{ "activation_headroom": 1.25, "activation_precision": 4, "calibration": { "created_at": "2026-10-04T03:14:39.274275Z", "input_files": [ { "path": "calibration-corpus.json", "sha256": "4ce5463b8b9e1e3e5998bd73e812208a209b3c7e989214140eba97225d9afcd5", "size_bytes": 1698 } ], "method": "observed-input-and-expert-replay-absmax", "runtime": "vllm", "runtime_version": "0.25.1", "schema_version": "axquant.cuda-activation.v1", "source_precision": "bfloat16", "statistics": [ { "absolute_maximum": 78.5, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.2.mlp.gate_proj.weight" }, { "absolute_maximum": 78.5, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.2.mlp.up_proj.weight" }, { "absolute_maximum": 17.625, "input_columns": 12288, "sample_count": 530, "tensor_name": "layers.2.mlp.down_proj.weight" }, { "absolute_maximum": 63.0, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.3.mlp.gate_proj.weight" }, { "absolute_maximum": 63.0, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.3.mlp.up_proj.weight" }, { "absolute_maximum": 21.875, "input_columns": 12288, "sample_count": 530, "tensor_name": "layers.3.mlp.down_proj.weight" }, { "absolute_maximum": 38.75, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.4.mlp.gate_proj.weight" }, { "absolute_maximum": 38.75, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.4.mlp.up_proj.weight" }, { "absolute_maximum": 22.875, "input_columns": 12288, "sample_count": 530, "tensor_name": "layers.4.mlp.down_proj.weight" }, { "absolute_maximum": 25.375, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.5.mlp.gate_proj.weight" }, { "absolute_maximum": 25.375, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.5.mlp.up_proj.weight" }, { "absolute_maximum": 13.625, "input_columns": 12288, "sample_count": 530, "tensor_name": "layers.5.mlp.down_proj.weight" }, { "absolute_maximum": 170.0, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.6.mlp.gate_proj.weight" }, { "absolute_maximum": 170.0, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.6.mlp.up_proj.weight" }, { "absolute_maximum": 4512.0, "input_columns": 12288, "sample_count": 530, "tensor_name": "layers.6.mlp.down_proj.weight" }, { "absolute_maximum": 39.0, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.7.mlp.gate_proj.weight" }, { "absolute_maximum": 39.0, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.7.mlp.up_proj.weight" }, { "absolute_maximum": 15.5, "input_columns": 12288, "sample_count": 530, "tensor_name": "layers.7.mlp.down_proj.weight" }, { "absolute_maximum": 7.125, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.8.mlp.gate_proj.weight" }, { "absolute_maximum": 7.125, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.8.mlp.up_proj.weight" }, { "absolute_maximum": 15.875, "input_columns": 12288, "sample_count": 530, "tensor_name": "layers.8.mlp.down_proj.weight" }, { "absolute_maximum": 7.71875, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.9.mlp.gate_proj.weight" }, { "absolute_maximum": 7.71875, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.9.mlp.up_proj.weight" }, { "absolute_maximum": 17.125, "input_columns": 12288, "sample_count": 530, "tensor_name": "layers.9.mlp.down_proj.weight" }, { "absolute_maximum": 8.5, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.10.mlp.gate_proj.weight" }, { "absolute_maximum": 8.5, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.10.mlp.up_proj.weight" }, { "absolute_maximum": 18.0, "input_columns": 12288, "sample_count": 530, "tensor_name": "layers.10.mlp.down_proj.weight" }, { "absolute_maximum": 8.5, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.11.mlp.gate_proj.weight" }, { "absolute_maximum": 8.5, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.11.mlp.up_proj.weight" }, { "absolute_maximum": 20.0, "input_columns": 12288, "sample_count": 530, "tensor_name": "layers.11.mlp.down_proj.weight" }, { "absolute_maximum": 9.3125, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.12.mlp.gate_proj.weight" }, { "absolute_maximum": 9.3125, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.12.mlp.up_proj.weight" }, { "absolute_maximum": 42.25, "input_columns": 12288, "sample_count": 530, "tensor_name": "layers.12.mlp.down_proj.weight" }, { "absolute_maximum": 10.875, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.13.mlp.gate_proj.weight" }, { "absolute_maximum": 10.875, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.13.mlp.up_proj.weight" }, { "absolute_maximum": 10.5625, "input_columns": 12288, "sample_count": 530, "tensor_name": "layers.13.mlp.down_proj.weight" }, { "absolute_maximum": 13.1875, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.14.mlp.gate_proj.weight" }, { "absolute_maximum": 13.1875, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.14.mlp.up_proj.weight" }, { "absolute_maximum": 49.0, "input_columns": 12288, "sample_count": 530, "tensor_name": "layers.14.mlp.down_proj.weight" }, { "absolute_maximum": 14.625, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.15.mlp.gate_proj.weight" }, { "absolute_maximum": 14.625, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.15.mlp.up_proj.weight" }, { "absolute_maximum": 21.875, "input_columns": 12288, "sample_count": 530, "tensor_name": "layers.15.mlp.down_proj.weight" }, { "absolute_maximum": 33.5, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.16.mlp.gate_proj.weight" }, { "absolute_maximum": 33.5, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.16.mlp.up_proj.weight" }, { "absolute_maximum": 90.5, "input_columns": 12288, "sample_count": 530, "tensor_name": "layers.16.mlp.down_proj.weight" }, { "absolute_maximum": 11.125, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.17.mlp.gate_proj.weight" }, { "absolute_maximum": 11.125, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.17.mlp.up_proj.weight" }, { "absolute_maximum": 8.4375, "input_columns": 12288, "sample_count": 530, "tensor_name": "layers.17.mlp.down_proj.weight" }, { "absolute_maximum": 10.6875, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.18.mlp.gate_proj.weight" }, { "absolute_maximum": 10.6875, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.18.mlp.up_proj.weight" }, { "absolute_maximum": 48.0, "input_columns": 12288, "sample_count": 530, "tensor_name": "layers.18.mlp.down_proj.weight" }, { "absolute_maximum": 10.4375, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.19.mlp.gate_proj.weight" }, { "absolute_maximum": 10.4375, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.19.mlp.up_proj.weight" }, { "absolute_maximum": 9.6875, "input_columns": 12288, "sample_count": 530, "tensor_name": "layers.19.mlp.down_proj.weight" }, { "absolute_maximum": 11.25, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.20.mlp.gate_proj.weight" }, { "absolute_maximum": 11.25, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.20.mlp.up_proj.weight" }, { "absolute_maximum": 14.5, "input_columns": 12288, "sample_count": 530, "tensor_name": "layers.20.mlp.down_proj.weight" }, { "absolute_maximum": 10.0625, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.21.mlp.gate_proj.weight" }, { "absolute_maximum": 10.0625, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.21.mlp.up_proj.weight" }, { "absolute_maximum": 14.5625, "input_columns": 12288, "sample_count": 530, "tensor_name": "layers.21.mlp.down_proj.weight" }, { "absolute_maximum": 9.1875, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.22.mlp.gate_proj.weight" }, { "absolute_maximum": 9.1875, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.22.mlp.up_proj.weight" }, { "absolute_maximum": 33.75, "input_columns": 12288, "sample_count": 530, "tensor_name": "layers.22.mlp.down_proj.weight" }, { "absolute_maximum": 9.6875, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.23.mlp.gate_proj.weight" }, { "absolute_maximum": 9.6875, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.23.mlp.up_proj.weight" }, { "absolute_maximum": 54.75, "input_columns": 12288, "sample_count": 530, "tensor_name": "layers.23.mlp.down_proj.weight" }, { "absolute_maximum": 10.875, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.24.mlp.gate_proj.weight" }, { "absolute_maximum": 10.875, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.24.mlp.up_proj.weight" }, { "absolute_maximum": 97.0, "input_columns": 12288, "sample_count": 530, "tensor_name": "layers.24.mlp.down_proj.weight" }, { "absolute_maximum": 11.1875, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.25.mlp.gate_proj.weight" }, { "absolute_maximum": 11.1875, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.25.mlp.up_proj.weight" }, { "absolute_maximum": 92.5, "input_columns": 12288, "sample_count": 