| { |
| "config_groups": { |
| "group_0": { |
| "targets": [ |
| "Linear" |
| ], |
| "weights": { |
| "num_bits": 4, |
| "type": "float", |
| "symmetric": true, |
| "group_size": 16, |
| "strategy": "tensor_group", |
| "dynamic": false, |
| "scale_dtype": "torch.float8_e4m3fn" |
| }, |
| "input_activations": { |
| "num_bits": 4, |
| "type": "float", |
| "symmetric": true, |
| "group_size": 16, |
| "strategy": "tensor_group", |
| "dynamic": "local", |
| "scale_dtype": "torch.float8_e4m3fn", |
| "observer": "static_minmax" |
| }, |
| "output_activations": null, |
| "format": null |
| } |
| }, |
| "quant_method": "compressed-tensors", |
| "kv_cache_scheme": null, |
| "format": "nvfp4-pack-quantized", |
| "quantization_status": "compressed", |
| "global_compression_ratio": null, |
| "ignore": [ |
| "re:.*vision_tower.*", |
| "vision_tower.blocks.0.attn.qkv", |
| "vision_tower.blocks.0.attn.proj", |
| "vision_tower.blocks.0.mlp.fc1", |
| "vision_tower.blocks.0.mlp.fc2", |
| "vision_tower.blocks.0.mlp.fc3", |
| "vision_tower.blocks.1.attn.qkv", |
| "vision_tower.blocks.1.attn.proj", |
| "vision_tower.blocks.1.mlp.fc1", |
| "vision_tower.blocks.1.mlp.fc2", |
| "vision_tower.blocks.1.mlp.fc3", |
| "vision_tower.blocks.2.attn.qkv", |
| "vision_tower.blocks.2.attn.proj", |
| "vision_tower.blocks.2.mlp.fc1", |
| "vision_tower.blocks.2.mlp.fc2", |
| "vision_tower.blocks.2.mlp.fc3", |
| "vision_tower.blocks.3.attn.qkv", |
| "vision_tower.blocks.3.attn.proj", |
| "vision_tower.blocks.3.mlp.fc1", |
| "vision_tower.blocks.3.mlp.fc2", |
| "vision_tower.blocks.3.mlp.fc3", |
| "vision_tower.blocks.4.attn.qkv", |
| "vision_tower.blocks.4.attn.proj", |
| "vision_tower.blocks.4.mlp.fc1", |
| "vision_tower.blocks.4.mlp.fc2", |
| "vision_tower.blocks.4.mlp.fc3", |
| "vision_tower.blocks.5.attn.qkv", |
| "vision_tower.blocks.5.attn.proj", |
| "vision_tower.blocks.5.mlp.fc1", |
| "vision_tower.blocks.5.mlp.fc2", |
| "vision_tower.blocks.5.mlp.fc3", |
| "vision_tower.blocks.6.attn.qkv", |
| "vision_tower.blocks.6.attn.proj", |
| "vision_tower.blocks.6.mlp.fc1", |
| "vision_tower.blocks.6.mlp.fc2", |
| "vision_tower.blocks.6.mlp.fc3", |
| "vision_tower.blocks.7.attn.qkv", |
| "vision_tower.blocks.7.attn.proj", |
| "vision_tower.blocks.7.mlp.fc1", |
| "vision_tower.blocks.7.mlp.fc2", |
| "vision_tower.blocks.7.mlp.fc3", |
| "vision_tower.blocks.8.attn.qkv", |
| "vision_tower.blocks.8.attn.proj", |
| "vision_tower.blocks.8.mlp.fc1", |
| "vision_tower.blocks.8.mlp.fc2", |
| "vision_tower.blocks.8.mlp.fc3", |
| "vision_tower.blocks.9.attn.qkv", |
| "vision_tower.blocks.9.attn.proj", |
| "vision_tower.blocks.9.mlp.fc1", |
| "vision_tower.blocks.9.mlp.fc2", |
