{
  "device": {
    "jax": "0.11.2",
    "backend": "gpu",
    "devices": [
      "cuda:0"
    ],
    "python": "3.13.18",
    "device_kind": "NVIDIA H100"
  },
  "config": {
    "output": "/home/tu/tu_tu/tu_iivgl01/tabdiffpfn/experiments/results/h100_branch/attention_tiles.json",
    "append": false,
    "dtypes": [
      "bfloat16",
      "float32"
    ],
    "head_dims": [
      64,
      128
    ],
    "seq_lens": [
      2048,
      8192
    ],
    "modes": [
      "fwd",
      "fwd_bwd"
    ],
    "batch": 4,
    "heads": 8,
    "repeats": 5,
    "auto_only": false,
    "tag": "h100_nvl",
    "fwd_block_q": [
      32,
      64,
      128
    ],
    "fwd_block_k": [
      32,
      64,
      128
    ],
    "fwd_stages": [
      1,
      2,
      3
    ],
    "fwd_warps": [
      4,
      8
    ],
    "bwd_outer": [
      32,
      64,
      128
    ],
    "bwd_inner": [
      16,
      32,
      64,
      128
    ],
    "bwd_stages": [
      1,
      2
    ],
    "bwd_warps": [
      4,
      8
    ],
    "bwd_fwd_config": null,
    "smem_limit_bytes": 232448
  },
  "results": [
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "auto",
      "rel_err": 0.002961658174172044,
      "compile_s": 0.46319316513836384,
      "inner": 200,
      "median_ms": 0.12052880483679473,
      "q1_ms": 0.12052005506120622,
      "q3_ms": 0.12468614033423364
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "auto",
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.5782660259865224,
      "inner": 77,
      "median_ms": 0.5764431537842596,
      "q1_ms": 0.530373559079387,
      "q3_ms": 0.5802679474332503
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.14845989481545985,
      "inner": 96,
      "median_ms": 0.41877515711045515,
      "q1_ms": 0.4145257917116396,
      "q3_ms": 0.4250805504852906
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.13420620583929121,
      "inner": 66,
      "median_ms": 0.6644364241792848,
      "q1_ms": 0.6628722114714257,
      "q3_ms": 0.6670921681110155
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.15394530002959073,
      "inner": 108,
      "median_ms": 0.394491371008809,
      "q1_ms": 0.3897664805494801,
      "q3_ms": 0.39718460067416783
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.14002714888192713,
      "inner": 67,
      "median_ms": 0.6614615376204698,
      "q1_ms": 0.6470763604087171,
      "q3_ms": 0.6645673420280218
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.15849723597057164,
      "inner": 107,
      "median_ms": 0.39826353106682544,
      "q1_ms": 0.39472419629308664,
      "q3_ms": 0.3992930646533164
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.1442433469928801,
      "inner": 68,
      "median_ms": 0.6600925151039573,
      "q1_ms": 0.6428268388845026,
      "q3_ms": 0.6602370755418259
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.15908661391586065,
      "inner": 118,
      "median_ms": 0.3450883303816288,
      "q1_ms": 0.3357694653983591,
      "q3_ms": 0.34671917881311504
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.14199354406446218,
      "inner": 84,
      "median_ms": 0.4930629871696944,
      "q1_ms": 0.4927541899849617,
      "q3_ms": 0.49404846227151294
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.1704583829268813,
      "inner": 133,
      "median_ms": 0.31297000762271254,
      "q1_ms": 0.30488348951129085,
      "q3_ms": 0.31482511291042303
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.149482652079314,
      "inner": 89,
      "median_ms": 0.4642542917281389,
      "q1_ms": 0.4600278870987423,
      "q3_ms": 0.46809135904807725
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.17563276714645326,
      "inner": 131,
      "median_ms": 0.31085444955539154,
      "q1_ms": 0.3098959853261254,
      "q3_ms": 0.31659267984956274
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.15632677194662392,
      "inner": 90,
      "median_ms": 0.46163334562960595,
      "q1_ms": 0.45262281006822985,
      "q3_ms": 0.4621741232565708
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.002961658174172044,
      "compile_s": 0.19784454395994544,
      "inner": 134,
      "median_ms": 0.29485221646392523,
      "q1_ms": 0.28599530750357394,
      "q3_ms": 0.2972157915303511
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.002961658174172044,
      "compile_s": 0.16758497105911374,
      "inner": 91,
      "median_ms": 0.405319605616259,
      "q1_ms": 0.40526818586411056,
      "q3_ms": 0.4063213607398691
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.002961658174172044,
      "compile_s": 0.2095520910806954,
      "inner": 133,
      "median_ms": 0.2675860073361406,
      "q1_ms": 0.2674639708173454,
      "q3_ms": 0.2784552192665581
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.002961658174172044,
      "compile_s": 0.17067255219444633,
      "inner": 101,
      "median_ms": 0.37420634871883557,
      "q1_ms": 0.37089001968281693,
      "q3_ms": 0.37988258461182073
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.002961658174172044,
      "compile_s": 0.2200074999127537,
      "inner": 136,
      "median_ms": 0.2748928159502719,
      "q1_ms": 0.27122951509869275,
      "q3_ms": 0.27628045577062843
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.002961658174172044,
      "compile_s": 0.18025455391034484,
      "inner": 105,
      "median_ms": 0.36523790497864994,
      "q1_ms": 0.3646308667070809,
      "q3_ms": 0.3681605703951347
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.1542573580518365,
      "inner": 168,
      "median_ms": 0.1843291129695163,
      "q1_ms": 0.17813591763288494,
      "q3_ms": 0.1856399467214942
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.13604988111183047,
      "inner": 132,
      "median_ms": 0.28971030913067586,
      "q1_ms": 0.2805085140139316,
      "q3_ms": 0.2914084841243246
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.16464490885846317,
      "inner": 183,
      "median_ms": 0.1745204265672164,
      "q1_ms": 0.17022869493710538,
      "q3_ms": 0.1793945134398553
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.14517518482171,
      "inner": 129,
      "median_ms": 0.29355512441013215,
      "q1_ms": 0.2892464561777752,
      "q3_ms": 0.2959713939028655
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.1680897988844663,
      "inner": 192,
      "median_ms": 0.1703666457615327,
      "q1_ms": 0.15949991090261997,
      "q3_ms": 0.17064086447741525
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.15103943808935583,
      "inner": 137,
      "median_ms": 0.27966941706836224,
      "q1_ms": 0.27886161369944573,
      "q3_ms": 0.2898727509662183
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.17330692382529378,
      "inner": 182,
      "median_ms": 0.15064181790656442,
      "q1_ms": 0.14099372028895132,
      "q3_ms": 0.15280698298138903
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.1787671889178455,
      "inner": 147,
      "median_ms": 0.2424005311311913,
      "q1_ms": 0.23892346205709336,
      "q3_ms": 0.25092391138832987
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.19147830200381577,
      "inner": 198,
      "median_ms": 0.14325327856108697,
      "q1_ms": 0.13508299963003156,
      "q3_ms": 0.14461549422280356
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.19751994498074055,
      "inner": 157,
      "median_ms": 0.23044494223917364,
      "q1_ms": 0.22331394913612276,
      "q3_ms": 0.2397088078888738
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.22737210895866156,
      "inner": 200,
      "median_ms": 0.14190988964401186,
      "q1_ms": 0.1306603301782161,
      "q3_ms": 0.14394748490303755
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.16814958886243403,
      "inner": 151,
      "median_ms": 0.23050695374015942,
      "q1_ms": 0.22653121168151596,
      "q3_ms": 0.24085547096583226
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.002961658174172044,
      "compile_s": 0.24608365586027503,
      "inner": 184,
      "median_ms": 0.15070788567359358,
      "q1_ms": 0.14346301095033795,
      "q3_ms": 0.15362722257597616
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.002961658174172044,
      "compile_s": 0.22298122197389603,
      "inner": 160,
      "median_ms": 0.2390814377577044,
      "q1_ms": 0.23130383633542806,
      "q3_ms": 0.2390987501712516
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.002961658174172044,
      "compile_s": 0.2497525259386748,
      "inner": 200,
      "median_ms": 0.12877393513917923,
      "q1_ms": 0.12590619502589107,
      "q3_ms": 0.14214883907698095
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.002961658174172044,
      "compile_s": 0.21068673813715577,
      "inner": 160,
      "median_ms": 0.2268682248541154,
      "q1_ms": 0.22114668099675328,
      "q3_ms": 0.22792985109845176
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.002961658174172044,
      "compile_s": 0.2612822721712291,
      "inner": 200,
      "median_ms": 0.1418708951678127,
      "q1_ms": 0.13507171417586505,
      "q3_ms": 0.1439576898701489
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.002961658174172044,
      "compile_s": 0.22043417999520898,
      "inner": 170,
      "median_ms": 0.21858931765617692,
      "q1_ms": 0.21414597628309445,
      "q3_ms": 0.22109883481307943
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.2066576590295881,
      "inner": 200,
      "median_ms": 0.14895107480697334,
      "q1_ms": 0.1423707406502217,
      "q3_ms": 0.15153141575865448
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.14903972391039133,
      "inner": 174,
      "median_ms": 0.19555036299701395,
      "q1_ms": 0.19439416081141467,
      "q3_ms": 0.1958092693346499
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.21990699088200927,
      "inner": 179,
      "median_ms": 0.18209836305269983,
      "q1_ms": 0.1775974800612507,
      "q3_ms": 0.18343696105371973
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.16017500101588666,
      "inner": 184,
      "median_ms": 0.17910242414749833,
      "q1_ms": 0.17487216214446918,
      "q3_ms": 0.1792057402148519
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.2234578609932214,
      "inner": 200,
      "median_ms": 0.1563436002470553,
      "q1_ms": 0.15200196532532573,
      "q3_ms": 0.15989419422112405
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.1655413571279496,
      "inner": 184,
      "median_ms": 0.16026752722530585,
      "q1_ms": 0.1511654396460432,
      "q3_ms": 0.16196692392796924
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.2617359219584614,
      "inner": 200,
      "median_ms": 0.15167421428486705,
      "q1_ms": 0.14518148498609662,
      "q3_ms": 0.1519498648121953
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.16655408288352191,
      "inner": 200,
      "median_ms": 0.1430011447519064,
      "q1_ms": 0.13479621498845518,
      "q3_ms": 0.14441373525187373
