| { |
| "add_skip_keys": true, |
| "dequantize_fp32": false, |
| "dynamic_loss_threshold": -1, |
| "group_size": 0, |
| "hadamard_group_size": 256, |
| "is_integer": true, |
| "is_training": false, |
| "modules_dtype_dict": { |
| "uint4": [ |
| "blocks.0.attn.gate.weight", |
| "blocks.0.attn.wk.weight", |
| "blocks.0.attn.wo.weight", |
| "blocks.0.attn.wq.weight", |
| "blocks.0.attn.wv.weight", |
| "blocks.0.mlp.down.weight", |
| "blocks.0.mlp.gate.weight", |
| "blocks.0.mlp.up.weight", |
| "blocks.1.attn.gate.weight", |
| "blocks.1.attn.wk.weight", |
| "blocks.1.attn.wo.weight", |
| "blocks.1.attn.wq.weight", |
| "blocks.1.attn.wv.weight", |
| "blocks.1.mlp.down.weight", |
| "blocks.1.mlp.gate.weight", |
| "blocks.1.mlp.up.weight", |
| "blocks.2.attn.gate.weight", |
| "blocks.2.attn.wk.weight", |
| "blocks.2.attn.wo.weight", |
| "blocks.2.attn.wq.weight", |
| "blocks.2.attn.wv.weight", |
| "blocks.2.mlp.down.weight", |
| "blocks.2.mlp.gate.weight", |
| "blocks.2.mlp.up.weight", |
| "blocks.3.attn.gate.weight", |
| "blocks.3.attn.wk.weight", |
| "blocks.3.attn.wo.weight", |
| "blocks.3.attn.wq.weight", |
| "blocks.3.attn.wv.weight", |
| "blocks.3.mlp.down.weight", |
| "blocks.3.mlp.gate.weight", |
| "blocks.3.mlp.up.weight", |
| "blocks.4.attn.gate.weight", |
| "blocks.4.attn.wk.weight", |
| "blocks.4.attn.wo.weight", |
| "blocks.4.attn.wq.weight", |
| "blocks.4.attn.wv.weight", |
| "blocks.4.mlp.down.weight", |
| "blocks.4.mlp.gate.weight", |
| "blocks.4.mlp.up.weight", |
| "blocks.5.attn.gate.weight", |
| "blocks.5.attn.wk.weight", |
| "blocks.5.attn.wo.weight", |
| "blocks.5.attn.wq.weight", |
| "blocks.5.attn.wv.weight", |
| "blocks.5.mlp.gate.weight", |
| "blocks.10.attn.gate.weight", |
| "blocks.10.attn.wk.weight", |
| "blocks.10.attn.wo.weight", |
| "blocks.10.attn.wq.weight", |
| "blocks.10.attn.wv.weight", |
| "blocks.10.mlp.down.weight", |
| "blocks.10.mlp.gate.weight", |
| "blocks.10.mlp.up.weight", |
| "blocks.11.attn.gate.weight", |
| "blocks.11.attn.wk.weight", |
| "blocks.11.attn.wq.weight", |
| "blocks.11.attn.wv.weight", |
| "blocks.5.mlp.down.weight", |
| "blocks.5.mlp.up.weight", |
| "blocks.6.attn.gate.weight", |
| "blocks.6.attn.wk.weight", |
| "blocks.6.attn.wo.weight", |
| "blocks.6.attn.wq.weight", |
| "blocks.6.attn.wv.weight", |
| "blocks.6.mlp.down.weight", |
| "blocks.6.mlp.gate.weight", |
| "blocks.6.mlp.up.weight", |
| "blocks.7.attn.gate.weight", |
| "blocks.7.attn.wk.weight", |
| "blocks.7.attn.wo.weight", |
| "blocks.7.attn.wq.weight", |
| "blocks.7.attn.wv.weight", |
| "blocks.7.mlp.down.weight", |
| "blocks.7.mlp.gate.weight", |
| "blocks.7.mlp.up.weight", |
| "blocks.8.attn.gate.weight", |
| "blocks.8.attn.wk.weight", |
| "blocks.8.attn.wo.weight", |
| "blocks.8.attn.wq.weight", |
| "blocks.8.attn.wv.weight", |
| "blocks.8.mlp.down.weight", |
| "blocks.8.mlp.gate.weight", |
| "blocks.8.mlp.up.weight", |
| "blocks.9.attn.gate.weight", |
| "blocks.9.attn.wk.weight", |
| "blocks.9.attn.wo.weight", |
| "blocks.9.attn.wq.weight", |
| "blocks.9.attn.wv.weight", |
| "blocks.9.mlp.down.weight", |
| "blocks.9.mlp.gate.weight", |
| "blocks.9.mlp.up.weight", |
| "blocks.11.attn.wo.weight", |
| "blocks.11.mlp.down.weight", |
| "blocks.11.mlp.gate.weight", |
| "blocks.11.mlp.up.weight", |
| "blocks.12.attn.gate.weight", |
| "blocks.12.attn.wk.weight", |
