validation/orbitquant_w4a4_cuda_gate.json

{
  "status": "PASS",
  "gpu": "NVIDIA L4",
  "compute_capability": [
    8,
    9
  ],
  "torch": "2.11.0+cu128",
  "cuda": "12.8",
  "dimensions": [
    5120,
    13824
  ],
  "real_projection_shapes": [
    [
      5120,
      5120
    ],
    [
      13824,
      5120
    ],
    [
      5120,
      13824
    ]
  ],
  "rows_tested": 17,
  "a4": "RPBH -> post-rotation L2 -> Lloyd-Max -> uint4",
  "a4_bf16_dequant_bit_exact": true,
  "w4a4": "packed nonuniform uint4 W [K/2,N] x packed uint4 A -> BF16 tile dequant -> FP32 accumulate",
  "weight_storage_layout": "K_half_by_N",
  "packed_artifact": {
    "packed_dir": "/content/OrbitQuant_Animate2_W4A4_PACKED",
    "manifest_sha256": "8f1f4e0acc97bc05433bf3ce34bb1bfb3b013eeca90a72c9d6c07299c5a97891",
    "shard_sizes": {
      "orbitquant-runtime-00001.safetensors": 794858280,
      "orbitquant-runtime-00002.safetensors": 794858320,
      "orbitquant-runtime-00003.safetensors": 794858320,
      "orbitquant-runtime-00004.safetensors": 794858296,
      "orbitquant-runtime-00005.safetensors": 772576096,
      "orbitquant-runtime-00006.safetensors": 772558672,
      "orbitquant-runtime-00007.safetensors": 794858304,
      "orbitquant-runtime-00008.safetensors": 794858328,
      "orbitquant-runtime-00009.safetensors": 794858296,
      "orbitquant-runtime-00010.safetensors": 794858232,
      "orbitquant-runtime-00011.safetensors": 175758600,
      "orbitquant-runtime-00012.safetensors": 493700016
    },
    "actual_targets": [
      {
        "target": "blocks.0.block.cross_attn.k.weight",
        "shape_NK": [
          5120,
          5120
        ],
        "max_abs": 0.0078125,
        "mean_abs": 4.6032124600969837e-07
      },
      {
        "target": "blocks.0.block.ffn.2.weight",
        "shape_NK": [
          5120,
          13824
        ],
        "max_abs": 0.015625,
        "mean_abs": 9.963549700842123e-07
      },
      {
        "target": "blocks.0.block.ffn.0.weight",
        "shape_NK": [
          13824,
          5120
        ],
        "max_abs": 0.0078125,
        "mean_abs": 2.0674802669873316e-07
      }
    ],
    "format": "OrbitQuant_WanAnimate2_direct_source_packed_nonuniform_W4A4_v3",
    "weight_storage_layout": "K_half_by_N"
  }
}