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.2755441819317639,
      "inner": 200,
      "median_ms": 0.1338973140809685,
      "q1_ms": 0.1337636192329228,
      "q3_ms": 0.14470608555711806
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.1819174790289253,
      "inner": 200,
      "median_ms": 0.1286876341328025,
      "q1_ms": 0.12682598549872637,
      "q3_ms": 0.14099149964749813
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.2841460769996047,
      "inner": 200,
      "median_ms": 0.12998373014852405,
      "q1_ms": 0.11860365979373455,
      "q3_ms": 0.13221886940300465
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0029095381032675505,
      "compile_s": 0.185633938992396,
      "inner": 200,
      "median_ms": 0.1401248446200043,
      "q1_ms": 0.1309140247758478,
      "q3_ms": 0.14164249529130757
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.002961658174172044,
      "compile_s": 0.4574567209929228,
      "inner": 200,
      "median_ms": 0.13932554982602596,
      "q1_ms": 0.1304093305952847,
      "q3_ms": 0.14004919910803437
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.002961658174172044,
      "compile_s": 0.22113643703050911,
      "inner": 168,
      "median_ms": 0.1775050241987975,
      "q1_ms": 0.1770209235560504,
      "q3_ms": 0.17901966791777385
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.002961658174172044,
      "compile_s": 0.4253086829558015,
      "inner": 200,
      "median_ms": 0.13315881951712072,
      "q1_ms": 0.12323369970545171,
      "q3_ms": 0.13599936035461724
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.002961658174172044,
      "compile_s": 0.23398672114126384,
      "inner": 200,
      "median_ms": 0.15846339403651655,
      "q1_ms": 0.15817029983736575,
      "q3_ms": 0.1626997848507017
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.002961658174172044,
      "compile_s": 0.48384992592036724,
      "inner": 195,
      "median_ms": 0.14512260945943686,
      "q1_ms": 0.13139998444761986,
      "q3_ms": 0.1460112213419798
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.002961658174172044,
      "compile_s": 0.24643560498952866,
      "inner": 200,
      "median_ms": 0.14546263497322798,
      "q1_ms": 0.14315319480374455,
      "q3_ms": 0.14770932495594025
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.7093704307917506,
      "inner": 29,
      "median_ms": 1.539659718501157,
      "q1_ms": 1.5327694087192931,
      "q3_ms": 1.5620027973862558
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.3173042822163552,
      "inner": 19,
      "median_ms": 2.386114265965788,
      "q1_ms": 2.376769522303029,
      "q3_ms": 2.398587371173658
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.5802227910608053,
      "inner": 34,
      "median_ms": 1.3349794111598063,
      "q1_ms": 1.3219712315784657,
      "q3_ms": 1.3680404997156823
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.3401731359772384,
      "inner": 19,
      "median_ms": 2.372203725635221,
      "q1_ms": 2.371063210854405,
      "q3_ms": 2.37577425708112
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.5468581849709153,
      "inner": 43,
      "median_ms": 1.0961405770463306,
      "q1_ms": 1.08448293081723,
      "q3_ms": 1.0975838346450135
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.304593687877059,
      "inner": 25,
      "median_ms": 1.828361600637436,
      "q1_ms": 1.8224896024912596,
      "q3_ms": 1.866446239873767
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.5717733649071306,
      "inner": 49,
      "median_ms": 0.946948304772377,
      "q1_ms": 0.9320959144709062,
      "q3_ms": 0.9499495297822417
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.32032088399864733,
      "inner": 28,
      "median_ms": 1.6548963197107827,
      "q1_ms": 1.653978501313499,
      "q3_ms": 1.661997745811407
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.562952539883554,
      "inner": 45,
      "median_ms": 1.0090764456739028,
      "q1_ms": 1.001610689693027,
      "q3_ms": 1.025636401027441
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.3055070359259844,
      "inner": 35,
      "median_ms": 1.3451581488230397,
      "q1_ms": 1.3160491162644967,
      "q3_ms": 1.3475515680121526
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.5913465381599963,
      "inner": 55,
      "median_ms": 0.8800464004955509,
      "q1_ms": 0.8284327108412981,
      "q3_ms": 0.8808029265227643
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.36370950983837247,
      "inner": 39,
      "median_ms": 1.1870058444448006,
      "q1_ms": 1.170378178358078,
      "q3_ms": 1.1961955386094558
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.5805863889399916,
      "inner": 45,
      "median_ms": 1.0046520218667057,
      "q1_ms": 0.9869649540632963,
      "q3_ms": 1.005632471707132
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.3356649598572403,
      "inner": 33,
      "median_ms": 1.3413779705649977,
      "q1_ms": 1.3389098175773115,
      "q3_ms": 1.3476918790150774
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.5751003751065582,
      "inner": 59,
      "median_ms": 0.7677418613901077,
      "q1_ms": 0.7663866252464763,
      "q3_ms": 0.787142494503977
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.34665606799535453,
      "inner": 39,
      "median_ms": 1.1336244344233704,
      "q1_ms": 1.1278749742091465,
      "q3_ms": 1.1390052043283598
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.5678845650982112,
      "inner": 46,
      "median_ms": 0.9615003060468513,
      "q1_ms": 0.9610159373234796,
      "q3_ms": 0.9631681124396297
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.32409941288642585,
      "inner": 33,
      "median_ms": 1.3526685493574901,
      "q1_ms": 1.3522652443498373,
      "q3_ms": 1.3588130911529968
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.5639617170672864,
      "inner": 49,
      "median_ms": 0.9022168149905545,
      "q1_ms": 0.8858235942541944,
      "q3_ms": 0.9160226933201965
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.32056975294835865,
      "inner": 33,
      "median_ms": 1.369608516776652,
      "q1_ms": 1.35988912150038,
      "q3_ms": 1.3792030301622369
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.5457459888420999,
      "inner": 57,
      "median_ms": 0.7498697908758594,
      "q1_ms": 0.7497353863232492,
      "q3_ms": 0.7530759096864547
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.3119331570342183,
      "inner": 46,
      "median_ms": 1.0010301712495477,
      "q1_ms": 0.9934667184058091,
      "q3_ms": 1.001689087032624
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.5620899118948728,
      "inner": 59,
      "median_ms": 0.7435424387518127,
      "q1_ms": 0.7351517787803028,
      "q3_ms": 0.7478793909362818
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.3309168932028115,
      "inner": 46,
      "median_ms": 0.9849171923554461,
      "q1_ms": 0.9818315232658515,
      "q3_ms": 0.9890643464725304
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.5588081309106201,
      "inner": 72,
      "median_ms": 0.6439616231040822,
      "q1_ms": 0.6417774744073136,
      "q3_ms": 0.6515322925729884
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.34098251396790147,
      "inner": 38,
      "median_ms": 1.1703240207249397,
      "q1_ms": 1.1570293174468371,
      "q3_ms": 1.187395819119717
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.5643566851504147,
      "inner": 80,
      "median_ms": 0.5615167727228254,
      "q1_ms": 0.5509725480806082,
      "q3_ms": 0.5745679751271382
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.3729336829856038,
      "inner": 41,
      "median_ms": 1.082440315768486,
      "q1_ms": 1.0626547595077172,
      "q3_ms": 1.0862854400240793
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.6337594070937485,
      "inner": 67,
      "median_ms": 0.6819390865571019,
      "q1_ms": 0.6612466106107875,
      "q3_ms": 0.6867250594407766
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.44326204201206565,
      "inner": 50,
      "median_ms": 0.87711569853127,
      "q1_ms": 0.8695401158183813,
      "q3_ms": 0.8817520597949624
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.6374498859513551,
      "inner": 85,
      "median_ms": 0.5159697057131459,
      "q1_ms": 0.5159486359094871,
      "q3_ms": 0.5381947054582483
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.44724380085244775,
      "inner": 57,
      "median_ms": 0.80915277398992,
      "q1_ms": 0.8011322648248129,
      "q3_ms": 0.8122246843158153
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.5231229430064559,
      "inner": 54,
      "median_ms": 0.7897345952827621,
      "q1_ms": 0.7867840575744156,
      "q3_ms": 0.7923636499240443
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.28645626991055906,
      "inner": 34,
      "median_ms": 1.3215709409183438,
      "q1_ms": 1.311569762251833,
      "q3_ms": 1.3224059126942473
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.5330659851897508,
      "inner": 58,
      "median_ms": 0.7567755869943006,
      "q1_ms": 0.740928394752073,
      "q3_ms": 0.7569330005810179
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.3085545180365443,
      "inner": 52,
      "median_ms": 0.8278276341465803,
      "q1_ms": 0.8251016708807304,
      "q3_ms": 0.8297845577964416
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.5211394890211523,
      "inner": 70,
      "median_ms": 0.6457593025905746,
      "q1_ms": 0.6454364263585636,
      "q3_ms": 0.6471445883757302
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.30834783893078566,
      "inner": 47,
      "median_ms": 0.9238801282295521,
      "q1_ms": 0.9204020594900594,
      "q3_ms": 0.9254881904083998
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.539122898131609,
      "inner": 69,
      "median_ms": 0.6757482170950675,
      "q1_ms": 0.6750857531754435,
      "q3_ms": 0.6768715372174114
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.3551490851677954,
      "inner": 49,
      "median_ms": 0.8491712470291828,
      "q1_ms": 0.8482224669079391,
      "q3_ms": 0.8494281827719237
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.6643160411622375,
      "inner": 71,
      "median_ms": 0.6553740437391777,
      "q1_ms": 0.6356207181660223,
      "q3_ms": 0.6567768614844117
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.43629620294086635,
      "inner": 52,
      "median_ms": 0.8143515173847286,
      "q1_ms": 0.8130784397228406,
      "q3_ms": 0.8144739997358277
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.6558062739204615,
      "inner": 74,
      "median_ms": 0.6141705388152922,
      "q1_ms": 0.6030238095066838,
      "q3_ms": 0.6168156616842827
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.49851559521630406,
      "inner": 60,
      "median_ms": 0.7549576849366229,
      "q1_ms": 0.7466006985244652,
      "q3_ms": 0.7558128330856562