| "blocks.12.attn.wo.weight", |
| "blocks.12.attn.wq.weight", |
| "blocks.12.attn.wv.weight", |
| "blocks.12.mlp.down.weight", |
| "blocks.12.mlp.gate.weight", |
| "blocks.12.mlp.up.weight", |
| "blocks.13.attn.gate.weight", |
| "blocks.13.attn.wk.weight", |
| "blocks.13.attn.wo.weight", |
| "blocks.13.attn.wq.weight", |
| "blocks.13.attn.wv.weight", |
| "blocks.13.mlp.down.weight", |
| "blocks.13.mlp.gate.weight", |
| "blocks.13.mlp.up.weight", |
| "blocks.14.attn.gate.weight", |
| "blocks.14.attn.wk.weight", |
| "blocks.14.attn.wo.weight", |
| "blocks.14.attn.wq.weight", |
| "blocks.14.attn.wv.weight", |
| "blocks.14.mlp.down.weight", |
| "blocks.14.mlp.gate.weight", |
| "blocks.14.mlp.up.weight", |
| "blocks.15.attn.gate.weight", |
| "blocks.15.attn.wk.weight", |
| "blocks.15.attn.wo.weight", |
| "blocks.15.attn.wq.weight", |
| "blocks.15.attn.wv.weight", |
| "blocks.15.mlp.down.weight", |
| "blocks.15.mlp.gate.weight", |
| "blocks.15.mlp.up.weight", |
| "blocks.16.attn.gate.weight", |
| "blocks.16.attn.wk.weight", |
| "blocks.16.attn.wo.weight", |
| "blocks.16.attn.wq.weight", |
| "blocks.16.attn.wv.weight", |
| "blocks.16.mlp.gate.weight", |
| "blocks.16.mlp.up.weight", |
| "blocks.16.mlp.down.weight", |
| "blocks.17.attn.gate.weight", |
| "blocks.17.attn.wk.weight", |
| "blocks.17.attn.wo.weight", |
| "blocks.17.attn.wq.weight", |
| "blocks.17.attn.wv.weight", |
| "blocks.17.mlp.down.weight", |
| "blocks.17.mlp.gate.weight", |
| "blocks.17.mlp.up.weight", |
| "blocks.18.attn.gate.weight", |
| "blocks.18.attn.wk.weight", |
| "blocks.18.attn.wo.weight", |
| "blocks.18.attn.wq.weight", |
| "blocks.18.attn.wv.weight", |
| "blocks.18.mlp.down.weight", |
| "blocks.18.mlp.gate.weight", |
| "blocks.18.mlp.up.weight", |
| "blocks.19.attn.gate.weight", |
| "blocks.19.attn.wk.weight", |
| "blocks.19.attn.wo.weight", |
| "blocks.19.attn.wq.weight", |
| "blocks.19.attn.wv.weight", |
| "blocks.19.mlp.down.weight", |
| "blocks.19.mlp.gate.weight", |
| "blocks.19.mlp.up.weight", |
| "blocks.20.attn.gate.weight", |
| "blocks.20.attn.wk.weight", |
| "blocks.20.attn.wo.weight", |
| "blocks.20.attn.wq.weight", |
| "blocks.20.attn.wv.weight", |
| "blocks.20.mlp.down.weight", |
| "blocks.20.mlp.gate.weight", |
| "blocks.20.mlp.up.weight", |
| "blocks.21.attn.gate.weight", |
| "blocks.21.attn.wk.weight", |
| "blocks.21.attn.wo.weight", |
| "blocks.21.attn.wq.weight", |
| "blocks.21.attn.wv.weight", |
| "blocks.21.mlp.down.weight", |
| "blocks.21.mlp.gate.weight", |
| "blocks.21.mlp.up.weight", |
| "blocks.22.attn.gate.weight", |
| "blocks.22.attn.wk.weight", |
| "blocks.22.attn.wo.weight", |
| "blocks.22.attn.wq.weight", |
| "blocks.22.attn.wv.weight", |
| "blocks.22.mlp.down.weight", |
| "blocks.22.mlp.gate.weight", |
| "blocks.22.mlp.up.weight", |
| "blocks.23.attn.gate.weight", |
| "blocks.23.attn.wk.weight", |
| "blocks.23.attn.wo.weight", |
| "blocks.23.attn.wq.weight", |
| "blocks.23.attn.wv.weight", |
| "blocks.23.mlp.down.weight", |
| "blocks.23.mlp.gate.weight", |
| "blocks.23.mlp.up.weight", |
| "blocks.24.attn.gate.weight", |
| "blocks.24.attn.wk.weight", |
| "blocks.24.attn.wo.weight", |
| "blocks.24.attn.wq.weight", |
| "blocks.24.attn.wv.weight", |
| "blocks.24.mlp.down.weight", |
| "blocks.24.mlp.gate.weight", |
| "blocks.24.mlp.up.weight", |