530, "tensor_name": "layers.25.mlp.down_proj.weight" }, { "absolute_maximum": 11.9375, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.26.mlp.gate_proj.weight" }, { "absolute_maximum": 11.9375, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.26.mlp.up_proj.weight" }, { "absolute_maximum": 80.5, "input_columns": 12288, "sample_count": 530, "tensor_name": "layers.26.mlp.down_proj.weight" }, { "absolute_maximum": 11.875, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.27.mlp.gate_proj.weight" }, { "absolute_maximum": 11.875, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.27.mlp.up_proj.weight" }, { "absolute_maximum": 137.0, "input_columns": 12288, "sample_count": 530, "tensor_name": "layers.27.mlp.down_proj.weight" }, { "absolute_maximum": 12.0625, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.28.mlp.gate_proj.weight" }, { "absolute_maximum": 12.0625, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.28.mlp.up_proj.weight" }, { "absolute_maximum": 129.0, "input_columns": 12288, "sample_count": 530, "tensor_name": "layers.28.mlp.down_proj.weight" }, { "absolute_maximum": 11.5625, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.29.mlp.gate_proj.weight" }, { "absolute_maximum": 11.5625, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.29.mlp.up_proj.weight" }, { "absolute_maximum": 104.5, "input_columns": 12288, "sample_count": 530, "tensor_name": "layers.29.mlp.down_proj.weight" }, { "absolute_maximum": 10.8125, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.30.mlp.gate_proj.weight" }, { "absolute_maximum": 10.8125, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.30.mlp.up_proj.weight" }, { "absolute_maximum": 92.5, "input_columns": 12288, "sample_count": 530, "tensor_name": "layers.30.mlp.down_proj.weight" }, { "absolute_maximum": 11.5625, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.31.mlp.gate_proj.weight" }, { "absolute_maximum": 11.5625, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.31.mlp.up_proj.weight" }, { "absolute_maximum": 95.5, "input_columns": 12288, "sample_count": 530, "tensor_name": "layers.31.mlp.down_proj.weight" }, { "absolute_maximum": 13.125, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.32.mlp.gate_proj.weight" }, { "absolute_maximum": 13.125, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.32.mlp.up_proj.weight" }, { "absolute_maximum": 203.0, "input_columns": 12288, "sample_count": 530, "tensor_name": "layers.32.mlp.down_proj.weight" }, { "absolute_maximum": 26.625, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.33.mlp.gate_proj.weight" }, { "absolute_maximum": 26.625, "input_columns": 4096, "sample_count": 530, "tensor_name": "layers.33.mlp.up_proj.weight" }, { "absolute_maximum": 101.0, "input_columns": 12288, "sample_count": 530, "tensor_name": "layers.33.mlp.down_proj.weight" } ], "status": "complete", "weight_plan_sha256": "072bc6e4412dc0b1447a8a3d3d0f0e3f5041e021ab9141830e60682b3a178921" }, "created_at": "2026-10-04T03:16:11.496533Z", "format": "nvfp4", "schema_version": "axquant.cuda-w4a4-plan.v1", "weight_plan": { "activation_dtype": "bfloat16", "activation_precision": 16, "algorithm": "rtn", "allocations": [ { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "embedding", "scale_group": null, "shape": [ 151665, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "embed_tokens.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.0.input_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.0.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "mlp", "scale_group": null, "shape": [ 12288, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.0.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "mlp", "scale_group": null, "shape": [ 12288, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.0.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.0.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.0.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.0.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.0.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.0.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.0.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.0.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.1.input_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.1.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "mlp", "scale_group": null, "shape": [ 12288, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.1.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "mlp", "scale_group": null, "shape": [ 12288, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.1.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.1.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.1.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.1.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.1.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.1.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.1.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.1.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.10.input_layernorm.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.10.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.10.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.10.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.10.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.10.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.10.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.10.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.10.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.10.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.10.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.10.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.10.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.11.input_layernorm.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.11.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.11.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.11.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.11.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.11.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.11.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.11.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.11.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.11.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.11.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.11.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.11.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.12.input_layernorm.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.12.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.12.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.12.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.12.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.12.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.12.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.12.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.12.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.12.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.12.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.12.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.12.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.13.input_layernorm.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.13.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.13.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.13.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.13.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.13.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.13.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.13.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.13.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.13.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.13.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.13.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.13.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.14.input_layernorm.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.14.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.14.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.14.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.14.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.14.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.14.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.14.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.14.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.14.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.14.