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0037013222463428974,
      "compile_s": 2.0151352880056947,
      "inner": 44,
      "median_ms": 1.0721573860130527,
      "q1_ms": 1.0613967262377795,
      "q3_ms": 1.0756303408098493
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.5796410921029747,
      "inner": 72,
      "median_ms": 0.6096603465266526,
      "q1_ms": 0.6040568890360495,
      "q3_ms": 0.6124346368273513
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0037013222463428974,
      "compile_s": 1.304162680869922,
      "inner": 42,
      "median_ms": 1.1000237593959485,
      "q1_ms": 1.0782447865321523,
      "q3_ms": 1.10283397537257
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0037013222463428974,
      "compile_s": 0.6514554570894688,
      "inner": 77,
      "median_ms": 0.584542222151702,
      "q1_ms": 0.5835552076966345,
      "q3_ms": 0.596695440494782
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "auto",
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.4205111670307815,
      "inner": 27,
      "median_ms": 1.9586155173817168,
      "q1_ms": 1.8824146267164636,
      "q3_ms": 1.9929283729719893
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "auto",
      "rel_err": 0.004734351299703121,
      "compile_s": 0.5481572421267629,
      "inner": 6,
      "median_ms": 8.75568832270801,
      "q1_ms": 8.338176179677248,
      "q3_ms": 8.76645150128752
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.1464472021907568,
      "inner": 8,
      "median_ms": 6.354771874612197,
      "q1_ms": 6.302704510744661,
      "q3_ms": 6.414582981960848
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.1385211341548711,
      "inner": 4,
      "median_ms": 10.394692711997777,
      "q1_ms": 10.322270507458597,
      "q3_ms": 10.429664980620146
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.15436390787363052,
      "inner": 8,
      "median_ms": 6.221804738743231,
      "q1_ms": 5.999641609378159,
      "q3_ms": 6.272378261201084
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.1428076308220625,
      "inner": 5,
      "median_ms": 10.138367023319006,
      "q1_ms": 10.04317537881434,
      "q3_ms": 10.368662234395742
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.15937335509806871,
      "inner": 8,
      "median_ms": 6.024177739163861,
      "q1_ms": 5.939949245657772,
      "q3_ms": 6.2278471305035055
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.14955797302536666,
      "inner": 5,
      "median_ms": 10.196230793371797,
      "q1_ms": 10.189946833997965,
      "q3_ms": 10.250672837719321
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.15973049378953874,
      "inner": 10,
      "median_ms": 5.094171408563852,
      "q1_ms": 5.019524693489075,
      "q3_ms": 5.181962205097079
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.15627416293136775,
      "inner": 6,
      "median_ms": 7.583658366153638,
      "q1_ms": 7.4372770031914115,
      "q3_ms": 7.692029660878082
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.1687735680025071,
      "inner": 11,
      "median_ms": 4.759453639218753,
      "q1_ms": 4.673188446428289,
      "q3_ms": 4.807710838080807
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.15245418390259147,
      "inner": 7,
      "median_ms": 7.21083371900022,
      "q1_ms": 7.0942925688411504,
      "q3_ms": 7.281897722610405
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.1745992200449109,
      "inner": 11,
      "median_ms": 4.665553908456455,
      "q1_ms": 4.618966824967752,
      "q3_ms": 4.886612444269386
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.15702038095332682,
      "inner": 7,
      "median_ms": 7.197686570829579,
      "q1_ms": 6.977492843621543,
      "q3_ms": 7.208159426227212
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.19753996492363513,
      "inner": 12,
      "median_ms": 4.3334892446485656,
      "q1_ms": 4.301241094556947,
      "q3_ms": 4.411780741065741
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.1654853771906346,
      "inner": 7,
      "median_ms": 6.192150858363935,
      "q1_ms": 6.188388015808804,
      "q3_ms": 6.215032289869019
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.21059532812796533,
      "inner": 13,
      "median_ms": 4.140005840991552,
      "q1_ms": 3.946730222266454,
      "q3_ms": 4.155171229145848
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.17218586802482605,
      "inner": 8,
      "median_ms": 5.844090861501172,
      "q1_ms": 5.793687130790204,
      "q3_ms": 5.887716979486868
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.22118979808874428,
      "inner": 13,
      "median_ms": 4.158094238776426,
      "q1_ms": 4.05302530942628,
      "q3_ms": 4.334386832152422
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.18072624411433935,
      "inner": 9,
      "median_ms": 5.591096227160758,
      "q1_ms": 5.538948656370242,
      "q3_ms": 5.609331652522087
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.15218137484043837,
      "inner": 21,
      "median_ms": 2.4877386167645454,
      "q1_ms": 2.424141624942422,
      "q3_ms": 2.523984183512983
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.13424693583510816,
      "inner": 12,
      "median_ms": 4.195267179359992,
      "q1_ms": 4.132963251322508,
      "q3_ms": 4.330570072246094
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.16270258417353034,
      "inner": 20,
      "median_ms": 2.5514301960356534,
      "q1_ms": 2.47102394932881,
      "q3_ms": 2.5570401456207037
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.14559442410245538,
      "inner": 12,
      "median_ms": 4.3284893229914205,
      "q1_ms": 4.2532411753199995,
      "q3_ms": 4.443865579863389
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.1668666829355061,
      "inner": 21,
      "median_ms": 2.367565619005334,
      "q1_ms": 2.3653475301606313,
      "q3_ms": 2.489252426173715
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.15086174895986915,
      "inner": 12,
      "median_ms": 4.2410703414740665,
      "q1_ms": 4.187138013852139,
      "q3_ms": 4.382749165718754
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.16951759508810937,
      "inner": 23,
      "median_ms": 2.227691427359115,
      "q1_ms": 2.1644542601121506,
      "q3_ms": 2.2504883413405525
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.15011354116722941,
      "inner": 14,
      "median_ms": 3.7377092854252885,
      "q1_ms": 3.605331776530615,
      "q3_ms": 3.767702795032944
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.18308553704991937,
      "inner": 25,
      "median_ms": 2.012567836791277,
      "q1_ms": 2.0109362434595823,
      "q3_ms": 2.136807879433036
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.1621148360427469,
      "inner": 15,
      "median_ms": 3.546044525379936,
      "q1_ms": 3.4480767945448556,
      "q3_ms": 3.591176401823759
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.19073322508484125,
      "inner": 25,
      "median_ms": 2.018582597374916,
      "q1_ms": 1.9941215217113497,
      "q3_ms": 2.078430037945509
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.16952674509957433,
      "inner": 15,
      "median_ms": 3.605322322497765,
      "q1_ms": 3.4962739950666823,
      "q3_ms": 3.646548263107737
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.22983699315227568,
      "inner": 23,
      "median_ms": 2.2776252185196983,
      "q1_ms": 2.2199506081802687,
      "q3_ms": 2.293311687105376
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.19612552877515554,
      "inner": 15,
      "median_ms": 3.4880180067072315,
      "q1_ms": 3.4712907392531633,
      "q3_ms": 3.5080565915753446
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.24671924393624067,
      "inner": 26,
      "median_ms": 1.9424763830521932,
      "q1_ms": 1.9345229240850759,
      "q3_ms": 2.050322964071081
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.21133447508327663,
      "inner": 15,
      "median_ms": 3.342941139514248,
      "q1_ms": 3.333515798052152,
      "q3_ms": 3.470308054238558
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.26014963584020734,
      "inner": 25,
      "median_ms": 1.9660331960767505,
      "q1_ms": 1.9459908362478018,
      "q3_ms": 2.097581196576357
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.21628583199344575,
      "inner": 16,
      "median_ms": 3.2877931225812063,
      "q1_ms": 3.2439644273836166,
      "q3_ms": 3.371592887560837
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.20629675989039242,
      "inner": 23,
      "median_ms": 2.2484983884446,
      "q1_ms": 2.1585373450880465,
      "q3_ms": 2.2611370431664195
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.14808867615647614,
      "inner": 20,
      "median_ms": 2.5048628565855324,
      "q1_ms": 2.4685839540325105,
      "q3_ms": 2.5399932404980063
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.22046162909828126,
      "inner": 22,
      "median_ms": 2.2748248672790146,
      "q1_ms": 2.2437258581207558,
      "q3_ms": 2.4236330487342044
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.15849795704707503,
      "inner": 21,
      "median_ms": 2.4283164336035648,
      "q1_ms": 2.356104663617554,
      "q3_ms": 2.442800617288975
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.225415145047009,
      "inner": 23,
      "median_ms": 2.215047123963418,
      "q1_ms": 2.1518486572186584,
      "q3_ms": 2.242989261108248
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.1637894520536065,
      "inner": 22,
      "median_ms": 2.3199656385589726,
      "q1_ms": 2.2513585546138613,
      "q3_ms": 2.3533223641358996
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.26540348096750677,
      "inner": 22,
      "median_ms": 2.251190408556299,
      "q1_ms": 2.2262922720983624,
      "q3_ms": 2.273132135583596
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.166859993012622,
      "inner": 23,
      "median_ms": 2.1708394785452145,
      "q1_ms": 2.158616001353316,
      "q3_ms": 2.218168917233529
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.2751592630520463,
      "inner": 24,
      "median_ms": 2.1643542083135494,
      "q1_ms": 2.074907417409122,
      "q3_ms": 2.1755383738006153
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.17832562001422048,
      "inner": 25,
      "median_ms": 2.0345161575824022,
      "q1_ms": 2.0151901990175247,
      "q3_ms": 2.1150571200996637
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.2811723160557449,
      "inner": 27,
      "median_ms": 1.968631779568063,
      "q1_ms": 1.8850868067669648,
      "q3_ms": 1.9798410314790629
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.18725858512334526,
      "inner": 25,
      "median_ms": 2.008220236748457,
      "q1_ms": 1.9845479261130092,
      "q3_ms": 2.094371635466814
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.4488619761541486,
      "inner": 25,
      "median_ms": 2.019506199285388,
      "q1_ms": 2.001926712691784,
