File size: 1,748 Bytes
9a16379 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 | {
"source_model": "microsoft/Lens",
"target_model": "WaveCut/Lens-SDNQ-uint4-static",
"method": "SDNQ uint4 static",
"corrected_recipe": true,
"weights_dtype": "uint4",
"quantized_matmul_dtype": "int8",
"group_size": 0,
"use_quantized_matmul": true,
"dequantize_fp32": false,
"modules_to_not_convert_user": [
".final_layer",
"pos_embed",
".norm_out",
".y_embedder",
".context_embedder",
".condition_embedder",
".x_embedder",
".vid_out",
".emb_out",
".img_in",
"patch_embed",
".time_embed",
".t_embedder",
"multi_modal_projector",
"patch_emb",
"norm",
".img_out",
"patch_embedding",
"lm_head",
".proj_out",
".vid_in",
".txt_in",
"wte",
"time_text_embed",
".txt_out",
".emb_in",
"*.img_mod.*",
"*.txt_mod.*"
],
"root_cause_from_turbo_ablation": "Do not quantize transformer modulation linears (*.img_mod.* and *.txt_mod.*); all-linear UINT4 caused periodic grid artifacts and text degradation on Lens-Turbo.",
"transformer_load_time_s": 3.677,
"transformer_load_peak_allocated_gb": 8.359,
"transformer_load_peak_reserved_gb": 8.424,
"quantization_time_s": 0.313,
"quantization_peak_allocated_gb": 8.425,
"quantization_peak_reserved_gb": 8.485,
"base_transformer_tensor_storage_gb": 16.417,
"quant_transformer_tensor_storage_gb": 4.301,
"transformer_storage_reduction_percent": 73.8,
"base_transformer_repo_files_gb": 16.417,
"quant_transformer_repo_files_gb": 4.302,
"base_transformer_dtypes": {
"FLOAT32": 16416900608
},
"quant_transformer_dtypes": {
"BFLOAT16": 2942501632,
"UINT8": 1358954496
},
"base_transformer_tensors": 1264,
"quant_transformer_tensors": 2224
} |