h3-blackwell-runtime/benchmarks/gb10-sage2-p1-real-parity.json

516 lines
12 KiB
JSON
Raw Normal View History

{
"device": "NVIDIA GB10",
"torch": "2.9.1+cu130",
"metadata": {
"resolution": [
1344,
768
],
"frames": 124,
"tokens": 37810,
"seed": 440420,
"blocks": [
0,
24,
49
]
},
"randomized": [
{
"sequence": 1,
"heads": 2,
"q_prepared": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"k_prepared": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"q_int8": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"q_scale": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
}
},
{
"sequence": 31,
"heads": 2,
"q_prepared": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"k_prepared": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"q_int8": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"q_scale": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
}
},
{
"sequence": 32,
"heads": 2,
"q_prepared": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"k_prepared": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"q_int8": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"q_scale": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
}
},
{
"sequence": 33,
"heads": 2,
"q_prepared": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"k_prepared": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"q_int8": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"q_scale": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
}
},
{
"sequence": 127,
"heads": 2,
"q_prepared": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"k_prepared": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"q_int8": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"q_scale": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
}
},
{
"sequence": 128,
"heads": 2,
"q_prepared": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"k_prepared": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"q_int8": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"q_scale": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
}
},
{
"sequence": 129,
"heads": 2,
"q_prepared": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"k_prepared": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"q_int8": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"q_scale": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
}
}
],
"real": [
{
"block": 0,
"q_shape": [
1,
37810,
56,
128
],
"q_stride": [
813066240,
21504,
128,
1
],
"q_prepared": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"k_prepared": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"q_int8": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"q_scale": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"k_mean": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"k_int8": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"k_scale": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"attention_output": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"baseline_entry": {
"timing": {
"count": 5,
"mean_s": 0.015943762969970705,
"p50_s": 0.01610553550720215,
"p90_s": 0.016237541961669922,
"p95_s": 0.016242866897583006,
"p99_s": 0.016247126846313478,
"min_s": 0.015570015907287597,
"max_s": 0.016248191833496094
},
"allocated_bytes_before": 12271139840,
"reserved_bytes_before": 18274582528,
"peak_allocated_bytes": 12542427136,
"peak_reserved_bytes": 18274582528,
"peak_allocated_delta_bytes": 271287296,
"peak_reserved_delta_bytes": 0
},
"candidate_entry": {
"timing": {
"count": 5,
"mean_s": 0.01246666259765625,
"p50_s": 0.012329312324523925,
"p90_s": 0.012692352104187012,
"p95_s": 0.01272153606414795,
"p99_s": 0.0127448832321167,
"min_s": 0.012323200225830078,
"max_s": 0.012750720024108887
},
"allocated_bytes_before": 12271139840,
"reserved_bytes_before": 18274582528,
"peak_allocated_bytes": 12542427136,
"peak_reserved_bytes": 18274582528,
"peak_allocated_delta_bytes": 271287296,
"peak_reserved_delta_bytes": 0
}
},
{
"block": 24,
"q_shape": [
1,
37810,
56,
128
],
"q_stride": [
813066240,
21504,
128,
1
],
"q_prepared": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"k_prepared": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"q_int8": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"q_scale": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"k_mean": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"k_int8": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"k_scale": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"attention_output": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"baseline_entry": {
"timing": {
"count": 5,
"mean_s": 0.015854873847961425,
"p50_s": 0.015605952262878418,
"p90_s": 0.016245626068115235,
"p95_s": 0.016249741363525392,
"p99_s": 0.016253033599853514,
"min_s": 0.015590432167053223,
"max_s": 0.016253856658935546
},
"allocated_bytes_before": 12271885312,
"reserved_bytes_before": 20443037696,
"peak_allocated_bytes": 12544190464,
"peak_reserved_bytes": 20443037696,
"peak_allocated_delta_bytes": 272305152,
"peak_reserved_delta_bytes": 0
},
"candidate_entry": {
"timing": {
"count": 5,
"mean_s": 0.012546982383728028,
"p50_s": 0.012344767570495605,
"p90_s": 0.012899776077270508,
"p95_s": 0.012985503959655762,
"p99_s": 0.013054086265563966,
"min_s": 0.012335616111755371,
"max_s": 0.013071231842041016
},
"allocated_bytes_before": 12271885312,
"reserved_bytes_before": 20443037696,
"peak_allocated_bytes": 12544190464,
"peak_reserved_bytes": 20443037696,
"peak_allocated_delta_bytes": 272305152,
"peak_reserved_delta_bytes": 0
}
},
{
"block": 49,
"q_shape": [
1,
37810,
56,
128
],
"q_stride": [
813066240,
21504,
128,
1
],
"q_prepared": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"k_prepared": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"q_int8": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"q_scale": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"k_mean": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"k_int8": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"k_scale": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"attention_output": {
"equal": true,
"different_elements": 0,
"max_abs": 0.0,
"mean_abs": 0.0
},
"baseline_entry": {
"timing": {
"count": 5,
"mean_s": 0.015745088005065917,
"p50_s": 0.01554531192779541,
"p90_s": 0.01607791976928711,
"p95_s": 0.016113983535766603,
"p99_s": 0.016142834548950193,
"min_s": 0.01552620792388916,
"max_s": 0.016150047302246093
},
"allocated_bytes_before": 12271139840,
"reserved_bytes_before": 20443037696,
"peak_allocated_bytes": 12542427136,
"peak_reserved_bytes": 20443037696,
"peak_allocated_delta_bytes": 271287296,
"peak_reserved_delta_bytes": 0
},
"candidate_entry": {
"timing": {
"count": 5,
"mean_s": 0.012486035346984864,
"p50_s": 0.012309951782226563,
"p90_s": 0.012827526664733887,
"p95_s": 0.012944339561462402,
"p99_s": 0.013037789878845216,
"min_s": 0.01228332805633545,
"max_s": 0.013061152458190919
},
"allocated_bytes_before": 12271139840,
"reserved_bytes_before": 20443037696,
"peak_allocated_bytes": 12542427136,
"peak_reserved_bytes": 20443037696,
"peak_allocated_delta_bytes": 271287296,
"peak_reserved_delta_bytes": 0
}
}
]
}