      "q3_ms": 2.0317253656685352
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.2212522269692272,
      "inner": 18,
      "median_ms": 2.5808970547384686,
      "q1_ms": 2.5794971216883926,
      "q3_ms": 2.6078703295853405
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.42507532495073974,
      "inner": 26,
      "median_ms": 1.9871519985966957,
      "q1_ms": 1.897523421435975,
      "q3_ms": 1.9901712299682772
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.23685412295162678,
      "inner": 21,
      "median_ms": 2.349244718927713,
      "q1_ms": 2.3204495664685965,
      "q3_ms": 2.3580727588740134
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.48282097885385156,
      "inner": 24,
      "median_ms": 2.0883252824811884,
      "q1_ms": 2.0003347114349403,
      "q3_ms": 2.127990999724716
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0026154140941798687,
      "compile_s": 0.23856528685428202,
      "inner": 23,
      "median_ms": 2.108943952328485,
      "q1_ms": 2.0890757846443546,
      "q3_ms": 2.215568431774559
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.5485212421044707,
      "inner": 2,
      "median_ms": 21.307109505869448,
      "q1_ms": 21.24200901016593,
      "q3_ms": 21.3797785108909
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.3337045260705054,
      "inner": 1,
      "median_ms": 34.81689002364874,
      "q1_ms": 34.669669810682535,
      "q3_ms": 34.961740020662546
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.5500841569155455,
      "inner": 2,
      "median_ms": 19.18015500996262,
      "q1_ms": 19.163085496984422,
      "q3_ms": 19.189384998753667
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.33150396193377674,
      "inner": 1,
      "median_ms": 35.29034904204309,
      "q1_ms": 35.16671899706125,
      "q3_ms": 36.14540700800717
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.5369877750054002,
      "inner": 3,
      "median_ms": 17.212057719007134,
      "q1_ms": 17.10044437398513,
      "q3_ms": 17.45746028609574
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.3202932041604072,
      "inner": 1,
      "median_ms": 29.0162181481719,
      "q1_ms": 29.01571709662676,
      "q3_ms": 29.97798495925963
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.5442778039723635,
      "inner": 3,
      "median_ms": 15.492362280686697,
      "q1_ms": 15.037766968210539,
      "q3_ms": 15.583272014434138
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.3284596900921315,
      "inner": 1,
      "median_ms": 26.150924852117896,
      "q1_ms": 26.05976490303874,
      "q3_ms": 26.925473008304834
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.5600613080896437,
      "inner": 3,
      "median_ms": 14.861837650338808,
      "q1_ms": 14.146732942511639,
      "q3_ms": 14.998707066600522
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.30189787689596415,
      "inner": 2,
      "median_ms": 21.056359983049333,
      "q1_ms": 20.967974909581244,
      "q3_ms": 21.118244505487382
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.5393520379438996,
      "inner": 4,
      "median_ms": 12.052722973749042,
      "q1_ms": 11.75790122942999,
      "q3_ms": 12.358742242213339
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.3078909390605986,
      "inner": 2,
      "median_ms": 18.417597515508533,
      "q1_ms": 18.271267996169627,
      "q3_ms": 18.49552197381854
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.5339908429887146,
      "inner": 3,
      "median_ms": 15.372715968017777,
      "q1_ms": 15.267649665474892,
      "q3_ms": 15.413269322986404
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.34295662911608815,
      "inner": 2,
      "median_ms": 21.44269901327789,
      "q1_ms": 21.38056408148259,
      "q3_ms": 21.631538402289152
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.5588315718341619,
      "inner": 4,
      "median_ms": 11.722978961188346,
      "q1_ms": 11.625859304331243,
      "q3_ms": 12.034145533107221
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.3385131920222193,
      "inner": 2,
      "median_ms": 17.652739421464503,
      "q1_ms": 17.5530849955976,
      "q3_ms": 17.751244478859007
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.5405992339365184,
      "inner": 3,
      "median_ms": 14.234449326371154,
      "q1_ms": 14.189122710376978,
      "q3_ms": 14.348588961487016
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.33706347714178264,
      "inner": 2,
      "median_ms": 21.83488244190812,
      "q1_ms": 21.70115301851183,
      "q3_ms": 21.858632448129356
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.5587653431575745,
      "inner": 3,
      "median_ms": 13.380058342590928,
      "q1_ms": 13.268598976234594,
      "q3_ms": 13.405875070020556
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.3176502110436559,
      "inner": 2,
      "median_ms": 21.26935892738402,
      "q1_ms": 21.243134513497353,
      "q3_ms": 21.574778016656637
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.5483409920707345,
      "inner": 3,
      "median_ms": 12.939089676365256,
      "q1_ms": 12.819839951892694,
      "q3_ms": 13.106629330043992
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.2947924470063299,
      "inner": 3,
      "median_ms": 15.506922344987592,
      "q1_ms": 15.454156013826529,
      "q3_ms": 15.550475645189485
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.5404537038411945,
      "inner": 4,
      "median_ms": 11.383179982658476,
      "q1_ms": 11.375712521839887,
      "q3_ms": 11.394207540433854
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.3223405280150473,
      "inner": 3,
      "median_ms": 15.694691954801481,
      "q1_ms": 15.231033011029163,
      "q3_ms": 15.701528328160446
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.5276657207868993,
      "inner": 5,
      "median_ms": 9.94602357968688,
      "q1_ms": 9.484438830986619,
      "q3_ms": 9.960073418915272
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.3291980850044638,
      "inner": 2,
      "median_ms": 18.71979341376573,
      "q1_ms": 18.70092307217419,
      "q3_ms": 18.83397251367569
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.5364942280575633,
      "inner": 6,
      "median_ms": 8.486856512414912,
      "q1_ms": 8.321645320393145,
      "q3_ms": 8.712539177698394
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.3636011390481144,
      "inner": 3,
      "median_ms": 16.81332670462628,
      "q1_ms": 16.642513684928417,
      "q3_ms": 16.883839620277286
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.596442581852898,
      "inner": 5,
      "median_ms": 10.128749813884497,
      "q1_ms": 10.008194204419851,
      "q3_ms": 10.303827375173569
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.4319203889463097,
      "inner": 3,
      "median_ms": 13.681588616843024,
      "q1_ms": 13.667838647961617,
      "q3_ms": 13.892878002176682
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.6171584238763899,
      "inner": 6,
      "median_ms": 7.689220326331754,
      "q1_ms": 7.6586554835860925,
      "q3_ms": 7.829308315801123
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.46928790491074324,
      "inner": 4,
      "median_ms": 12.261740979738533,
      "q1_ms": 12.175748765002936,
      "q3_ms": 12.578775000292808
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.5475047279614955,
      "inner": 4,
      "median_ms": 12.437457975465804,
      "q1_ms": 12.37136300187558,
      "q3_ms": 12.743307277560234
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.3175635270308703,
      "inner": 2,
      "median_ms": 21.148231462575495,
      "q1_ms": 21.12103591207415,
      "q3_ms": 21.20781608391553
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.5412107149604708,
      "inner": 4,
      "median_ms": 11.319646029733121,
      "q1_ms": 11.173231527209282,
      "q3_ms": 11.324910970870405
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.3195769509766251,
      "inner": 3,
      "median_ms": 12.530688662081957,
      "q1_ms": 12.430642343436679,
      "q3_ms": 12.75167769442002
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.5300618461333215,
      "inner": 5,
      "median_ms": 9.77122480981052,
      "q1_ms": 9.755094768479466,
      "q3_ms": 9.817292587831616
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.3158663120120764,
      "inner": 3,
      "median_ms": 14.621956351523599,
      "q1_ms": 14.581032718221346,
      "q3_ms": 14.62361627879242
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.6452894760295749,
      "inner": 5,
      "median_ms": 10.557742603123188,
      "q1_ms": 10.058839991688728,
      "q3_ms": 10.693006217479706
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.37487105815671384,
      "inner": 3,
      "median_ms": 12.934930700187882,
      "q1_ms": 12.93164398521185,
      "q3_ms": 12.942050661270818
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.6583627900108695,
      "inner": 5,
      "median_ms": 9.540151618421078,
      "q1_ms": 9.363471996039152,
      "q3_ms": 9.562311414629221
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.4386759400367737,
      "inner": 4,
      "median_ms": 12.050691526383162,
      "q1_ms": 12.047536496538669,
      "q3_ms": 12.138391262851655
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.6837419790681452,
      "inner": 6,
      "median_ms": 9.207181166857481,
      "q1_ms": 8.784620518175265,
      "q3_ms": 9.236985972772041
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.5063081018161029,
      "inner": 4,
      "median_ms": 11.381668213289231,
      "q1_ms": 11.135441774968058,
      "q3_ms": 11.399468523450196
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.004734351299703121,
      "compile_s": 2.0108451389241964,
      "inner": 2,
      "median_ms": 18.344579031690955,
      "q1_ms": 18.338344059884548,
      "q3_ms": 18.370194011367857
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.5801585470326245,
      "inner": 5,
      "median_ms": 9.332487965002656,
      "q1_ms": 9.321330208331347,
      "q3_ms": 9.341564029455185
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.004734351299703121,
      "compile_s": 1.3072974951937795,
      "inner": 3,
      "median_ms": 16.42885773132245,
      "q1_ms": 16.204618305588763,
      "q3_ms": 16.992492678885657
    },
    {
      "dtype": "bfloat16",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.004734351299703121,
      "compile_s": 0.660159874940291,
      "inner": 6,
      "median_ms": 8.518914653298756,
      "q1_ms": 8.462206499340633,
      "q3_ms": 8.601649470316866
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "auto",
      "rel_err": 0.002930226270109415,
      "compile_s": 0.3211783170700073,
      "inner": 139,
      "median_ms": 0.23129245219417183,
      "q1_ms": 0.23061690664441467,
      "q3_ms": 0.2428076695650816
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "auto",
      "rel_err": 0.003966696560382843,
