h3-blackwell-runtime/benchmarks/gb10-sage2-temporal-pair-timing-20260825-p0.json

199 lines
5.4 KiB
JSON
Raw Normal View History

{
"metadata": {
"width": 1344,
"height": 768,
"frames": 124,
"steps": 12,
"sampler_step": 1,
"seed": 440420,
"text_tokens": 100,
"block_index": 24,
"attention": "sage2",
"hidden_shape": [
37810,
5376
],
"h_msa_shape": [
37810,
5376
],
"segments": [
[
0,
100,
1
],
[
100,
514,
2
],
[
514,
37810,
0
]
]
},
"source_commit": "d1a57a546c3d395b1ffcbeecc66d81db76f3b4b5",
"reference_output_sha256": "4c666c20f5f8f651158a2ced33ccff08f3bada07665c595b99008d171db30574",
"q_shape": [
1,
37810,
56,
128
],
"scheduler": {
"cta_q": 128,
"cta_k": 64,
"warp_q": 32,
"warp_k": 64,
"threads_per_cta": 128,
"dynamic_shared_memory_bytes": 32768,
"grid_ctas": 16576
},
"rounds": 5,
"warmup": 3,
"iterations_per_round": 10,
"variants": [
{
"name": "baseline",
"path": "/variants/baseline/lib/sageattention/_qattn_sm89.cpython-312-aarch64-linux-gnu.so",
"output_sha256": "4c666c20f5f8f651158a2ced33ccff08f3bada07665c595b99008d171db30574",
"byte_exact": true,
"timing": {
"count": 50,
"mean_s": 0.2445254299926758,
"p50_s": 0.2454372787475586,
"p90_s": 0.24733047790527343,
"p95_s": 0.2478844924926758,
"p99_s": 0.24838713729858397,
"min_s": 0.2394311065673828,
"max_s": 0.24867494201660156
},
"rounds": [
{
"count": 10,
"mean_s": 0.24067847290039063,
"p50_s": 0.24043912506103515,
"p90_s": 0.24227873840332034,
"p95_s": 0.24245605773925782,
"p99_s": 0.2425979132080078,
"min_s": 0.2394311065673828,
"max_s": 0.2426333770751953
},
{
"count": 10,
"mean_s": 0.24265675506591794,
"p50_s": 0.2430414047241211,
"p90_s": 0.2439772232055664,
"p95_s": 0.24400416030883787,
"p99_s": 0.24402570999145506,
"min_s": 0.2410487060546875,
"max_s": 0.24403109741210938
},
{
"count": 10,
"mean_s": 0.24536304473876952,
"p50_s": 0.2454372787475586,
"p90_s": 0.245896842956543,
"p95_s": 0.2459307289123535,
"p99_s": 0.24595783767700197,
"min_s": 0.24456390380859375,
"max_s": 0.24596461486816407
},
{
"count": 10,
"mean_s": 0.24639998474121091,
"p50_s": 0.24632894134521482,
"p90_s": 0.24672851867675782,
"p95_s": 0.2470246139526367,
"p99_s": 0.24726149017333982,
"min_s": 0.24594912719726564,
"max_s": 0.24732070922851562
},
{
"count": 10,
"mean_s": 0.24752889251708984,
"p50_s": 0.2473423309326172,
"p90_s": 0.24814632110595702,
"p95_s": 0.2484106315612793,
"p99_s": 0.24862207992553711,
"min_s": 0.2467729034423828,
"max_s": 0.24867494201660156
}
],
"speedup_vs_baseline_percent": 0.0
},
{
"name": "temporal-pair",
"path": "/variants/temporal-pair/lib/sageattention/_qattn_sm89.cpython-312-aarch64-linux-gnu.so",
"output_sha256": "4c666c20f5f8f651158a2ced33ccff08f3bada07665c595b99008d171db30574",
"byte_exact": true,
"timing": {
"count": 50,
"mean_s": 0.24434641784667968,
"p50_s": 0.24519876861572265,
"p90_s": 0.2470915344238281,
"p95_s": 0.2473576820373535,
"p99_s": 0.24812875885009766,
"min_s": 0.23884719848632813,
"max_s": 0.2482449035644531
},
"rounds": [
{
"count": 10,
"mean_s": 0.24035887298583986,
"p50_s": 0.2405825424194336,
"p90_s": 0.24129256286621092,
"p95_s": 0.24146988220214843,
"p99_s": 0.24161173767089844,
"min_s": 0.23884719848632813,
"max_s": 0.24164720153808594
},
{
"count": 10,
"mean_s": 0.2422781478881836,
"p50_s": 0.2422403030395508,
"p90_s": 0.24364151611328125,
"p95_s": 0.2438028366088867,
"p99_s": 0.2439318930053711,
"min_s": 0.24067245483398436,
"max_s": 0.24396415710449218
},
{
"count": 10,
"mean_s": 0.24521179351806638,
"p50_s": 0.24519876861572265,
"p90_s": 0.2460178466796875,
"p95_s": 0.2461315963745117,
"p99_s": 0.2462225961303711,
"min_s": 0.24424879455566406,
"max_s": 0.24624534606933593
},
{
"count": 10,
"mean_s": 0.24673807983398435,
"p50_s": 0.2465936508178711,
"p90_s": 0.24718292388916016,
"p95_s": 0.2475953987121582,
"p99_s": 0.24792537857055663,
"min_s": 0.24614927673339843,
"max_s": 0.24800787353515624
},
{
"count": 10,
"mean_s": 0.24714519500732424,
"p50_s": 0.24705561828613282,
"p90_s": 0.24748206787109375,
"p95_s": 0.24786348571777345,
"p99_s": 0.24816861999511716,
"min_s": 0.24656553649902344,
"max_s": 0.2482449035644531
}
],
"speedup_vs_baseline_percent": 0.09727215727162175
}
]
}