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.14.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.14.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.15.input_layernorm.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.15.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.15.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.15.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.15.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.15.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.15.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.15.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.15.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.15.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.15.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.15.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.15.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.16.input_layernorm.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.16.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.16.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.16.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.16.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.16.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.16.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.16.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.16.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.16.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.16.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.16.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.16.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.17.input_layernorm.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.17.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.17.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.17.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.17.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.17.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.17.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.17.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.17.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.17.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.17.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.17.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.17.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.18.input_layernorm.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.18.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.18.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.18.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.18.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.18.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.18.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.18.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.18.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.18.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.18.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.18.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.18.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.19.input_layernorm.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.19.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.19.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.19.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.19.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.19.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.19.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.19.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.19.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.19.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.19.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.19.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.19.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.2.input_layernorm.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.2.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.2.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.2.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.2.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.2.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.2.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.2.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.2.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.2.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.2.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.2.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.2.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.20.input_layernorm.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.20.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.20.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.20.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.20.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.20.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.20.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.20.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.20.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.20.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.20.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.20.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.20.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.21.input_layernorm.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.21.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.21.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.21.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.21.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.21.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.21.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.21.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.21.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.21.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.21.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.21.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.21.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.22.input_layernorm.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.22.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.22.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.22.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.22.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.22.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.22.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.22.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.22.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.22.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.22.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.22.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.22.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.23.input_layernorm.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.23.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.23.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.23.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.23.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.23.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.23.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.23.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.23.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.23.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.23.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.23.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.23.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.24.input_layernorm.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.24.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.24.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.24.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.24.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.24.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.24.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.24.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.24.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.24.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.24.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.24.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.24.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.25.input_layernorm.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.25.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.25.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.25.