.to_v.weight", "transformer_blocks.21.attn2.to_k.weight", "transformer_blocks.21.attn2.to_out.0.weight", "transformer_blocks.21.attn2.to_q.weight", "transformer_blocks.21.attn2.to_v.weight", "transformer_blocks.21.ff.net.0.proj.weight", "transformer_blocks.21.ff.net.2.weight", "transformer_blocks.21.norm1.linear_1.weight", "transformer_blocks.21.norm1.linear_2.weight", "transformer_blocks.21.norm2.linear_1.weight", "transformer_blocks.21.norm2.linear_2.weight", "transformer_blocks.21.norm3.linear_1.weight", "transformer_blocks.21.norm3.linear_2.weight", "transformer_blocks.22.attn1.to_k.weight", "transformer_blocks.22.attn1.to_out.0.weight", "transformer_blocks.22.attn1.to_q.weight", "transformer_blocks.22.attn1.to_v.weight", "transformer_blocks.22.attn2.to_k.weight", "transformer_blocks.22.attn2.to_out.0.weight", "transformer_blocks.22.attn2.to_q.weight", "transformer_blocks.22.attn2.to_v.weight", "transformer_blocks.22.ff.net.0.proj.weight", "transformer_blocks.22.ff.net.2.weight", "transformer_blocks.22.norm1.linear_1.weight", "transformer_blocks.22.norm1.linear_2.weight", "transformer_blocks.22.norm2.linear_1.weight", "transformer_blocks.22.norm2.linear_2.weight", "transformer_blocks.22.norm3.linear_1.weight", "transformer_blocks.22.norm3.linear_2.weight", "transformer_blocks.23.attn1.to_k.weight", "transformer_blocks.23.attn1.to_out.0.weight", "transformer_blocks.23.attn1.to_q.weight", "transformer_blocks.23.attn1.to_v.weight", "transformer_blocks.23.attn2.to_k.weight", "transformer_blocks.23.attn2.to_out.0.weight", "transformer_blocks.23.attn2.to_q.weight", "transformer_blocks.23.attn2.to_v.weight", "transformer_blocks.23.ff.net.0.proj.weight", "transformer_blocks.23.ff.net.2.weight", "transformer_blocks.23.norm1.linear_1.weight", "transformer_blocks.23.norm1.linear_2.weight", "transformer_blocks.23.norm2.linear_1.weight", "transformer_blocks.23.norm2.linear_2.weight", "transformer_blocks.23.norm3.linear_1.weight", "transformer_blocks.23.norm3.linear_2.weight", "transformer_blocks.24.attn1.to_k.weight", "transformer_blocks.24.attn1.to_out.0.weight", "transformer_blocks.24.attn1.to_q.weight", "transformer_blocks.24.attn1.to_v.weight", "transformer_blocks.24.attn2.to_k.weight", "transformer_blocks.24.attn2.to_out.0.weight", "transformer_blocks.24.attn2.to_q.weight", "transformer_blocks.24.attn2.to_v.weight", "transformer_blocks.24.ff.net.0.proj.weight", "transformer_blocks.24.ff.net.2.weight", "transformer_blocks.24.norm1.linear_1.weight", "transformer_blocks.24.norm1.linear_2.weight", "transformer_blocks.24.norm2.linear_1.weight", "transformer_blocks.24.norm2.linear_2.weight", "transformer_blocks.24.norm3.linear_1.weight", "transformer_blocks.24.norm3.linear_2.weight", "transformer_blocks.25.attn1.to_k.weight", "transformer_blocks.25.attn1.to_out.0.weight", "transformer_blocks.25.attn1.to_q.weight", "transformer_blocks.25.attn1.to_v.weight", "transformer_blocks.25.attn2.to_k.weight", "transformer_blocks.25.attn2.to_out.0.weight", "transformer_blocks.25.attn2.to_q.weight", "transformer_blocks.25.attn2.to_v.weight", "transformer_