{
  "source": "alphakek/Qwen3.8-27B-heretic-ara-DFlash2",
  "output": "magiccodingman/Qwen3.8-27B-heretic-ara-DFlash2-fp8",
  "profile": "mlp-only",
  "format": "experimental selective native block-FP8 E4M3",
  "weight_block_size": [
    128,
    128
  ],
  "activation_scheme": "dynamic",
  "scale_storage": "float32 weight_scale_inv",
  "quantized_tensors": 15,
  "quantized_modules": [
    "layers.0.mlp.down_proj",
    "layers.0.mlp.gate_proj",
    "layers.0.mlp.up_proj",
    "layers.1.mlp.down_proj",
    "layers.1.mlp.gate_proj",
    "layers.1.mlp.up_proj",
    "layers.2.mlp.down_proj",
    "layers.2.mlp.gate_proj",
    "layers.2.mlp.up_proj",
    "layers.3.mlp.down_proj",
    "layers.3.mlp.gate_proj",
    "layers.3.mlp.up_proj",
    "layers.4.mlp.down_proj",
    "layers.4.mlp.gate_proj",
    "layers.4.mlp.up_proj"
  ],
  "preserved_strategy": {
    "qkv": "BF16; required by current DFlash fused context-KV precompute",
    "fc": "BF16 target-hidden-state adapter",
    "convolution": "BF16; runtime constructs these with quant_config=None",
    "candidate_selector": "BF16; runtime constructs it with quant_config=None",
    "norms_and_small_parameters": "BF16"
  },
  "source_tensor_entries": 81,
  "output_tensor_entries": 96,
  "output_tensor_bytes": 2512200960,
  "torch_version": "2.13.0+cu130",
  "python_version": "3.12.3"
}