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.25.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.25.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.25.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.25.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.25.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.25.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.25.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.25.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.25.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.26.input_layernorm.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.26.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.26.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.26.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.26.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.26.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.26.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.26.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.26.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.26.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.26.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.26.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.26.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.27.input_layernorm.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.27.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.27.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.27.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.27.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.27.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.27.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.27.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.27.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.27.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.27.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.27.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.27.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.28.input_layernorm.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.28.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.28.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.28.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.28.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.28.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.28.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.28.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.28.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.28.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.28.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.28.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.28.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.29.input_layernorm.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.29.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.29.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.29.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.29.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.29.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.29.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.29.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.29.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.29.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.29.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.29.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.29.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.3.input_layernorm.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.3.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.3.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.3.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.3.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.3.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.3.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.3.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.3.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.3.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.3.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.3.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.3.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.30.input_layernorm.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.30.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.30.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.30.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.30.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.30.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.30.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.30.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.30.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.30.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.30.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.30.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.30.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.31.input_layernorm.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.31.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.31.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.31.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.31.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.31.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.31.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.31.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.31.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.31.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.31.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.31.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.31.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.32.input_layernorm.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.32.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.32.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.32.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.32.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.32.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.32.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.32.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.32.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.32.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.32.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.32.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.32.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.33.input_layernorm.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.33.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.33.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.33.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.33.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.33.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.33.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.33.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.33.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.33.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.33.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.33.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.33.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.34.input_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.34.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "mlp", "scale_group": null, "shape": [ 12288, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.34.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "mlp", "scale_group": null, "shape": [ 12288, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.34.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.34.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.34.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.34.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.34.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.34.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.34.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.34.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00004-of-00004.safetensors", "tensor_name": "layers.35.input_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00004-of-00004.safetensors", "tensor_name": "layers.35.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "mlp", "scale_group": null, "shape": [ 12288, 4096 ], "source_file": "model-00004-of-00004.safetensors", "tensor_name": "layers.35.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "mlp", "scale_group": null, "shape": [ 12288, 4096 ], "source_file": "model-00004-of-00004.safetensors", "tensor_name": "layers.35.