97 lines
1.8 KiB
JSON
97 lines
1.8 KiB
JSON
|
{
|
||
|
|
"device": "NVIDIA GB10",
|
||
|
|
"torch": "2.9.1+cu130",
|
||
|
|
"cutlass_dsl": "4.6.2",
|
||
|
|
"metadata": {
|
||
|
|
"width": 1344,
|
||
|
|
"height": 768,
|
||
|
|
"frames": 124,
|
||
|
|
"steps": 12,
|
||
|
|
"sampler_step": 1,
|
||
|
|
"seed": 440420,
|
||
|
|
"text_tokens": 100,
|
||
|
|
"hidden_shape": [
|
||
|
|
37810,
|
||
|
|
5376
|
||
|
|
],
|
||
|
|
"segments": [
|
||
|
|
[
|
||
|
|
0,
|
||
|
|
100,
|
||
|
|
1
|
||
|
|
],
|
||
|
|
[
|
||
|
|
100,
|
||
|
|
514,
|
||
|
|
2
|
||
|
|
],
|
||
|
|
[
|
||
|
|
514,
|
||
|
|
37810,
|
||
|
|
0
|
||
|
|
]
|
||
|
|
]
|
||
|
|
},
|
||
|
|
"linear": "mlp_fc1",
|
||
|
|
"rows": 128,
|
||
|
|
"mnk": [
|
||
|
|
128,
|
||
|
|
28672,
|
||
|
|
5376
|
||
|
|
],
|
||
|
|
"tile_shape_mnk": [
|
||
|
|
128,
|
||
|
|
128,
|
||
|
|
128
|
||
|
|
],
|
||
|
|
"swap_nibbles": false,
|
||
|
|
"reencode_fp4": true,
|
||
|
|
"zero_a": false,
|
||
|
|
"fuse_alpha": true,
|
||
|
|
"stream_a": true,
|
||
|
|
"streamed_activation_materialization": {
|
||
|
|
"global_qdata": false,
|
||
|
|
"global_sfa": false
|
||
|
|
},
|
||
|
|
"timing": {
|
||
|
|
"warmup": 5,
|
||
|
|
"iterations": 20,
|
||
|
|
"total_ms": 114.3238754272461,
|
||
|
|
"mean_ms": 5.716193771362304
|
||
|
|
},
|
||
|
|
"cute_shapes": {
|
||
|
|
"a": "(128, 5376, 1)",
|
||
|
|
"b": "(28672, 5376, 1)",
|
||
|
|
"sfa": "(1,)",
|
||
|
|
"sfb": "(28672, 336, 1)",
|
||
|
|
"c": "(128, 28672, 1)"
|
||
|
|
},
|
||
|
|
"tensor_scales": {
|
||
|
|
"a": 0.00494384765625,
|
||
|
|
"b": 0.0023651123046875,
|
||
|
|
"alpha": 1.1692754924297333e-05
|
||
|
|
},
|
||
|
|
"raw_output": {
|
||
|
|
"dtype": "torch.bfloat16",
|
||
|
|
"checksum": -3814965.5,
|
||
|
|
"max_abs": 95.0,
|
||
|
|
"nonzero": 3670016,
|
||
|
|
"finite": true
|
||
|
|
},
|
||
|
|
"raw_blockscaled_parity": {
|
||
|
|
"applicable": false,
|
||
|
|
"reference_checksum": -326266716160.0,
|
||
|
|
"candidate_checksum": -3814965.5,
|
||
|
|
"equal": false,
|
||
|
|
"max_abs": 8126369.0,
|
||
|
|
"mean_abs": 646601.3125
|
||
|
|
},
|
||
|
|
"reference_checksum": -3814965.5,
|
||
|
|
"candidate_checksum": -3814965.5,
|
||
|
|
"equal": true,
|
||
|
|
"max_abs": 0.0,
|
||
|
|
"mean_abs": 0.0,
|
||
|
|
"relative_l2": 0.0,
|
||
|
|
"numerical_note": "The experimental epilogue applies the FP32 global-scale product before BF16 conversion."
|
||
|
|
}
|