blocks.25.ff.net.0.proj.weight", "transformer_blocks.25.ff.net.2.weight", "transformer_blocks.25.norm1.linear_1.weight", "transformer_blocks.25.norm1.linear_2.weight", "transformer_blocks.25.norm2.linear_1.weight", "transformer_blocks.25.norm2.linear_2.weight", "transformer_blocks.25.norm3.linear_1.weight", "transformer_blocks.25.norm3.linear_2.weight", "transformer_blocks.26.attn1.to_k.weight", "transformer_blocks.26.attn1.to_out.0.weight", "transformer_blocks.26.attn1.to_q.weight", "transformer_blocks.26.attn1.to_v.weight", "transformer_blocks.26.attn2.to_k.weight", "transformer_blocks.26.attn2.to_out.0.weight", "transformer_blocks.26.attn2.to_q.weight", "transformer_blocks.26.attn2.to_v.weight", "transformer_blocks.26.ff.net.0.proj.weight", "transformer_blocks.26.ff.net.2.weight", "transformer_blocks.26.norm1.linear_1.weight", "transformer_blocks.26.norm1.linear_2.weight", "transformer_blocks.26.norm2.linear_1.weight", "transformer_blocks.26.norm2.linear_2.weight", "transformer_blocks.26.norm3.linear_1.weight", "transformer_blocks.26.norm3.linear_2.weight", "transformer_blocks.27.attn1.to_k.weight", "transformer_blocks.27.attn1.to_out.0.weight", "transformer_blocks.27.attn1.to_q.weight", "transformer_blocks.27.attn1.to_v.weight", "transformer_blocks.27.attn2.to_k.weight", "transformer_blocks.27.attn2.to_out.0.weight", "transformer_blocks.27.attn2.to_q.weight", "transformer_blocks.27.attn2.to_v.weight", "transformer_blocks.27.ff.net.0.proj.weight", "transformer_blocks.27.ff.net.2.weight", "transformer_blocks.27.norm1.linear_1.weight", "transformer_blocks.27.norm1.linear_2.weight", "transformer_blocks.27.norm2.linear_1.weight", "transformer_blocks.27.norm2.linear_2.weight", "transformer_blocks.27.norm3.linear_1.weight", "transformer_blocks.27.norm3.linear_2.weight", "transformer_blocks.3.attn1.to_k.weight", "transformer_blocks.3.attn1.to_out.0.weight", "transformer_blocks.3.attn1.to_q.weight", "transformer_blocks.3.attn1.to_v.weight", "transformer_blocks.3.attn2.to_k.weight", "transformer_blocks.3.attn2.to_out.0.weight", "transformer_blocks.3.attn2.to_q.weight", "transformer_blocks.3.attn2.to_v.weight", "transformer_blocks.3.ff.net.0.proj.weight", "transformer_blocks.3.ff.net.2.weight", "transformer_blocks.3.norm1.linear_1.weight", "transformer_blocks.3.norm1.linear_2.weight", "transformer_blocks.3.norm2.linear_1.weight", "transformer_blocks.3.norm2.linear_2.weight", "transformer_blocks.3.norm3.linear_1.weight", "transformer_blocks.3.norm3.linear_2.weight", "transformer_blocks.4.attn1.to_k.weight", "transformer_blocks.4.attn1.to_out.0.weight", "transformer_blocks.4.attn1.to_q.weight", "transformer_blocks.4.attn1.to_v.weight", "transformer_blocks.4.attn2.to_k.weight", "transformer_blocks.4.attn2.to_out.0.weight", "transformer_blocks.4.attn2.to_q.weight", "transformer_blocks.4.attn2.to_v.weight", "transformer_blocks.4.ff.net.0.proj.weight", "transformer_blocks.4.ff.net.2.weight", "transformer_blocks.4.norm1.linear_1.weight", "transformer_blocks.4.norm1.linear_2.weight", "transformer_blocks.4.norm2.linear_1.weight", "transformer_blocks