      "compile_s": 0.6287662319373339,
      "inner": 52,
      "median_ms": 0.9246255190541538,
      "q1_ms": 0.917345137657741,
      "q3_ms": 0.9473250560963956
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.16729778493754566,
      "inner": 65,
      "median_ms": 0.6617212453140663,
      "q1_ms": 0.657733862933058,
      "q3_ms": 0.6657096772239758
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.14738047984428704,
      "inner": 44,
      "median_ms": 1.0295546327886935,
      "q1_ms": 1.0224578648128293,
      "q3_ms": 1.0301144052804871
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.17355665704235435,
      "inner": 73,
      "median_ms": 0.6153514379099624,
      "q1_ms": 0.6152989744680794,
      "q3_ms": 0.6169362331753316
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.15148305799812078,
      "inner": 47,
      "median_ms": 0.9631932752722121,
      "q1_ms": 0.9529454475070568,
      "q3_ms": 0.9836347060317688
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.18321766005828977,
      "inner": 73,
      "median_ms": 0.6187219740440175,
      "q1_ms": 0.6098248900156723,
      "q3_ms": 0.6239375750189775
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.15754204196855426,
      "inner": 47,
      "median_ms": 0.9530786167275398,
      "q1_ms": 0.9239370231219428,
      "q3_ms": 0.9573260619443782
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.19672055286355317,
      "inner": 82,
      "median_ms": 0.5524176110445362,
      "q1_ms": 0.5444364150895215,
      "q3_ms": 0.5640169753261456
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.16143765090964735,
      "inner": 53,
      "median_ms": 0.7772166984823514,
      "q1_ms": 0.7766057728385588,
      "q3_ms": 0.7816965094293064
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.208765430143103,
      "inner": 85,
      "median_ms": 0.5096106926965362,
      "q1_ms": 0.5040050703374779,
      "q3_ms": 0.525091599453898
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.17133831302635372,
      "inner": 63,
      "median_ms": 0.6980110802465962,
      "q1_ms": 0.6962029987739192,
      "q3_ms": 0.702980255323743
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.21857598307542503,
      "inner": 86,
      "median_ms": 0.5166128597243927,
      "q1_ms": 0.5054804523540444,
      "q3_ms": 0.5240104063738916
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.176680559059605,
      "inner": 66,
      "median_ms": 0.6814246228190534,
      "q1_ms": 0.6769360007831093,
      "q3_ms": 0.6852503654293038
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.26122920401394367,
      "inner": 81,
      "median_ms": 0.5342831984622243,
      "q1_ms": 0.5276007520287861,
      "q3_ms": 0.53609097606422
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.18803282687440515,
      "inner": 74,
      "median_ms": 0.6108677574760608,
      "q1_ms": 0.5985265786482676,
      "q3_ms": 0.6140922038891429
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.2753285341896117,
      "inner": 100,
      "median_ms": 0.4448233707807958,
      "q1_ms": 0.44290626188740134,
      "q3_ms": 0.46225050929933786
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.20195440901443362,
      "inner": 82,
      "median_ms": 0.5367014403795688,
      "q1_ms": 0.5315455846579337,
      "q3_ms": 0.5441119031208317
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.2897932240739465,
      "inner": 86,
      "median_ms": 0.500163954119523,
      "q1_ms": 0.4924511872665134,
      "q3_ms": 0.5050449994890842
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.20959977712482214,
      "inner": 71,
      "median_ms": 0.5987800303584253,
      "q1_ms": 0.5967342688269179,
      "q3_ms": 0.6062774787801253
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.1940553500317037,
      "inner": 145,
      "median_ms": 0.27174952536307534,
      "q1_ms": 0.26175562003306274,
      "q3_ms": 0.27581330324555264
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.15372243314050138,
      "inner": 80,
      "median_ms": 0.5077997135231271,
      "q1_ms": 0.5075633525848389,
      "q3_ms": 0.5078482005046681
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.20700095500797033,
      "inner": 150,
      "median_ms": 0.26697499367098015,
      "q1_ms": 0.26196120617290336,
      "q3_ms": 0.2713816457738479
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.16567137907259166,
      "inner": 93,
      "median_ms": 0.4632160197743165,
      "q1_ms": 0.45368325726319386,
      "q3_ms": 0.46355753726456117
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.21514786197803915,
      "inner": 122,
      "median_ms": 0.31706140964429397,
      "q1_ms": 0.3084697222459268,
      "q3_ms": 0.31982329445051366
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.17346091801300645,
      "inner": 95,
      "median_ms": 0.451450537596094,
      "q1_ms": 0.44815822161341967,
      "q3_ms": 0.46112060939010824
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.23307834193110466,
      "inner": 141,
      "median_ms": 0.26513642552209665,
      "q1_ms": 0.2545270992852483,
      "q3_ms": 0.26563635419605053
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.1781607740558684,
      "inner": 106,
      "median_ms": 0.4149775288753071,
      "q1_ms": 0.39633097048005406,
      "q3_ms": 0.41540685759962725
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.2490128881763667,
      "inner": 140,
      "median_ms": 0.2613527716935745,
      "q1_ms": 0.2516066572362823,
      "q3_ms": 0.26700597954913974
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.1931112229358405,
      "inner": 107,
      "median_ms": 0.37475513015812806,
      "q1_ms": 0.35553798297089395,
      "q3_ms": 0.38102622335411124
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.2621370709966868,
      "inner": 152,
      "median_ms": 0.2615050654735808,
      "q1_ms": 0.2522297574231695,
      "q3_ms": 0.2624331513958934
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.20093020098283887,
      "inner": 112,
      "median_ms": 0.38996820512693375,
      "q1_ms": 0.3874188384673159,
      "q3_ms": 0.39263560860750396
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.326580542139709,
      "inner": 144,
      "median_ms": 0.2657132232949759,
      "q1_ms": 0.26405565343641985,
      "q3_ms": 0.27253146496756625
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.24215942714363337,
      "inner": 82,
      "median_ms": 0.500624221408876,
      "q1_ms": 0.5005248311180167,
      "q3_ms": 0.5010683657374324
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.35641973884776235,
      "inner": 127,
      "median_ms": 0.2895363635845541,
      "q1_ms": 0.28906258255771294,
      "q3_ms": 0.29741003962717655
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.25908224005252123,
      "inner": 105,
      "median_ms": 0.3924225323966571,
      "q1_ms": 0.3867462587853273,
      "q3_ms": 0.40472478029273806
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.37220688513480127,
      "inner": 139,
      "median_ms": 0.274198516707519,
      "q1_ms": 0.26501300610709105,
      "q3_ms": 0.27427010124267864
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.27055756794288754,
      "inner": 110,
      "median_ms": 0.3858515632931482,
      "q1_ms": 0.38247502789917315,
      "q3_ms": 0.39142954628914595
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.33062595105729997,
      "inner": 132,
      "median_ms": 0.2696206140557699,
      "q1_ms": 0.26900804545165913,
      "q3_ms": 0.272789008731982
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.19104503910057247,
      "inner": 132,
      "median_ms": 0.25291959966786887,
      "q1_ms": 0.2488375687266164,
      "q3_ms": 0.2578755692494187
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.34445816301740706,
      "inner": 136,
      "median_ms": 0.2589472946163048,
      "q1_ms": 0.23526669931871927,
      "q3_ms": 0.27084137614378156
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.2036899239756167,
      "inner": 145,
      "median_ms": 0.25960755181209794,
      "q1_ms": 0.257269564437969,
      "q3_ms": 0.2678245659275302
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.3523930811788887,
      "inner": 160,
      "median_ms": 0.238277213065885,
      "q1_ms": 0.2299697996932082,
      "q3_ms": 0.23841977526899427
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.20931662805378437,
      "inner": 153,
      "median_ms": 0.25437248937688234,
      "q1_ms": 0.2470570266417234,
      "q3_ms": 0.2559217718715956
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.4698533830232918,
      "inner": 156,
      "median_ms": 0.23050224468207514,
      "q1_ms": 0.2214556667380608,
      "q3_ms": 0.23134378268598363
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.22571154311299324,
      "inner": 100,
      "median_ms": 0.37807474844157696,
      "q1_ms": 0.37638584850355983,
      "q3_ms": 0.3787851519882679
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.4424745000433177,
      "inner": 167,
      "median_ms": 0.21143198782901565,
      "q1_ms": 0.19788346671594118,
      "q3_ms": 0.22212692263151357
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.24157335795462132,
      "inner": 122,
      "median_ms": 0.2853209343662516,
      "q1_ms": 0.2809379013163633,
      "q3_ms": 0.285594867252302
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.456188591895625,
      "inner": 161,
      "median_ms": 0.23047283858708711,
      "q1_ms": 0.223331552531038,
      "q3_ms": 0.24077634801451836
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.24781610118225217,
      "inner": 154,
      "median_ms": 0.2404180969196287,
      "q1_ms": 0.23444727299033435,
      "q3_ms": 0.24411061706093998
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.9385182908736169,
      "inner": 88,
      "median_ms": 0.5062085904435001,
      "q1_ms": 0.488339669325135,
      "q3_ms": 0.516186624405567
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.3055135200265795,
      "inner": 133,
      "median_ms": 0.2731489388734326,
      "q1_ms": 0.27279999226163654,
      "q3_ms": 0.2823955260385248
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.8763104840181768,
      "inner": 86,
      "median_ms": 0.4943978138874437,
      "q1_ms": 0.4930838585159806,
      "q3_ms": 0.500085465826614
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.32422121800482273,
      "inner": 142,
      "median_ms": 0.23523110627445956,
      "q1_ms": 0.2334209021285806,
      "q3_ms": 0.24684269840396203
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.9513403850141913,
      "inner": 83,
      "median_ms": 0.5212302392640388,
      "q1_ms": 0.5205674573823988,
      "q3_ms": 0.5247305927192232
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.002930226270109415,
      "compile_s": 0.33464600006118417,
      "inner": 147,
      "median_ms": 0.22830120382868513,
      "q1_ms": 0.22449019067242842,
