| {"ts": "2025-10-31T20:13:46Z", "run": "0cb1af490a594cbca21d4dd4012a3c10", "impl": "torch_flash_ma", "tags": {"family": "torch-sdpa", "backend": "FLASH", "compile": "max-autotune"}, "wl": {"name": "cuda_attn_L128_bfloat16", "batch": 1, "seq_len": 4224, "heads": 24, "head_dim": 128, "dtype": "bfloat16", "device": "cuda", "seed": 0}, "env": {"torch": "2.8.0+cu128", "cuda": "12.8", "gpu": "NVIDIA L40S", "sm": "8.9", "py": "3.11.14", "plat": "Linux-5.10.244-240.970.amzn2.x86_64-x86_64-with-glibc2.35"}, "lat_ms": {"p10": 1.208432000112225, "p50": 1.215130999980829, "p90": 1.2198710001030122, "mean": 1.215487200033749, "iqr": 0.006680000069536618, "raw_times": [1.2208109999392036, 1.208432000112225, 1.2198710001030122, 1.2131910000334756, 1.215130999980829], "has_warnings": false, "reps": 5, "warmup": 2}, "compile_ms": 1.2240119999660237, "peak_bytes": 295567360, "ok": true, "absmax": 0.0625, "corr": {"ok": true, "rtol": 0.02, "atol": 0.02, "absmax": 0.0625, "mae": 0.000354766845703125, "mse": 2.7567148208618164e-06, "ref": "sdpa_math_fp32"}, "err": null} | |
| {"ts": "2025-10-31T20:13:46Z", "run": "0cb1af490a594cbca21d4dd4012a3c10", "impl": "torch_flash_ma", "tags": {"family": "torch-sdpa", "backend": "FLASH", "compile": "max-autotune"}, "wl": {"name": "cuda_attn_L256_bfloat16", "batch": 1, "seq_len": 4352, "heads": 24, "head_dim": 128, "dtype": "bfloat16", "device": "cuda", "seed": 0}, "env": {"torch": "2.8.0+cu128", "cuda": "12.8", "gpu": "NVIDIA L40S", "sm": "8.9", "py": "3.11.14", "plat": "Linux-5.10.244-240.970.amzn2.x86_64-x86_64-with-glibc2.35"}, "lat_ms": {"p10": 1.26713200006634, "p50": 1.2766830000146001, "p90": 1.277253000125711, "mean": 1.2749268000789016, "iqr": 0.004750000016429112, "raw_times": [1.277253000125711, 1.26713200006634, 1.2766830000146001, 1.281063000078575, 1.2725030001092819], "has_warnings": false, "reps": 5, "warmup": 2}, "compile_ms": 1.2717629999769997, "peak_bytes": 304742400, "ok": true, "absmax": 0.0625, "corr": {"ok": true, "rtol": 0.02, "atol": 0.02, "absmax": 0.0625, "mae": 0.000354766845703125, "mse": 2.7418136596679688e-06, "ref": "sdpa_math_fp32"}, "err": null} | |
| {"ts": "2025-10-31T20:13:46Z", "run": "0cb1af490a594cbca21d4dd4012a3c10", "impl": "torch_flash_ma", "tags": {"family": "torch-sdpa", "backend": "FLASH", "compile": "max-autotune"}, "wl": {"name": "cuda_attn_L320_bfloat16", "batch": 1, "seq_len": 4416, "heads": 24, "head_dim": 128, "dtype": "bfloat16", "device": "cuda", "seed": 0}, "env": {"torch": "2.8.0+cu128", "cuda": "12.8", "gpu": "NVIDIA L40S", "sm": "8.9", "py": "3.11.14", "plat": "Linux-5.10.244-240.970.amzn2.x86_64-x86_64-with-glibc2.35"}, "lat_ms": {"p10": 1.2928539999847999, "p50": 1.3003640001443273, "p90": 1.3163240000721999, "mean": 1.3067478000721167, "iqr": 0.01689100008661626, "raw_times": [1.3003640001443273, 1.2928539999847999, 1.2994329999855836, 1.3163240000721999, 1.3247640001736727], "has_warnings": false, "reps": 5, "warmup": 2}, "compile_ms": 1.3026630001604644, "peak_bytes": 307494912, "ok": true, "absmax": 0.0625, "corr": {"ok": true, "rtol": 0.02, "atol": 0.02, "absmax": 0.0625, "mae": 0.000354766845703125, "mse": 2.7567148208618164e-06, "ref": "sdpa_math_fp32"}, "err": null} | |