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00004-of-00004.safetensors", "tensor_name": "layers.35.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00004-of-00004.safetensors", "tensor_name": "layers.35.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.35.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00004-of-00004.safetensors", "tensor_name": "layers.35.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00004-of-00004.safetensors", "tensor_name": "layers.35.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.35.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00003-of-00004.safetensors", "tensor_name": "layers.35.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.4.input_layernorm.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.4.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.4.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.4.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.4.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.4.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.4.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.4.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.4.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.4.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.4.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.4.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.4.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.5.input_layernorm.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.5.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.5.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.5.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.5.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.5.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.5.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.5.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.5.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.5.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.5.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.5.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.5.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.6.input_layernorm.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.6.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.6.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.6.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.6.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.6.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.6.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.6.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.6.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.6.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.6.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.6.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.6.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.7.input_layernorm.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.7.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.7.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.7.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.7.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.7.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.7.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.7.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.7.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.7.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.7.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.7.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.7.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.8.input_layernorm.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.8.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.8.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.8.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.8.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.8.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.8.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.8.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.8.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.8.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.8.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.8.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.8.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.9.input_layernorm.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": null, "shape": [ 4096, 12288 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.9.mlp.down_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.9.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.9.mlp.gate_proj.weight" }, { "dtype": "BF16", "method": "nvfp4", "reason": "Unmeasured native NVFP4 RTN, block 16, activations preserved", "role": "mlp", "scale_group": "layers.9.mlp.gate_up_proj", "shape": [ 12288, 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.9.mlp.up_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00002-of-00004.safetensors", "tensor_name": "layers.9.post_attention_layernorm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.9.self_attn.k_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.9.self_attn.k_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.9.self_attn.o_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 128 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.9.self_attn.q_norm.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 4096, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.9.self_attn.q_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Preserve complete fused runtime unit", "role": "attention", "scale_group": null, "shape": [ 1024, 4096 ], "source_file": "model-00001-of-00004.safetensors", "tensor_name": "layers.9.self_attn.v_proj.weight" }, { "dtype": "BF16", "method": "preserve", "reason": "Source precision preserved by protection, shape, dtype or keep policy", "role": "norm", "scale_group": null, "shape": [ 4096 ], "source_file": "model-00004-of-00004.safetensors", "tensor_name": "norm.weight" } ], "block_size": 16, "config_sha256": "565586ba86accfa587a2b8adbd49008e7ab61d55b69102c2ee75571c27ef58c8", "created_at": "2026-10-04T03:01:10.440996Z", "estimated_weight_bytes": 8188823936, "evidence_kind": "unmeasured", "format": "nvfp4", "keep_patterns": [ "layers.*.self_attn.*", "layers.0.mlp.*", "layers.1.mlp.*", "layers.34.mlp.*", "layers.35.mlp.*" ], "model_id": "Qwen/Qwen3-Embedding-8B", "model_type": "qwen3", "revision": "1d8ad4ca9b3dd8059ad90a75d4983776a23d44af", "schema_version": "axquant.cuda-plan.v1", "source_files": [ { "path": "1_Pooling/config.json", "sha256": "2e1da26b3fd65cf7e370d2fabf28a8c59efa7edb525d1b8b50be8e5ca1048ea6", "size_bytes": 313 }, { "path": "config.json", "sha256": "dcccf7c7890c8debc4af8dace8d6acd8dd50bd56d50d4d8b60529381ba3de3d7", "size_bytes": 729 }, { "path": "config_sentence_transformers.json", "sha256": "10667c72ddb772627bf1780cb7f86af8e2ae0032b8c243c731172064105c6961", "size_bytes": 215 }, { "path": "generation_config.json", "sha256": "28396d421a2108acce96383f6a7de78008f7f1b17f807958f3c14c51dbfb65fb", "size_bytes": 117 }, { "path": "merges.txt", "sha256": "8831e4f1a044471340f7c0a83d7bd71306a5b867e95fd870f74d0c5308a904d5", "size_bytes": 1671853 }, { "path": "model-00001-of-00004.safetensors", "sha256": "99b343597fe840706146144699a8b9188dd3387e43eb61faf0231b70b249d451", "size_bytes": 4900037024 }, { "path": "model-00002-of-00004.safetensors", "sha256": "dff635b0f6dbbaad2a2d633ef037ec0a39bc165cc1806c712fbd6fcbcb4526c0", "size_bytes": 4915959512 }, { "path": "model-00003-of-00004.safetensors", "sha256": "30b1d4c53d84eb018f642cad7b373f0aabf79699872d8702c1f38577c0a59a2f", "size_bytes": 4983067656 }, { "path": "model-00004-of-00004.safetensors", "sha256": "36cbc9c60375693629f25743c1e77ebb1724af58e671b2376463193c7fd21ef6", "size_bytes": 335570376 }, { "path": "model.safetensors.index.json", "sha256": "ceaf1429bfa18eecc75d079737e8d1fe8dc0627f5ec7e6a985514626d1ac6a77", "size_bytes": 30432 }, { "path": "modules.json", "sha256": "84e40c8e006c9b1d6c122e02cba9b02458120b5fb0c87b746c41e0207cf642cf", "size_bytes": 349 }, { "path": "tokenizer.json", "sha256": "83cdf8c3a34f68862319cb1810ee7b1e2c0a44e0864ae930194ddb76bb7feb8d", "size_bytes": 11422947 }, { "path": "tokenizer_config.json", "sha256": "2f58f4bbd7bbce15d683f525954ef3a92cd82f5e06415a9c513859bf8ab72436", "size_bytes": 7256 }, { "path": "vocab.json", "sha256": "ca10d7e9fb3ed18575dd1e277a2579c16d108e32f27439684afa0e10b1440910", "size_bytes": 2776833 } ] } }