      "q3_ms": 0.23887783675422877
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.003966696560382843,
      "compile_s": 1.02063846308738,
      "inner": 16,
      "median_ms": 3.054166561923921,
      "q1_ms": 2.988821070175618,
      "q3_ms": 3.113625687547028
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.003966696560382843,
      "compile_s": 0.3941211241763085,
      "inner": 11,
      "median_ms": 4.127431279894981,
      "q1_ms": 4.113801268183373,
      "q3_ms": 4.133125809444623
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.003966696560382843,
      "compile_s": 1.013316682074219,
      "inner": 18,
      "median_ms": 2.612775502105554,
      "q1_ms": 2.611169449260665,
      "q3_ms": 2.6458787292035093
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.003966696560382843,
      "compile_s": 0.4290182269178331,
      "inner": 13,
      "median_ms": 3.768180299980136,
      "q1_ms": 3.745173390668172,
      "q3_ms": 3.794616392741983
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.003966696560382843,
      "compile_s": 1.012884653871879,
      "inner": 19,
      "median_ms": 2.539968154834289,
      "q1_ms": 2.5337671640476116,
      "q3_ms": 2.5446239466729916
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.003966696560382843,
      "compile_s": 0.37743923999369144,
      "inner": 17,
      "median_ms": 2.9007801649105898,
      "q1_ms": 2.8970278251697033,
      "q3_ms": 2.919269520241548
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.003966696560382843,
      "compile_s": 0.9832670360337943,
      "inner": 23,
      "median_ms": 2.029651300171795,
      "q1_ms": 2.0264135167488586,
      "q3_ms": 2.035740044210916
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.003966696560382843,
      "compile_s": 0.38887027907185256,
      "inner": 18,
      "median_ms": 2.7244163340785437,
      "q1_ms": 2.7095925114634962,
      "q3_ms": 2.7312407295943966
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.003966696560382843,
      "compile_s": 1.0177653888240457,
      "inner": 25,
      "median_ms": 2.0455791987478733,
      "q1_ms": 2.040302436798811,
      "q3_ms": 2.0693987514823675
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.003966696560382843,
      "compile_s": 0.3802781901322305,
      "inner": 16,
      "median_ms": 2.912204508902505,
      "q1_ms": 2.905846427893266,
      "q3_ms": 2.925413878983818
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.003966696560382843,
      "compile_s": 0.9542171210050583,
      "inner": 30,
      "median_ms": 1.653749798424542,
      "q1_ms": 1.6397482017055154,
      "q3_ms": 1.660897466354072
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.003966696560382843,
      "compile_s": 0.3923534268978983,
      "inner": 20,
      "median_ms": 2.368760493118316,
      "q1_ms": 2.3396286414936185,
      "q3_ms": 2.3926509427838027
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.003966696560382843,
      "compile_s": 1.0257817150559276,
      "inner": 25,
      "median_ms": 2.008130121976137,
      "q1_ms": 1.9918745569884777,
      "q3_ms": 2.0669119618833065
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.003966696560382843,
      "compile_s": 0.39501755009405315,
      "inner": 21,
      "median_ms": 2.2195006188537394,
      "q1_ms": 2.2194349051763615,
      "q3_ms": 2.2234219969028515
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.003966696560382843,
      "compile_s": 0.9749482041224837,
      "inner": 30,
      "median_ms": 1.6073272641127307,
      "q1_ms": 1.573875336907804,
      "q3_ms": 1.6117962620531519
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.003966696560382843,
      "compile_s": 0.4014936320018023,
      "inner": 28,
      "median_ms": 1.7680223183041173,
      "q1_ms": 1.7576023237779737,
      "q3_ms": 1.775483351333865
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.003966696560382843,
      "compile_s": 1.0167924400884658,
      "inner": 25,
      "median_ms": 1.8603805173188448,
      "q1_ms": 1.842720154672861,
      "q3_ms": 1.8894568365067244
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.003966696560382843,
      "compile_s": 0.39302184502594173,
      "inner": 14,
      "median_ms": 3.3004360073911294,
      "q1_ms": 3.287722439771252,
      "q3_ms": 3.3153673534148504
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.003966696560382843,
      "compile_s": 0.9772915369831026,
      "inner": 27,
      "median_ms": 1.6990638924417671,
      "q1_ms": 1.6864987038489845,
      "q3_ms": 1.7231445569820978
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.003966696560382843,
      "compile_s": 0.4169720890931785,
      "inner": 15,
      "median_ms": 3.058488992974162,
      "q1_ms": 3.057804921021064,
      "q3_ms": 3.0702869407832623
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.003966696560382843,
      "compile_s": 1.0232018218375742,
      "inner": 32,
      "median_ms": 1.5285161571227945,
      "q1_ms": 1.5220767818391323,
      "q3_ms": 1.5331961185438558
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.003966696560382843,
      "compile_s": 0.36949310917407274,
      "inner": 21,
      "median_ms": 2.195213998978337,
      "q1_ms": 2.192086861690595,
      "q3_ms": 2.1959887595758554
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.003966696560382843,
      "compile_s": 0.9545237498823553,
      "inner": 34,
      "median_ms": 1.4083276174085981,
      "q1_ms": 1.4044417355976562,
      "q3_ms": 1.4097184997380656
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.003966696560382843,
      "compile_s": 0.39338798518292606,
      "inner": 23,
      "median_ms": 2.03864399910621,
      "q1_ms": 2.026577085337561,
      "q3_ms": 2.0594930519228396
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.003966696560382843,
      "compile_s": 1.0165257898624986,
      "inner": 35,
      "median_ms": 1.3640268572739194,
      "q1_ms": 1.363322857235159,
      "q3_ms": 1.3731387976024834
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.003966696560382843,
      "compile_s": 0.4062198188621551,
      "inner": 25,
      "median_ms": 1.8562689144164324,
      "q1_ms": 1.8478841613978148,
      "q3_ms": 1.8751904740929604
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.003966696560382843,
      "compile_s": 0.9533186231274158,
      "inner": 41,
      "median_ms": 1.1677738824268666,
      "q1_ms": 1.157989271166848,
      "q3_ms": 1.169271437769256
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.003966696560382843,
      "compile_s": 0.4758691689930856,
      "inner": 30,
      "median_ms": 1.5988559617350497,
      "q1_ms": 1.5973779605701566,
      "q3_ms": 1.6025069635361433
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.003966696560382843,
      "compile_s": 1.177420609863475,
      "inner": 30,
      "median_ms": 1.6245975391939282,
      "q1_ms": 1.6073345982780058,
      "q3_ms": 1.6606694630657632
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.003966696560382843,
      "compile_s": 0.5433904139790684,
      "inner": 33,
      "median_ms": 1.4663184216866891,
      "q1_ms": 1.4450363320947597,
      "q3_ms": 1.473509001009392
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.003966696560382843,
      "compile_s": 0.9568917031865567,
      "inner": 26,
      "median_ms": 1.8022619993784106,
      "q1_ms": 1.7772724207204122,
      "q3_ms": 1.8057223910895677
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.003966696560382843,
      "compile_s": 0.5934216568712145,
      "inner": 39,
      "median_ms": 1.2589424597815824,
      "q1_ms": 1.2213979453708117,
      "q3_ms": 1.2761980223541076
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.003966696560382843,
      "compile_s": 1.0388016290962696,
      "inner": 20,
      "median_ms": 2.5609819451346993,
      "q1_ms": 2.5002512033097446,
      "q3_ms": 2.6223722961731255
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.003966696560382843,
      "compile_s": 0.37999005103483796,
      "inner": 25,
      "median_ms": 1.8094138335436583,
      "q1_ms": 1.8003954738378525,
      "q3_ms": 1.8097510375082493
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.003966696560382843,
      "compile_s": 0.9685384309850633,
      "inner": 20,
      "median_ms": 2.5136056472547352,
      "q1_ms": 2.482919197063893,
      "q3_ms": 2.527054597157985
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.003966696560382843,
      "compile_s": 0.4145069560036063,
      "inner": 28,
      "median_ms": 1.663891856359052,
      "q1_ms": 1.658432252172913,
      "q3_ms": 1.6759697131679527
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.003966696560382843,
      "compile_s": 1.0253185669425875,
      "inner": 26,
      "median_ms": 1.9409573869779706,
      "q1_ms": 1.8964501885840526,
      "q3_ms": 1.9443766124403248
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.003966696560382843,
      "compile_s": 0.45329472003504634,
      "inner": 34,
      "median_ms": 1.358445966616273,
      "q1_ms": 1.3480319175869226,
      "q3_ms": 1.3692618539447292
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.003966696560382843,
      "compile_s": 1.1998607190325856,
      "inner": 18,
      "median_ms": 2.740196392147077,
      "q1_ms": 2.6829682238813906,
      "q3_ms": 2.7547369374790125
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.003966696560382843,
      "compile_s": 0.5152927509043366,
      "inner": 36,
      "median_ms": 1.2549459998909798,
      "q1_ms": 1.2480163067165349,
      "q3_ms": 1.2573407212686207
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.003966696560382843,
      "compile_s": 1.5758622598368675,
      "inner": 14,
      "median_ms": 3.3957278688571284,
      "q1_ms": 3.3775664965755174,
      "q3_ms": 3.416576356227909
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.003966696560382843,
      "compile_s": 0.6677410139236599,
      "inner": 31,
      "median_ms": 1.6075930362867734,
      "q1_ms": 1.6022149638663377,
      "q3_ms": 1.615576228247054
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.003966696560382843,
      "compile_s": 1.1947186219040304,
      "inner": 20,
      "median_ms": 2.4923176504671574,
      "q1_ms": 2.4852911941707134,
      "q3_ms": 2.4977341410703957
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.003966696560382843,
      "compile_s": 0.7348312409594655,
      "inner": 35,
      "median_ms": 1.3647851135049547,
      "q1_ms": 1.3587725681385825,
      "q3_ms": 1.3654979677604777
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.003966696560382843,
      "compile_s": 6.179280700162053,
      "inner": 12,
      "median_ms": 3.9827657359031337,
      "q1_ms": 3.971946658566594,
      "q3_ms": 3.9889132604002953
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.003966696560382843,
      "compile_s": 0.894182374002412,
      "inner": 40,
      "median_ms": 1.1631440720520914,
      "q1_ms": 1.162418071180582,
      "q3_ms": 1.1706820456311107