.4.norm2.linear_2.weight", "transformer_blocks.4.norm3.linear_1.weight", "transformer_blocks.4.norm3.linear_2.weight", "transformer_blocks.5.attn1.to_k.weight", "transformer_blocks.5.attn1.to_out.0.weight", "transformer_blocks.5.attn1.to_q.weight", "transformer_blocks.5.attn1.to_v.weight", "transformer_blocks.5.attn2.to_k.weight", "transformer_blocks.5.attn2.to_out.0.weight", "transformer_blocks.5.attn2.to_q.weight", "transformer_blocks.5.attn2.to_v.weight", "transformer_blocks.5.ff.net.0.proj.weight", "transformer_blocks.5.ff.net.2.weight", "transformer_blocks.5.norm1.linear_1.weight", "transformer_blocks.5.norm1.linear_2.weight", "transformer_blocks.5.norm2.linear_1.weight", "transformer_blocks.5.norm2.linear_2.weight", "transformer_blocks.5.norm3.linear_1.weight", "transformer_blocks.5.norm3.linear_2.weight", "transformer_blocks.6.attn1.to_k.weight", "transformer_blocks.6.attn1.to_out.0.weight", "transformer_blocks.6.attn1.to_q.weight", "transformer_blocks.6.attn1.to_v.weight", "transformer_blocks.6.attn2.to_k.weight", "transformer_blocks.6.attn2.to_out.0.weight", "transformer_blocks.6.attn2.to_q.weight", "transformer_blocks.6.attn2.to_v.weight", "transformer_blocks.6.ff.net.0.proj.weight", "transformer_blocks.6.ff.net.2.weight", "transformer_blocks.6.norm1.linear_1.weight", "transformer_blocks.6.norm1.linear_2.weight", "transformer_blocks.6.norm2.linear_1.weight", "transformer_blocks.6.norm2.linear_2.weight", "transformer_blocks.6.norm3.linear_1.weight", "transformer_blocks.6.norm3.linear_2.weight", "transformer_blocks.7.attn1.to_k.weight", "transformer_blocks.7.attn1.to_out.0.weight", "transformer_blocks.7.attn1.to_q.weight", "transformer_blocks.7.attn1.to_v.weight", "transformer_blocks.7.attn2.to_k.weight", "transformer_blocks.7.attn2.to_out.0.weight", "transformer_blocks.7.attn2.to_q.weight", "transformer_blocks.7.attn2.to_v.weight", "transformer_blocks.7.ff.net.0.proj.weight", "transformer_blocks.7.ff.net.2.weight", "transformer_blocks.7.norm1.linear_1.weight", "transformer_blocks.7.norm1.linear_2.weight", "transformer_blocks.7.norm2.linear_1.weight", "transformer_blocks.7.norm3.linear_1.weight", "transformer_blocks.7.norm3.linear_2.weight", "transformer_blocks.8.attn1.to_k.weight", "transformer_blocks.8.attn1.to_out.0.weight", "transformer_blocks.8.attn1.to_q.weight", "transformer_blocks.8.attn1.to_v.weight", "transformer_blocks.8.attn2.to_k.weight", "transformer_blocks.8.attn2.to_out.0.weight", "transformer_blocks.8.attn2.to_q.weight", "transformer_blocks.8.attn2.to_v.weight", "transformer_blocks.8.ff.net.0.proj.weight", "transformer_blocks.8.ff.net.2.weight", "transformer_blocks.8.norm1.linear_1.weight", "transformer_blocks.8.norm1.linear_2.weight", "transformer_blocks.8.norm2.linear_1.weight", "transformer_blocks.8.norm3.linear_1.weight", "transformer_blocks.8.norm3.linear_2.weight", "transformer_blocks.9.attn1.to_k.weight", "transformer_blocks.9.attn1.to_out.0.weight", "transformer_blocks.9.attn1.to_q.weight", "transformer_blocks.9.attn1.to_v.weight", "transformer_blocks.9.attn2.to_k.weight", "transformer_blocks.9.attn2