| {"ts": "2025-10-31T20:13:46Z", "run": "0cb1af490a594cbca21d4dd4012a3c10", "impl": "torch_flash_ma", "tags": {"family": "torch-sdpa", "backend": "FLASH", "compile": "max-autotune"}, "wl": {"name": "cuda_attn_L384_bfloat16", "batch": 1, "seq_len": 4480, "heads": 24, "head_dim": 128, "dtype": "bfloat16", "device": "cuda", "seed": 0}, "env": {"torch": "2.8.0+cu128", "cuda": "12.8", "gpu": "NVIDIA L40S", "sm": "8.9", "py": "3.11.14", "plat": "Linux-5.10.244-240.970.amzn2.x86_64-x86_64-with-glibc2.35"}, "lat_ms": {"p10": 1.3232850001259067, "p50": 1.3295650001055037, "p90": 1.3361950000216893, "mean": 1.332684600038192, "iqr": 0.007890999995652237, "raw_times": [1.328304000026037, 1.3361950000216893, 1.3295650001055037, 1.3232850001259067, 1.3460739999118232], "has_warnings": false, "reps": 5, "warmup": 2}, "compile_ms": 1.3245140000890387, "peak_bytes": 311296000, "ok": true, "absmax": 0.0625, "corr": {"ok": true, "rtol": 0.02, "atol": 0.02, "absmax": 0.0625, "mae": 0.000354766845703125, "mse": 2.7418136596679688e-06, "ref": "sdpa_math_fp32"}, "err": null} | |
| {"ts": "2025-10-31T20:13:46Z", "run": "0cb1af490a594cbca21d4dd4012a3c10", "impl": "torch_flash_ma", "tags": {"family": "torch-sdpa", "backend": "FLASH", "compile": "max-autotune"}, "wl": {"name": "cuda_attn_L448_bfloat16", "batch": 1, "seq_len": 4544, "heads": 24, "head_dim": 128, "dtype": "bfloat16", "device": "cuda", "seed": 0}, "env": {"torch": "2.8.0+cu128", "cuda": "12.8", "gpu": "NVIDIA L40S", "sm": "8.9", "py": "3.11.14", "plat": "Linux-5.10.244-240.970.amzn2.x86_64-x86_64-with-glibc2.35"}, "lat_ms": {"p10": 1.4790479999646777, "p50": 1.4950690001569455, "p90": 1.4989779999723396, "mean": 1.4914904000306706, "iqr": 0.017840000055002747, "raw_times": [1.5032190001420531, 1.4950690001569455, 1.4790479999646777, 1.4811379999173369, 1.4989779999723396], "has_warnings": false, "reps": 5, "warmup": 2}, "compile_ms": 1.5107090000583412, "peak_bytes": 315621376, "ok": true, "absmax": 0.0625, "corr": {"ok": true, "rtol": 0.02, "atol": 0.02, "absmax": 0.0625, "mae": 0.000354766845703125, "mse": 2.7418136596679688e-06, "ref": "sdpa_math_fp32"}, "err": null} | |
| {"ts": "2025-10-31T20:13:46Z", "run": "0cb1af490a594cbca21d4dd4012a3c10", "impl": "torch_flash_ma", "tags": {"family": "torch-sdpa", "backend": "FLASH", "compile": "max-autotune"}, "wl": {"name": "cuda_attn_L512_bfloat16", "batch": 1, "seq_len": 4608, "heads": 24, "head_dim": 128, "dtype": "bfloat16", "device": "cuda", "seed": 0}, "env": {"torch": "2.8.0+cu128", "cuda": "12.8", "gpu": "NVIDIA L40S", "sm": "8.9", "py": "3.11.14", "plat": "Linux-5.10.244-240.970.amzn2.x86_64-x86_64-with-glibc2.35"}, "lat_ms": {"p10": 1.511368999899787, "p50": 1.5117090001695033, "p90": 1.512698999931672, "mean": 1.516499199988175, "iqr": 0.00113999999484804, "raw_times": [1.511368999899787, 1.512698999931672, 1.5117090001695033, 1.511558999936824, 1.5351600000030885], "has_warnings": false, "reps": 5, "warmup": 2}, "compile_ms": 1.5183190000698232, "peak_bytes": 319946752, "ok": true, "absmax": 0.125, "corr": {"ok": true, "rtol": 0.02, "atol": 0.02, "absmax": 0.125, "mae": 0.0003566741943359375, "mse": 2.7567148208618164e-06, "ref": "sdpa_math_fp32"}, "err": null} | |