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.003966696560382843,
      "compile_s": 2.379718122072518,
      "inner": 14,
      "median_ms": 3.3177330790619766,
      "q1_ms": 3.3155137872589484,
      "q3_ms": 3.340020858948784
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.003966696560382843,
      "compile_s": 0.8951716318260878,
      "inner": 48,
      "median_ms": 0.99350333524247,
      "q1_ms": 0.9934545620732631,
      "q3_ms": 1.000658521661535
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "auto",
      "rel_err": 0.0017128211911767721,
      "compile_s": 0.32501724106259644,
      "inner": 15,
      "median_ms": 3.4551738761365414,
      "q1_ms": 3.2749397369722524,
      "q3_ms": 3.516905739282568
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "auto",
      "rel_err": 0.00294348387978971,
      "compile_s": 0.648009727941826,
      "inner": 3,
      "median_ms": 14.086401322856545,
      "q1_ms": 13.954338307181994,
      "q3_ms": 14.098351666082939
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.17575114988721907,
      "inner": 5,
      "median_ms": 10.015364596620202,
      "q1_ms": 10.007394570857286,
      "q3_ms": 10.10197438299656
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.1589130749925971,
      "inner": 3,
      "median_ms": 15.616914024576545,
      "q1_ms": 15.500544027114907,
      "q3_ms": 15.754213634257514
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.1816959730349481,
      "inner": 5,
      "median_ms": 9.612591611221433,
      "q1_ms": 9.237078810110688,
      "q3_ms": 9.643245628103614
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.16500257910229266,
      "inner": 3,
      "median_ms": 14.979605718205372,
      "q1_ms": 14.696506705755988,
      "q3_ms": 15.004195660973588
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.1901293401606381,
      "inner": 5,
      "median_ms": 9.489578008651733,
      "q1_ms": 9.119006991386414,
      "q3_ms": 9.695271635428071
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.1706609830725938,
      "inner": 3,
      "median_ms": 14.748176326975226,
      "q1_ms": 14.664796646684408,
      "q3_ms": 14.948535632962981
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.2071394429076463,
      "inner": 6,
      "median_ms": 8.598458177099625,
      "q1_ms": 8.325922341706852,
      "q3_ms": 8.633726315262416
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.1714994008652866,
      "inner": 4,
      "median_ms": 11.717537767253816,
      "q1_ms": 11.717348010279238,
      "q3_ms": 11.726505472324789
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.21569628012366593,
      "inner": 7,
      "median_ms": 7.633326275806342,
      "q1_ms": 7.562622128586684,
      "q3_ms": 7.9133412800729275
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.17916330113075674,
      "inner": 4,
      "median_ms": 10.884430259466171,
      "q1_ms": 10.637308238074183,
      "q3_ms": 11.111449741292745
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.22620925190858543,
      "inner": 6,
      "median_ms": 7.781676986875634,
      "q1_ms": 7.531839306466281,
      "q3_ms": 8.185712659421066
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.18500934285111725,
      "inner": 4,
      "median_ms": 10.652500961441547,
      "q1_ms": 10.356801736634225,
      "q3_ms": 10.761615762021393
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0017128211911767721,
      "compile_s": 0.26652315095998347,
      "inner": 6,
      "median_ms": 7.969369840187332,
      "q1_ms": 7.9564516587803755,
      "q3_ms": 8.147810973847905
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0017128211911767721,
      "compile_s": 0.19602277386002243,
      "inner": 5,
      "median_ms": 9.393344214186072,
      "q1_ms": 9.314068406820297,
      "q3_ms": 9.409008221700788
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0017128211911767721,
      "compile_s": 0.2806655231397599,
      "inner": 7,
      "median_ms": 6.8496541519250185,
      "q1_ms": 6.7495915573090315,
      "q3_ms": 7.128500412883503
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0017128211911767721,
      "compile_s": 0.2078289019409567,
      "inner": 6,
      "median_ms": 8.495723518232504,
      "q1_ms": 8.211102491865555,
      "q3_ms": 8.526313351467252
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0017128211911767721,
      "compile_s": 0.29518032306805253,
      "inner": 6,
      "median_ms": 7.964443298988044,
      "q1_ms": 7.897496650305888,
      "q3_ms": 8.05929466150701
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0017128211911767721,
      "compile_s": 0.22002951800823212,
      "inner": 5,
      "median_ms": 9.238514769822359,
      "q1_ms": 9.121085004881024,
      "q3_ms": 9.274450596421957
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.19695175089873374,
      "inner": 13,
      "median_ms": 4.190923162521078,
      "q1_ms": 3.954012305117571,
      "q3_ms": 4.215579229192092
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.15885205590166152,
      "inner": 7,
      "median_ms": 7.243653027606862,
      "q1_ms": 7.229598704725504,
      "q3_ms": 7.2554172948002815
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.21050291415303946,
      "inner": 13,
      "median_ms": 3.942923477062812,
      "q1_ms": 3.7927814544393468,
      "q3_ms": 4.08701145520004
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.1707743878941983,
      "inner": 7,
      "median_ms": 6.833557571683611,
      "q1_ms": 6.815053422802261,
      "q3_ms": 7.283633541581886
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.21923912013880908,
      "inner": 12,
      "median_ms": 4.479394837593039,
      "q1_ms": 4.360849251194547,
      "q3_ms": 4.538427165243775
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.17689433204941452,
      "inner": 7,
      "median_ms": 6.971832996766482,
      "q1_ms": 6.902508870033281,
      "q3_ms": 7.314896410597222
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.23359218193218112,
      "inner": 14,
      "median_ms": 3.6850616535437957,
      "q1_ms": 3.6746160060699498,
      "q3_ms": 3.8136427756398916
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.18397369282320142,
      "inner": 8,
      "median_ms": 6.15287249092944,
      "q1_ms": 6.006792857078835,
      "q3_ms": 6.219008500920609
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.25178007408976555,
      "inner": 13,
      "median_ms": 3.9637818431051874,
      "q1_ms": 3.901218923811729,
      "q3_ms": 4.0330293122678995
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.19691899907775223,
      "inner": 9,
      "median_ms": 5.854461238616043,
      "q1_ms": 5.4680777765396575,
      "q3_ms": 5.864971110390292
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.267077743075788,
      "inner": 14,
      "median_ms": 3.7259379168972373,
      "q1_ms": 3.715061565994152,
      "q3_ms": 3.881153999827802
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.20443356991745532,
      "inner": 9,
      "median_ms": 5.892802236808671,
      "q1_ms": 5.650121666904952,
      "q3_ms": 5.950936452589101
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0017128211911767721,
      "compile_s": 0.32884021988138556,
      "inner": 13,
      "median_ms": 3.9135757833719254,
      "q1_ms": 3.7411655514286113,
      "q3_ms": 3.9276165457872247
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0017128211911767721,
      "compile_s": 0.2469424670562148,
      "inner": 7,
      "median_ms": 6.602045414703233,
      "q1_ms": 6.509491308991398,
      "q3_ms": 6.613836856558919
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0017128211911767721,
      "compile_s": 0.3530685850419104,
      "inner": 12,
      "median_ms": 4.2380854914275305,
      "q1_ms": 4.171515601531913,
      "q3_ms": 4.375673423055559
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0017128211911767721,
      "compile_s": 0.2630893439054489,
      "inner": 9,
      "median_ms": 5.735667008492682,
      "q1_ms": 5.675716118680106,
      "q3_ms": 6.040378421958949
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0017128211911767721,
      "compile_s": 0.36994375102221966,
      "inner": 13,
      "median_ms": 4.113599138620954,
      "q1_ms": 3.8970773860525627,
      "q3_ms": 4.1211337710802365
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0017128211911767721,
      "compile_s": 0.27560109994374216,
      "inner": 9,
      "median_ms": 5.771921456067099,
      "q1_ms": 5.621868444399701,
      "q3_ms": 5.946103224737777
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.33241351088508964,
      "inner": 13,
      "median_ms": 4.064789220977287,
      "q1_ms": 3.9121458450189,
      "q3_ms": 4.068876150995493
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.19325491902418435,
      "inner": 14,
      "median_ms": 3.552226998311068,
      "q1_ms": 3.540625562891364,
      "q3_ms": 3.674426714756659
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.34676501294597983,
      "inner": 13,
      "median_ms": 3.911298842957387,
      "q1_ms": 3.7209455353709364,
      "q3_ms": 3.967014929422965
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.208966176956892,
      "inner": 13,
      "median_ms": 3.9972886801339107,
      "q1_ms": 3.8549128489998674,
      "q3_ms": 4.004631692973467
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.3542781420983374,
      "inner": 15,
      "median_ms": 3.4565448605765896,
      "q1_ms": 3.264600007484357,
      "q3_ms": 3.4689394601931176
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.2112107709981501,
      "inner": 14,
      "median_ms": 3.745411489424961,
      "q1_ms": 3.629454710919942,
      "q3_ms": 3.824698439400111
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.4703612560406327,
      "inner": 16,
      "median_ms": 3.285163198597729,
      "q1_ms": 3.2641057478031144,
      "q3_ms": 3.373854182427749
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.2289252351038158,
      "inner": 10,
      "median_ms": 4.649985698051751,
      "q1_ms": 4.643273679539561,
      "q3_ms": 4.653660696931183
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.4428282578010112,
      "inner": 16,
      "median_ms": 3.245665750000626,
      "q1_ms": 3.107725497102365,
      "q3_ms": 3.2603563740849495
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.24276598799042404,
      "inner": 12,
      "median_ms": 3.931458748411387,
      "q1_ms": 3.8864189215625324,
      "q3_ms": 3.963642831270893
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.4550819359719753,
      "inner": 15,
      "median_ms": 3.5231180023401976,
      "q1_ms": 3.51279399668177,
      "q3_ms": 3.551290665442745
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.001778920879587531,
      "compile_s": 0.2491916110739112,
      "inner": 14,
      "median_ms": 3.4997214263837253,
      "q1_ms": 3.4564972821889177,
      "q3_ms": 3.6785345624334047
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0017128211911767721,