.to_out.0.weight", "transformer_blocks.9.attn2.to_q.weight", "transformer_blocks.9.attn2.to_v.weight", "transformer_blocks.9.ff.net.0.proj.weight", "transformer_blocks.9.ff.net.2.weight", "transformer_blocks.9.norm1.linear_1.weight", "transformer_blocks.9.norm1.linear_2.weight", "transformer_blocks.9.norm2.linear_1.weight", "transformer_blocks.9.norm2.linear_2.weight", "transformer_blocks.9.norm3.linear_1.weight", "transformer_blocks.9.norm3.linear_2.weight" ] }, "modules_quant_config": {}, "modules_to_not_convert": [ "time_embed", "proj_out", "patch_embed", "learnable_pos_embed", "crossattn_proj", "norm_out", "transformer_blocks.0.norm*", "transformer_blocks.0.attn1.norm_k.weight", "transformer_blocks.0.attn1.norm_q.weight", "transformer_blocks.0.attn2.norm_k.weight", "transformer_blocks.0.attn2.norm_q.weight", "transformer_blocks.1.attn1.norm_k.weight", "transformer_blocks.1.attn1.norm_q.weight", "transformer_blocks.1.attn2.norm_k.weight", "transformer_blocks.1.attn2.norm_q.weight", "transformer_blocks.10.attn1.norm_k.weight", "transformer_blocks.10.attn1.norm_q.weight", "transformer_blocks.10.attn2.norm_k.weight", "transformer_blocks.10.attn2.norm_q.weight", "transformer_blocks.11.attn1.norm_k.weight", "transformer_blocks.11.attn1.norm_q.weight", "transformer_blocks.11.attn2.norm_k.weight", "transformer_blocks.11.attn2.norm_q.weight", "transformer_blocks.12.attn1.norm_k.weight", "transformer_blocks.12.attn1.norm_q.weight", "transformer_blocks.12.attn2.norm_k.weight", "transformer_blocks.12.attn2.norm_q.weight", "transformer_blocks.13.attn1.norm_k.weight", "transformer_blocks.13.attn1.norm_q.weight", "transformer_blocks.13.attn2.norm_k.weight", "transformer_blocks.13.attn2.norm_q.weight", "transformer_blocks.14.attn1.norm_k.weight", "transformer_blocks.14.attn1.norm_q.weight", "transformer_blocks.14.attn2.norm_k.weight", "transformer_blocks.14.attn2.norm_q.weight", "transformer_blocks.15.attn1.norm_k.weight", "transformer_blocks.15.attn1.norm_q.weight", "transformer_blocks.15.attn2.norm_k.weight", "transformer_blocks.15.attn2.norm_q.weight", "transformer_blocks.16.attn1.norm_k.weight", "transformer_blocks.16.attn1.norm_q.weight", "transformer_blocks.16.attn2.norm_k.weight", "transformer_blocks.16.attn2.norm_q.weight", "transformer_blocks.17.attn1.norm_k.weight", "transformer_blocks.17.attn1.norm_q.weight", "transformer_blocks.17.attn2.norm_k.weight", "transformer_blocks.17.attn2.norm_q.weight", "transformer_blocks.18.attn1.norm_k.weight", "transformer_blocks.18.attn1.norm_q.weight", "transformer_blocks.18.attn2.norm_k.weight", "transformer_blocks.18.attn2.norm_q.weight", "transformer_blocks.19.attn1.norm_k.weight", "transformer_blocks.19.attn1.norm_q.weight", "transformer_blocks.19.attn2.norm_k.weight", "transformer_blocks.19.attn2.norm_q.weight", "transformer_blocks.2.attn1.norm_k.weight", "transformer_blocks.2.attn1.norm_q.weight", "transformer_blocks.2.attn2.norm_k.weight", "transformer_blocks.2.attn2.norm_q.weight", "transformer_blocks.20.attn1.norm_k.weight", "transformer_blocks.20.attn1.norm_q.weight", "transformer_blocks.20