h3-blackwell-runtime/benchmarks/gb10-cute-nvfp4-tile-producer.json

93 lines
1.6 KiB
JSON
Raw Normal View History

{
"device": "NVIDIA GB10",
"cutlass_dsl": "4.6.2",
"tile": [
128,
128
],
"all_equal": true,
"cases": [
{
"name": "random",
"tensor_scale": 0.0013885498046875,
"fp4_difference_count": 0,
"fp4_equal": true,
"block_scale_difference_count": 0,
"block_scales_equal": true,
"scalar_block_scales_equal": true,
"actual_block_scale_bytes": [
107,
110,
111,
112,
113,
114,
115,
116,
117,
118,
119,
120,
121,
122,
123,
124,
125,
126
],
"expected_block_scale_bytes": [
107,
110,
111,
112,
113,
114,
115,
116,
117,
118,
119,
120,
121,
122,
123,
124,
125,
126
]
},
{
"name": "zeros",
"tensor_scale": 0.0,
"fp4_difference_count": 0,
"fp4_equal": true,
"block_scale_difference_count": 0,
"block_scales_equal": true,
"scalar_block_scales_equal": true,
"actual_block_scale_bytes": [
126
],
"expected_block_scale_bytes": [
126
]
},
{
"name": "sparse_extremes",
"tensor_scale": 0.037109375,
"fp4_difference_count": 0,
"fp4_equal": true,
"block_scale_difference_count": 0,
"block_scales_equal": true,
"scalar_block_scales_equal": true,
"actual_block_scale_bytes": [
0,
126
],
"expected_block_scale_bytes": [
0,
126
]
}
]
}