| "blocks.25.attn.gate.weight", |
| "blocks.25.attn.wk.weight", |
| "blocks.25.attn.wo.weight", |
| "blocks.25.attn.wq.weight", |
| "blocks.25.attn.wv.weight", |
| "blocks.25.mlp.down.weight", |
| "blocks.25.mlp.gate.weight", |
| "blocks.25.mlp.up.weight", |
| "blocks.26.attn.gate.weight", |
| "blocks.26.attn.wk.weight", |
| "blocks.26.attn.wo.weight", |
| "blocks.26.attn.wq.weight", |
| "blocks.26.attn.wv.weight", |
| "blocks.26.mlp.down.weight", |
| "blocks.26.mlp.gate.weight", |
| "blocks.26.mlp.up.weight", |
| "blocks.27.attn.gate.weight", |
| "blocks.27.attn.wk.weight", |
| "blocks.27.attn.wo.weight", |
| "blocks.27.attn.wq.weight", |
| "blocks.27.attn.wv.weight", |
| "blocks.27.mlp.down.weight", |
| "blocks.27.mlp.gate.weight", |
| "blocks.27.mlp.up.weight", |
| "txtmlp.1.weight", |
| "txtmlp.3.weight", |
| "txtfusion.layerwise_blocks.0.attn.gate.weight", |
| "txtfusion.layerwise_blocks.0.attn.wk.weight", |
| "txtfusion.layerwise_blocks.0.attn.wo.weight", |
| "txtfusion.layerwise_blocks.0.attn.wq.weight", |
| "txtfusion.layerwise_blocks.0.attn.wv.weight", |
| "txtfusion.layerwise_blocks.0.mlp.down.weight", |
| "txtfusion.layerwise_blocks.0.mlp.gate.weight", |
| "txtfusion.layerwise_blocks.0.mlp.up.weight", |
| "txtfusion.layerwise_blocks.1.attn.gate.weight", |
| "txtfusion.layerwise_blocks.1.attn.wk.weight", |
| "txtfusion.layerwise_blocks.1.attn.wo.weight", |
| "txtfusion.layerwise_blocks.1.attn.wq.weight", |
| "txtfusion.layerwise_blocks.1.attn.wv.weight", |
| "txtfusion.layerwise_blocks.1.mlp.down.weight", |
| "txtfusion.layerwise_blocks.1.mlp.gate.weight", |
| "txtfusion.layerwise_blocks.1.mlp.up.weight", |
| "txtfusion.refiner_blocks.0.attn.gate.weight", |
| "txtfusion.refiner_blocks.0.attn.wk.weight", |
| "txtfusion.refiner_blocks.0.attn.wo.weight", |
| "txtfusion.refiner_blocks.0.attn.wq.weight", |
| "txtfusion.refiner_blocks.0.attn.wv.weight", |
| "txtfusion.refiner_blocks.0.mlp.down.weight", |
| "txtfusion.refiner_blocks.0.mlp.gate.weight", |
| "txtfusion.refiner_blocks.0.mlp.up.weight", |
| "txtfusion.refiner_blocks.1.attn.gate.weight", |
| "txtfusion.refiner_blocks.1.attn.wk.weight", |
| "txtfusion.refiner_blocks.1.attn.wo.weight", |
| "txtfusion.refiner_blocks.1.attn.wq.weight", |
| "txtfusion.refiner_blocks.1.attn.wv.weight", |
| "txtfusion.refiner_blocks.1.mlp.down.weight", |
| "txtfusion.refiner_blocks.1.mlp.gate.weight", |
| "txtfusion.refiner_blocks.1.mlp.up.weight" |
| ] |
| }, |
| "modules_quant_config": {}, |
| "modules_to_not_convert": [ |
| ".txt_out", |
| ".time_embed", |
| ".vid_out", |
| ".vid_in", |
| ".txt_in", |
| ".img_in", |
| ".emb_out", |
| "first", |
| ".norm_out", |
| "multi_modal_projector", |
| ".img_out", |
| "tproj", |
| ".t_embedder", |
| ".emb_in", |
| ".final_layer", |
| "last", |
| "patch_embedding", |
| ".proj_out", |
| "time_text_embed", |
| "tmlp", |
| "patch_embed", |
| ".x_embedder", |
| "patch_emb", |
| "wte", |
| ".y_embedder", |
| ".condition_embedder", |
| ".context_embedder", |
| "projector", |
| "lm_head" |
| ], |
| "modules_to_not_use_matmul": [], |
| "non_blocking": false, |
| "quant_conv": false, |
| "quant_embedding": false, |
| "quant_method": "sdnq", |
| "quantization_device": "cuda", |
| "quantized_matmul_dtype": null, |
| "return_device": "cpu", |
| "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" |
| } |
|
|
