{ "scenario": "F3", "backend": "auto", "sequence": 41778, "batch": 1, "heads": 28, "head_dim": 128, "pid": 2815988, "torch": "2.11.0+cu130", "cuda": "13.0", "cudnn": 91900, "gpu": "NVIDIA RTX 6000D", "capability": [ 12, 0 ], "started_at": 1787624711.5848253, "results": [ { "layout": "model_strided", "dtype": "torch.bfloat16", "q_shape": [ 1, 28, 41778, 128 ], "q_stride": [ 449197056, 128, 10752, 1 ], "k_stride": [ 449197056, 128, 10752, 1 ], "v_stride": [ 449197056, 128, 10752, 1 ], "q_contiguous": false, "warm": { "samples_ms": [ 175.44146728515625, 175.4617919921875, 175.4637451171875, 175.46966552734375, 175.47244262695312, 175.4766387939453, 175.503173828125, 175.52345275878906, 175.5325164794922, 175.5386199951172, 175.54576110839844, 175.62742614746094 ], "min_ms": 175.44146728515625, "median_ms": 175.48990631103516, "mean_ms": 175.50472513834634, "p90_ms": 175.54576110839844, "max_ms": 175.62742614746094, "stdev_ms": 0.05187238846501639, "cv": 0.00029556120739271597 }, "cold": { "samples_ms": [ 175.39590454101562, 175.42054748535156, 175.4443817138672, 175.48626708984375, 175.5114288330078 ], "min_ms": 175.39590454101562, "median_ms": 175.4443817138672, "mean_ms": 175.4517059326172, "p90_ms": 175.5114288330078, "max_ms": 175.5114288330078, "stdev_ms": 0.04716881051309542, "cv": 0.00026884213101473496 }, "cold_over_warm": 0.9997405856660082, "effective_tflops": 142.58411400069545, "io_lower_bound_gbps": 6.825798937272987, "cuda_graph": { "supported": true, "samples_ms": [ 175.3736572265625, 175.3738250732422, 175.4015350341797, 175.40924072265625, 175.44566345214844, 175.48825073242188, 175.48886108398438, 175.4950408935547, 175.4990692138672, 175.52517700195312, 175.5259246826172, 175.55328369140625 ], "min_ms": 175.3736572265625, "median_ms": 175.48855590820312, "mean_ms": 175.46496073404947, "p90_ms": 175.5259246826172, "max_ms": 175.55328369140625, "stdev_ms": 0.06213428653591246, "cv": 0.00035411221862174973 }, "output_checksum": 2.1508976715267636e-06, "trace": "/data/wxy/profile_results/h3-sdpa-kernel-matrix-20260825-run1/jobs/gpu3-F3-auto-model_strided.trace.json", "trace_error": null }, { "layout": "contiguous", "dtype": "torch.bfloat16", "q_shape": [ 1, 28, 41778, 128 ], "q_stride": [ 149732352, 5347584, 128, 1 ], "k_stride": [ 149732352, 5347584, 128, 1 ], "v_stride": [ 149732352, 5347584, 128, 1 ], "q_contiguous": true, "warm": { "samples_ms": [ 175.44003295898438, 175.4427490234375, 175.46585083007812, 175.4673309326172, 175.46815490722656, 175.4784393310547, 175.48629760742188, 175.49551391601562, 175.50111389160156, 175.53024291992188, 175.53753662109375, 175.55616760253906 ], "min_ms": 175.44003295898438, "median_ms": 175.48236846923828, "mean_ms": 175.48911921183267, "p90_ms": 175.53753662109375, "max_ms": 175.55616760253906, "stdev_ms": 0.0367460335903592, "cv": 0.00020939209083386599 }, "cold": { "samples_ms": [ 175.28627014160156, 175.37564086914062, 175.38265991210938, 175.44447326660156, 175.54141235351562 ], "min_ms": 175.28627014160156, "median_ms": 175.38265991210938, "mean_ms": 175.40609130859374, "p90_ms": 175.54141235351562, "max_ms": 175.54141235351562, "stdev_ms": 0.09436409746657848, "cv": 0.0005379750313263793 }, "cold_over_warm": 0.9994318029896754, "effective_tflops": 142.59023869859791, "io_lower_bound_gbps": 6.826092139336393, "cuda_graph": { "supported": true, "samples_ms": [ 175.12002563476562, 175.12786865234375, 175.12991333007812, 175.13381958007812, 175.15740966796875, 175.20147705078125, 175.2263641357422, 175.29405212402344, 175.31491088867188, 175.41734313964844, 175.46829223632812, 175.48562622070312 ], "min_ms": 175.12002563476562, "median_ms": 175.21392059326172, "mean_ms": 175.25642522176108, "p90_ms": 175.46829223632812, "max_ms": 175.48562622070312, "stdev_ms": 0.1375261146588285, "cv": 0.0007847136815943241 }, "output_checksum": 2.1051430394436466e-06, "trace": "/data/wxy/profile_results/h3-sdpa-kernel-matrix-20260825-run1/jobs/gpu3-F3-auto-contiguous.trace.json", "trace_error": null } ], "status": "success", "finished_at": 1787624726.5770328 }