.attn2.norm_k.weight", "transformer_blocks.20.attn2.norm_q.weight", "transformer_blocks.21.attn1.norm_k.weight", "transformer_blocks.21.attn1.norm_q.weight", "transformer_blocks.21.attn2.norm_k.weight", "transformer_blocks.21.attn2.norm_q.weight", "transformer_blocks.22.attn1.norm_k.weight", "transformer_blocks.22.attn1.norm_q.weight", "transformer_blocks.22.attn2.norm_k.weight", "transformer_blocks.22.attn2.norm_q.weight", "transformer_blocks.23.attn1.norm_k.weight", "transformer_blocks.23.attn1.norm_q.weight", "transformer_blocks.23.attn2.norm_k.weight", "transformer_blocks.23.attn2.norm_q.weight", "transformer_blocks.24.attn1.norm_k.weight", "transformer_blocks.24.attn1.norm_q.weight", "transformer_blocks.24.attn2.norm_k.weight", "transformer_blocks.24.attn2.norm_q.weight", "transformer_blocks.25.attn1.norm_k.weight", "transformer_blocks.25.attn1.norm_q.weight", "transformer_blocks.25.attn2.norm_k.weight", "transformer_blocks.25.attn2.norm_q.weight", "transformer_blocks.26.attn1.norm_k.weight", "transformer_blocks.26.attn1.norm_q.weight", "transformer_blocks.26.attn2.norm_k.weight", "transformer_blocks.26.attn2.norm_q.weight", "transformer_blocks.27.attn1.norm_k.weight", "transformer_blocks.27.attn1.norm_q.weight", "transformer_blocks.27.attn2.norm_k.weight", "transformer_blocks.27.attn2.norm_q.weight", "transformer_blocks.3.attn1.norm_k.weight", "transformer_blocks.3.attn1.norm_q.weight", "transformer_blocks.3.attn2.norm_k.weight", "transformer_blocks.3.attn2.norm_q.weight", "transformer_blocks.4.attn1.norm_k.weight", "transformer_blocks.4.attn1.norm_q.weight", "transformer_blocks.4.attn2.norm_k.weight", "transformer_blocks.4.attn2.norm_q.weight", "transformer_blocks.5.attn1.norm_k.weight", "transformer_blocks.5.attn1.norm_q.weight", "transformer_blocks.5.attn2.norm_k.weight", "transformer_blocks.5.attn2.norm_q.weight", "transformer_blocks.6.attn1.norm_k.weight", "transformer_blocks.6.attn1.norm_q.weight", "transformer_blocks.6.attn2.norm_k.weight", "transformer_blocks.6.attn2.norm_q.weight", "transformer_blocks.7.attn1.norm_k.weight", "transformer_blocks.7.attn1.norm_q.weight", "transformer_blocks.7.attn2.norm_k.weight", "transformer_blocks.7.attn2.norm_q.weight", "transformer_blocks.8.attn1.norm_k.weight", "transformer_blocks.8.attn1.norm_q.weight", "transformer_blocks.8.attn2.norm_k.weight", "transformer_blocks.8.attn2.norm_q.weight", "transformer_blocks.9.attn1.norm_k.weight", "transformer_blocks.9.attn1.norm_q.weight", "transformer_blocks.9.attn2.norm_k.weight", "transformer_blocks.9.attn2.norm_q.weight" ], "modules_to_not_use_matmul": [], "non_blocking": true, "quant_conv": false, "quant_embedding": false, "quant_method": "sdnq", "quantization_device": "cuda", "quantized_matmul_dtype": null, "return_device": "cuda", "sdnq_version": "0.2.1", "svd_rank": 32, "svd_steps": 8, "use_dynamic_quantization": true, "use_grad_ckpt": true, "use_hadamard": true, "use_quantized_matmul": true, "use_quantized_matmul_conv": true, "use_static_quantization": true, "use_stochastic_rounding": false, "use_svd": false, "weights_dtype": "uint4" }