| "vision_tower.blocks.9.mlp.fc3", |
| "vision_tower.blocks.10.attn.qkv", |
| "vision_tower.blocks.10.attn.proj", |
| "vision_tower.blocks.10.mlp.fc1", |
| "vision_tower.blocks.10.mlp.fc2", |
| "vision_tower.blocks.10.mlp.fc3", |
| "vision_tower.blocks.11.attn.qkv", |
| "vision_tower.blocks.11.attn.proj", |
| "vision_tower.blocks.11.mlp.fc1", |
| "vision_tower.blocks.11.mlp.fc2", |
| "vision_tower.blocks.11.mlp.fc3", |
| "vision_tower.blocks.12.attn.qkv", |
| "vision_tower.blocks.12.attn.proj", |
| "vision_tower.blocks.12.mlp.fc1", |
| "vision_tower.blocks.12.mlp.fc2", |
| "vision_tower.blocks.12.mlp.fc3", |
| "vision_tower.blocks.13.attn.qkv", |
| "vision_tower.blocks.13.attn.proj", |
| "vision_tower.blocks.13.mlp.fc1", |
| "vision_tower.blocks.13.mlp.fc2", |
| "vision_tower.blocks.13.mlp.fc3", |
| "vision_tower.blocks.14.attn.qkv", |
| "vision_tower.blocks.14.attn.proj", |
| "vision_tower.blocks.14.mlp.fc1", |
| "vision_tower.blocks.14.mlp.fc2", |
| "vision_tower.blocks.14.mlp.fc3", |
| "vision_tower.blocks.15.attn.qkv", |
| "vision_tower.blocks.15.attn.proj", |
| "vision_tower.blocks.15.mlp.fc1", |
| "vision_tower.blocks.15.mlp.fc2", |
| "vision_tower.blocks.15.mlp.fc3", |
| "vision_tower.blocks.16.attn.qkv", |
| "vision_tower.blocks.16.attn.proj", |
| "vision_tower.blocks.16.mlp.fc1", |
| "vision_tower.blocks.16.mlp.fc2", |
| "vision_tower.blocks.16.mlp.fc3", |
| "vision_tower.blocks.17.attn.qkv", |
| "vision_tower.blocks.17.attn.proj", |
| "vision_tower.blocks.17.mlp.fc1", |
| "vision_tower.blocks.17.mlp.fc2", |
| "vision_tower.blocks.17.mlp.fc3", |
| "vision_tower.blocks.18.attn.qkv", |
| "vision_tower.blocks.18.attn.proj", |
| "vision_tower.blocks.18.mlp.fc1", |
| "vision_tower.blocks.18.mlp.fc2", |
| "vision_tower.blocks.18.mlp.fc3", |
| "vision_tower.blocks.19.attn.qkv", |
| "vision_tower.blocks.19.attn.proj", |
| "vision_tower.blocks.19.mlp.fc1", |
| "vision_tower.blocks.19.mlp.fc2", |
| "vision_tower.blocks.19.mlp.fc3", |
| "vision_tower.blocks.20.attn.qkv", |
| "vision_tower.blocks.20.attn.proj", |
| "vision_tower.blocks.20.mlp.fc1", |
| "vision_tower.blocks.20.mlp.fc2", |
| "vision_tower.blocks.20.mlp.fc3", |
| "vision_tower.blocks.21.attn.qkv", |
| "vision_tower.blocks.21.attn.proj", |
| "vision_tower.blocks.21.mlp.fc1", |
| "vision_tower.blocks.21.mlp.fc2", |
| "vision_tower.blocks.21.mlp.fc3", |
| "vision_tower.blocks.22.attn.qkv", |
| "vision_tower.blocks.22.attn.proj", |
| "vision_tower.blocks.22.mlp.fc1", |
| "vision_tower.blocks.22.mlp.fc2", |
| "vision_tower.blocks.22.mlp.fc3", |
| "vision_tower.blocks.23.attn.qkv", |
| "vision_tower.blocks.23.attn.proj", |
| "vision_tower.blocks.23.mlp.fc1", |
