h3-blackwell-runtime/benchmarks/gb10-sage2-vprep-randomized.json

358 lines
7.1 KiB
JSON
Raw Normal View History

{
"status": "pass",
"contract": {
"tensor_layout": "NHD",
"head_dim": 128,
"scale_max": 2.25,
"output": "Sage2 padded/permuted E4M3 V and FP32 per-channel scale"
},
"cases": [
{
"sequence": 1,
"heads": 56,
"stride": [
7168,
7168,
128,
1
],
"fp8": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"scale": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
}
},
{
"sequence": 31,
"heads": 56,
"stride": [
666624,
21504,
128,
1
],
"fp8": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"scale": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
}
},
{
"sequence": 32,
"heads": 56,
"stride": [
688128,
21504,
128,
1
],
"fp8": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"scale": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
}
},
{
"sequence": 33,
"heads": 56,
"stride": [
709632,
21504,
128,
1
],
"fp8": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"scale": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
}
},
{
"sequence": 63,
"heads": 56,
"stride": [
1354752,
21504,
128,
1
],
"fp8": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"scale": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
}
},
{
"sequence": 64,
"heads": 56,
"stride": [
1376256,
21504,
128,
1
],
"fp8": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"scale": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
}
},
{
"sequence": 65,
"heads": 56,
"stride": [
1397760,
21504,
128,
1
],
"fp8": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"scale": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
}
},
{
"sequence": 127,
"heads": 56,
"stride": [
2731008,
21504,
128,
1
],
"fp8": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"scale": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
}
},
{
"sequence": 128,
"heads": 56,
"stride": [
2752512,
21504,
128,
1
],
"fp8": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"scale": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
}
},
{
"sequence": 129,
"heads": 56,
"stride": [
2774016,
21504,
128,
1
],
"fp8": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"scale": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
}
},
{
"sequence": 37760,
"heads": 56,
"stride": [
811991040,
21504,
128,
1
],
"fp8": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"scale": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"baseline_timing": {
"count": 8,
"mean_s": 0.010765039920806884,
"p50_s": 0.01076302433013916,
"p90_s": 0.010959852409362793,
"p95_s": 0.010992982006072998,
"p99_s": 0.011019485683441162,
"min_s": 0.010565183639526366,
"max_s": 0.011026111602783203
},
"candidate_timing": {
"count": 8,
"mean_s": 0.006614172041416168,
"p50_s": 0.006597951889038086,
"p90_s": 0.006795004796981812,
"p95_s": 0.006805454468727112,
"p99_s": 0.0068138142061233515,
"min_s": 0.006458144187927246,
"max_s": 0.006815904140472412
}
},
{
"sequence": 37761,
"heads": 56,
"stride": [
812012544,
21504,
128,
1
],
"fp8": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"scale": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"baseline_timing": {
"count": 8,
"mean_s": 0.0109060640335083,
"p50_s": 0.010748064041137696,
"p90_s": 0.011263570976257326,
"p95_s": 0.011430585479736327,
"p99_s": 0.011564197082519531,
"min_s": 0.010667872428894044,
"max_s": 0.011597599983215332
},
"candidate_timing": {
"count": 8,
"mean_s": 0.0065507440567016606,
"p50_s": 0.006485184192657471,
"p90_s": 0.006748016023635865,
"p95_s": 0.006786712002754211,
"p99_s": 0.006817668786048889,
"min_s": 0.006461887836456299,
"max_s": 0.0068254079818725584
}
},
{
"sequence": 37810,
"heads": 56,
"stride": [
813066240,
21504,
128,
1
],
"fp8": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"scale": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"baseline_timing": {
"count": 8,
"mean_s": 0.01064847207069397,
"p50_s": 0.010628528118133545,
"p90_s": 0.010686016178131105,
"p95_s": 0.010728799867630005,
"p99_s": 0.010763026819229126,
"min_s": 0.010623680114746094,
"max_s": 0.010771583557128907
},
"candidate_timing": {
"count": 8,
"mean_s": 0.006505947947502136,
"p50_s": 0.006436048030853271,
"p90_s": 0.00664213752746582,
"p95_s": 0.006790044784545898,
"p99_s": 0.006908370590209961,
"min_s": 0.006423327922821045,
"max_s": 0.006937952041625977
}
}
]
}