{ "scenario": "F3", "backend": "flash", "sequence": 41778, "batch": 1, "heads": 28, "head_dim": 128, "pid": 2815985, "torch": "2.11.0+cu130", "cuda": "13.0", "cudnn": 91900, "gpu": "NVIDIA RTX 6000D", "capability": [ 12, 0 ], "started_at": 1787624711.7681105, "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": [ 175.5344696044922, 175.5581512451172, 175.56346130371094, 175.59347534179688, 175.62399291992188, 175.65011596679688, 175.6781768798828, 175.68582153320312, 175.6874542236328, 175.68917846679688, 175.76194763183594, 175.8026580810547 ], "min_ms": 175.5344696044922, "median_ms": 175.66414642333984, "mean_ms": 175.65240859985352, "p90_ms": 175.76194763183594, "max_ms": 175.8026580810547, "stdev_ms": 0.08197885888112125, "cv": 0.00046671070174661764 }, "cold": { "samples_ms": [ 175.55372619628906, 175.5543670654297, 175.5680389404297, 175.68063354492188, 175.75669860839844 ], "min_ms": 175.55372619628906, "median_ms": 175.5680389404297, "mean_ms": 175.62269287109376, "p90_ms": 175.75669860839844, "max_ms": 175.75669860839844, "stdev_ms": 0.09182392036463183, "cv": 0.0005228476961803004 }, "cold_over_warm": 0.9994528907300267, "effective_tflops": 142.44268575513604, "io_lower_bound_gbps": 6.819028472168894, "cuda_graph": { "supported": true, "samples_ms": [ 175.3698272705078, 175.50086975097656, 175.54559326171875, 175.55609130859375, 175.5925750732422, 175.5975341796875, 175.59866333007812, 175.6071014404297, 175.60882568359375, 175.61386108398438, 175.63682556152344, 176.9344024658203 ], "min_ms": 175.3698272705078, "median_ms": 175.5980987548828, "mean_ms": 175.68018086751303, "p90_ms": 175.63682556152344, "max_ms": 176.9344024658203, "stdev_ms": 0.40146506832100187, "cv": 0.0022852040926788528 }, "output_checksum": 2.1508976715267636e-06, "trace": "/data/wxy/profile_results/h3-sdpa-kernel-matrix-20260825-run1/jobs/gpu0-F3-flash-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": [ 175.30905151367188, 175.322021484375, 175.32546997070312, 175.3474884033203, 175.3837127685547, 175.40304565429688, 175.4245147705078, 175.438232421875, 175.5750732421875, 175.5750732421875, 175.58546447753906, 175.6988525390625 ], "min_ms": 175.30905151367188, "median_ms": 175.41378021240234, "mean_ms": 175.4490000406901, "p90_ms": 175.58546447753906, "max_ms": 175.6988525390625, "stdev_ms": 0.12834642610474586, "cv": 0.0007315312488243295 }, "cold": { "samples_ms": [ 175.0148162841797, 175.14720153808594, 175.1736602783203, 175.22560119628906, 175.3553009033203 ], "min_ms": 175.0148162841797, "median_ms": 175.1736602783203, "mean_ms": 175.18331604003907, "p90_ms": 175.3553009033203, "max_ms": 175.3553009033203, "stdev_ms": 0.12364918757043826, "cv": 0.0007058274176187964 }, "cold_over_warm": 0.9986311227442264, "effective_tflops": 142.64599267586422, "io_lower_bound_gbps": 6.82876119851904, "cuda_graph": { "supported": true, "samples_ms": [ 175.08787536621094, 175.10890197753906, 175.10963439941406, 175.11289978027344, 175.11900329589844, 175.1321258544922, 175.13241577148438, 175.13539123535156, 175.17955017089844, 175.18316650390625, 175.1844482421875, 175.2764434814453 ], "min_ms": 175.08787536621094, "median_ms": 175.13227081298828, "mean_ms": 175.14682133992514, "p90_ms": 175.1844482421875, "max_ms": 175.2764434814453, "stdev_ms": 0.05168422112012699, "cv": 0.00029509083136495065 }, "output_checksum": 2.1051430394436466e-06, "trace": "/data/wxy/profile_results/h3-sdpa-kernel-matrix-20260825-run1/jobs/gpu0-F3-flash-contiguous.trace.json", "trace_error": null } ], "status": "success", "finished_at": 1787624726.5346925 }