| "vision_tower.blocks.23.mlp.fc2", |
| "vision_tower.blocks.23.mlp.fc3", |
| "vision_tower.blocks.24.attn.qkv", |
| "vision_tower.blocks.24.attn.proj", |
| "vision_tower.blocks.24.mlp.fc1", |
| "vision_tower.blocks.24.mlp.fc2", |
| "vision_tower.blocks.24.mlp.fc3", |
| "vision_tower.blocks.25.attn.qkv", |
| "vision_tower.blocks.25.attn.proj", |
| "vision_tower.blocks.25.mlp.fc1", |
| "vision_tower.blocks.25.mlp.fc2", |
| "vision_tower.blocks.25.mlp.fc3", |
| "vision_tower.blocks.26.attn.qkv", |
| "vision_tower.blocks.26.attn.proj", |
| "vision_tower.blocks.26.mlp.fc1", |
| "vision_tower.blocks.26.mlp.fc2", |
| "vision_tower.blocks.26.mlp.fc3", |
| "vision_tower.blocks.27.attn.qkv", |
| "vision_tower.blocks.27.attn.proj", |
| "vision_tower.blocks.27.mlp.fc1", |
| "vision_tower.blocks.27.mlp.fc2", |
| "vision_tower.blocks.27.mlp.fc3", |
| "vision_tower.blocks.28.attn.qkv", |
| "vision_tower.blocks.28.attn.proj", |
| "vision_tower.blocks.28.mlp.fc1", |
| "vision_tower.blocks.28.mlp.fc2", |
| "vision_tower.blocks.28.mlp.fc3", |
| "vision_tower.blocks.29.attn.qkv", |
| "vision_tower.blocks.29.attn.proj", |
| "vision_tower.blocks.29.mlp.fc1", |
| "vision_tower.blocks.29.mlp.fc2", |
| "vision_tower.blocks.29.mlp.fc3", |
| "vision_tower.blocks.30.attn.qkv", |
| "vision_tower.blocks.30.attn.proj", |
| "vision_tower.blocks.30.mlp.fc1", |
| "vision_tower.blocks.30.mlp.fc2", |
| "vision_tower.blocks.30.mlp.fc3", |
| "vision_tower.blocks.31.attn.qkv", |
| "vision_tower.blocks.31.attn.proj", |
| "vision_tower.blocks.31.mlp.fc1", |
| "vision_tower.blocks.31.mlp.fc2", |
| "vision_tower.blocks.31.mlp.fc3", |
| "vision_tower.blocks.32.attn.qkv", |
| "vision_tower.blocks.32.attn.proj", |
| "vision_tower.blocks.32.mlp.fc1", |
| "vision_tower.blocks.32.mlp.fc2", |
| "vision_tower.blocks.32.mlp.fc3", |
| "vision_tower.blocks.33.attn.qkv", |
| "vision_tower.blocks.33.attn.proj", |
| "vision_tower.blocks.33.mlp.fc1", |
| "vision_tower.blocks.33.mlp.fc2", |
| "vision_tower.blocks.33.mlp.fc3", |
| "vision_tower.blocks.34.attn.qkv", |
| "vision_tower.blocks.34.attn.proj", |
| "vision_tower.blocks.34.mlp.fc1", |
| "vision_tower.blocks.34.mlp.fc2", |
| "vision_tower.blocks.34.mlp.fc3", |
| "vision_tower.blocks.35.attn.qkv", |
| "vision_tower.blocks.35.attn.proj", |
| "vision_tower.blocks.35.mlp.fc1", |
| "vision_tower.blocks.35.mlp.fc2", |
| "vision_tower.blocks.35.mlp.fc3", |
| "vision_tower.blocks.36.attn.qkv", |
| "vision_tower.blocks.36.attn.proj", |
| "vision_tower.blocks.36.mlp.fc1", |
| "vision_tower.blocks.36.mlp.fc2", |
| "vision_tower.blocks.36.mlp.fc3", |
| "vision_tower.blocks.37.attn.qkv", |
| "vision_tower.blocks.37.attn.proj", |
| "vision_tower.blocks.37.mlp.fc1", |
| "vision_tower.blocks.37.mlp.fc2", |
| "vision_tower.blocks.37.mlp.fc3", |
| "vision_tower.blocks.38.attn.qkv", |
| "vision_tower.blocks.38.attn.proj", |
| "vision_tower.blocks.38.mlp.fc1", |
| "vision_tower.blocks.38.mlp.fc2", |
| "vision_tower.blocks.38.mlp.fc3", |
| "vision_tower.blocks.39.attn.qkv", |
| "vision_tower.blocks.39.attn.proj", |
| "vision_tower.blocks.39.mlp.fc1", |
| "vision_tower.blocks.39.mlp.fc2", |
| "vision_tower.blocks.39.mlp.fc3", |
| "vision_tower.blocks.40.attn.qkv", |
| "vision_tower.blocks.40.attn.proj", |
| "vision_tower.blocks.40.mlp.fc1", |
| "vision_tower.blocks.40.mlp.fc2", |
| "vision_tower.blocks.40.mlp.fc3", |
| "vision_tower.blocks.41.attn.qkv", |
| "vision_tower.blocks.41.attn.proj", |
| "vision_tower.blocks.41.mlp.fc1", |
| "vision_tower.blocks.41.mlp.fc2", |
| "vision_tower.blocks.41.mlp.fc3", |
| "vision_tower.merger.mlp.0", |
| "vision_tower.merger.mlp.2", |
| "lm_head" |
| ], |
| "moe_sparsity_config": null, |
| "provider": "auto-round", |
| "autoround_config": { |
| "autoround_version": "0.17.0", |
| "bits": 4, |
| "act_bits": 4, |
| "data_type": "nv_fp", |
| "act_data_type": "nv_fp4_with_static_gs", |
| "group_size": 16, |
| "act_group_size": 16, |
| "sym": true, |
| "act_sym": true, |
| "act_dynamic": true, |
| "amp": true, |
| "batch_size": 2, |
| "low_gpu_mem_usage": false, |
| "nsamples": 512, |
| "quant_block_list": [ |
| [ |
| "model.layers.0", |
| "model.layers.1", |
| "model.layers.2", |
| "model.layers.3", |
| "model.layers.4", |
| "model.layers.5", |
| "model.layers.6", |
| "model.layers.7", |
| "model.layers.8", |
| "model.layers.9", |
| "model.layers.10", |
| "model.layers.11", |
| "model.layers.12", |
| "model.layers.13", |
| "model.layers.14", |
| "model.layers.15", |
| "model.layers.16", |
| "model.layers.17", |
| "model.layers.18", |
| "model.layers.19", |
| "model.layers.20", |
| "model.layers.21", |
| "model.layers.22", |
| "model.layers.23", |
| "model.layers.24", |
| "model.layers.25", |
| "model.layers.26", |
| "model.layers.27" |
| ] |
| ], |
| "scale_dtype": "torch.float16", |
| "seqlen": 4096, |
| "supported_types": [ |
| "torch.nn.modules.linear.Linear", |
| "transformers.pytorch_utils.Conv1D" |
| ], |
| "static_attention_granularity": "tensor", |
| "static_kv_granularity": "tensor", |
| "to_quant_block_names": "model.layers", |
| "rotation_configs": [], |
| "enable_minmax_tuning": true, |
| "enable_norm_bias_tuning": false, |
| "enable_quanted_input": true, |
| "gradient_accumulate_steps": 4, |
| "iters": 1200, |
| "lr": 0.0008333333333333334, |
| "minmax_lr": 0.0008333333333333334 |
| } |
| } |