      "compile_s": 0.9479989921674132,
      "inner": 6,
      "median_ms": 7.267445636292298,
      "q1_ms": 7.200994179584086,
      "q3_ms": 7.4687168234959245
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0017128211911767721,
      "compile_s": 0.3057508000638336,
      "inner": 14,
      "median_ms": 3.7665342845554863,
      "q1_ms": 3.6245875027296797,
      "q3_ms": 3.8190434320962856
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0017128211911767721,
      "compile_s": 0.8813973881769925,
      "inner": 6,
      "median_ms": 7.469693703266482,
      "q1_ms": 7.4550185042123,
      "q3_ms": 7.526088467178245
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0017128211911767721,
      "compile_s": 0.32504164008423686,
      "inner": 15,
      "median_ms": 3.5100420005619526,
      "q1_ms": 3.261921399583419,
      "q3_ms": 3.5230267327278852
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0017128211911767721,
      "compile_s": 0.9561915500089526,
      "inner": 6,
      "median_ms": 7.867267665763697,
      "q1_ms": 7.822772682023546,
      "q3_ms": 7.925185685356459
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0017128211911767721,
      "compile_s": 0.33504518400877714,
      "inner": 16,
      "median_ms": 3.3255130547331646,
      "q1_ms": 3.160566629958339,
      "q3_ms": 3.3538755087647587
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.00294348387978971,
      "compile_s": 1.0630978769622743,
      "inner": 1,
      "median_ms": 44.792642118409276,
      "q1_ms": 44.61735184304416,
      "q3_ms": 45.07397091947496
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.00294348387978971,
      "compile_s": 0.4500386801082641,
      "inner": 1,
      "median_ms": 64.1170300077647,
      "q1_ms": 64.08209097571671,
      "q3_ms": 64.20456012710929
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.00294348387978971,
      "compile_s": 0.9873114980291575,
      "inner": 1,
      "median_ms": 41.58906009979546,
      "q1_ms": 41.00035200826824,
      "q3_ms": 41.84370907023549
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.00294348387978971,
      "compile_s": 0.45104677602648735,
      "inner": 1,
      "median_ms": 60.23531104438007,
      "q1_ms": 59.811791870743036,
      "q3_ms": 60.6944290921092
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.00294348387978971,
      "compile_s": 1.0536491321399808,
      "inner": 1,
      "median_ms": 38.51452795788646,
      "q1_ms": 38.475617999210954,
      "q3_ms": 39.09215587191284
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.00294348387978971,
      "compile_s": 0.4105140140745789,
      "inner": 1,
      "median_ms": 45.65600003115833,
      "q1_ms": 45.639388961717486,
      "q3_ms": 45.68497906439006
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.00294348387978971,
      "compile_s": 0.9884880939498544,
      "inner": 1,
      "median_ms": 34.55587779171765,
      "q1_ms": 34.13754003122449,
      "q3_ms": 34.85310799442232
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.00294348387978971,
      "compile_s": 0.4279662878252566,
      "inner": 1,
      "median_ms": 42.850536992773414,
      "q1_ms": 42.19629801809788,
      "q3_ms": 43.02385589107871
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.00294348387978971,
      "compile_s": 1.0458031331654638,
      "inner": 1,
      "median_ms": 30.968938022851944,
      "q1_ms": 30.79328895546496,
      "q3_ms": 31.288056867197156
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.00294348387978971,
      "compile_s": 0.4311922488268465,
      "inner": 1,
      "median_ms": 44.59403082728386,
      "q1_ms": 44.16063311509788,
      "q3_ms": 45.20380008034408
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.00294348387978971,
      "compile_s": 0.9804415558464825,
      "inner": 2,
      "median_ms": 26.200120570138097,
      "q1_ms": 26.085170917212963,
      "q3_ms": 27.78667153324932
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.00294348387978971,
      "compile_s": 0.44382720603607595,
      "inner": 1,
      "median_ms": 36.18599404580891,
      "q1_ms": 36.02050384506583,
      "q3_ms": 36.41540394164622
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.00294348387978971,
      "compile_s": 1.0421091029420495,
      "inner": 1,
      "median_ms": 30.66904889419675,
      "q1_ms": 30.383770121261477,
      "q3_ms": 31.449965899810195
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.00294348387978971,
      "compile_s": 0.4447702928446233,
      "inner": 1,
      "median_ms": 33.613661071285605,
      "q1_ms": 33.39515207335353,
      "q3_ms": 33.812239998951554
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.00294348387978971,
      "compile_s": 1.011891283094883,
      "inner": 1,
      "median_ms": 35.19070683978498,
      "q1_ms": 34.76830804720521,
      "q3_ms": 35.240706987679005
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.00294348387978971,
      "compile_s": 0.43644763506017625,
      "inner": 1,
      "median_ms": 27.145677944645286,
      "q1_ms": 27.099509024992585,
      "q3_ms": 27.148247929289937
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.00294348387978971,
      "compile_s": 1.0402249379549176,
      "inner": 1,
      "median_ms": 28.317224932834506,
      "q1_ms": 28.25513598509133,
      "q3_ms": 28.553444892168045
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.00294348387978971,
      "compile_s": 0.44476613285951316,
      "inner": 1,
      "median_ms": 50.87657505646348,
      "q1_ms": 50.48811691813171,
      "q3_ms": 50.97643518820405
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.00294348387978971,
      "compile_s": 1.0055852299556136,
      "inner": 2,
      "median_ms": 25.97536100074649,
      "q1_ms": 25.93221643473953,
      "q3_ms": 26.02836152072996
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.00294348387978971,
      "compile_s": 0.43603929597884417,
      "inner": 1,
      "median_ms": 47.58412390947342,
      "q1_ms": 47.56362410262227,
      "q3_ms": 47.629063948988914
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.00294348387978971,
      "compile_s": 1.0378828539978713,
      "inner": 2,
      "median_ms": 23.395878029987216,
      "q1_ms": 23.161873454228044,
      "q3_ms": 23.847762029618025
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.00294348387978971,
      "compile_s": 0.42808202700689435,
      "inner": 1,
      "median_ms": 33.373111858963966,
      "q1_ms": 33.361752051860094,
      "q3_ms": 33.944350900128484
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.00294348387978971,
      "compile_s": 0.977247575065121,
      "inner": 2,
      "median_ms": 21.423323545604944,
      "q1_ms": 21.155503927730024,
      "q3_ms": 21.43009298015386
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.00294348387978971,
      "compile_s": 0.45626960205845535,
      "inner": 1,
      "median_ms": 31.719806138426065,
      "q1_ms": 31.57924604602158,
      "q3_ms": 32.336795004084706
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.00294348387978971,
      "compile_s": 1.0440538879483938,
      "inner": 2,
      "median_ms": 20.15030151233077,
      "q1_ms": 20.064087002538145,
      "q3_ms": 20.16763191204518
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.00294348387978971,
      "compile_s": 0.4293720140121877,
      "inner": 1,
      "median_ms": 28.095915913581848,
      "q1_ms": 28.026466025039554,
      "q3_ms": 28.75657402910292
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.00294348387978971,
      "compile_s": 0.9738528931047767,
      "inner": 2,
      "median_ms": 18.033082596957684,
      "q1_ms": 17.590847914107144,
      "q3_ms": 19.213628955185413
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.00294348387978971,
      "compile_s": 0.5154572259634733,
      "inner": 2,
      "median_ms": 24.77246941998601,
      "q1_ms": 24.24980595242232,
      "q3_ms": 25.393537478521466
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.00294348387978971,
      "compile_s": 1.2116227250080556,
      "inner": 1,
      "median_ms": 32.387354178354144,
      "q1_ms": 32.29455486871302,
      "q3_ms": 32.63647295534611
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.00294348387978971,
      "compile_s": 0.5642673971597105,
      "inner": 2,
      "median_ms": 21.701022516936064,
      "q1_ms": 21.628223010338843,
      "q3_ms": 21.74250246025622
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.00294348387978971,
      "compile_s": 1.004356101853773,
      "inner": 1,
      "median_ms": 26.952248997986317,
      "q1_ms": 26.782939210534096,
      "q3_ms": 27.12280792184174
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.00294348387978971,
      "compile_s": 0.6138224161695689,
      "inner": 2,
      "median_ms": 19.087679451331496,
      "q1_ms": 18.384416471235454,
      "q3_ms": 19.146304461173713
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.00294348387978971,
      "compile_s": 1.0573831021320075,
      "inner": 1,
      "median_ms": 40.730491979047656,
      "q1_ms": 40.296563878655434,
      "q3_ms": 41.15079203620553
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.00294348387978971,
      "compile_s": 0.417203996097669,
      "inner": 1,
      "median_ms": 27.093967888504267,
      "q1_ms": 27.067708782851696,
      "q3_ms": 27.242728043347597
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.00294348387978971,
      "compile_s": 0.9912206679582596,
      "inner": 1,
      "median_ms": 39.37566583044827,
      "q1_ms": 38.63749699667096,
      "q3_ms": 40.09375302121043
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.00294348387978971,
      "compile_s": 0.44609599909745157,
      "inner": 1,
      "median_ms": 25.051264092326164,
      "q1_ms": 25.029304204508662,
      "q3_ms": 25.10069403797388
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.00294348387978971,
      "compile_s": 1.0555139069911093,
      "inner": 1,
      "median_ms": 30.623069033026695,
      "q1_ms": 30.495668994262815,
      "q3_ms": 32.28252497501671
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.00294348387978971,
      "compile_s": 0.4451676309108734,
      "inner": 2,
      "median_ms": 20.49170550890267,
      "q1_ms": 20.28367144521326,
      "q3_ms": 20.49613557755947
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.00294348387978971,
      "compile_s": 1.2355649240780622,
      "inner": 1,
      "median_ms": 41.76434106193483,
      "q1_ms": 41.68460005894303,
      "q3_ms": 41.821049991995096
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.00294348387978971,
      "compile_s": 0.5516064912080765,
      "inner": 2,
      "median_ms": 18.71439057867974,
      "q1_ms": 18.62740598153323,
      "q3_ms": 18.762546009384096
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.00294348387978971,
      "compile_s": 1.634667525999248,
      "inner": 1,
      "median_ms": 53.60554903745651,
      "q1_ms": 53.58939990401268,
      "q3_ms": 53.80739807151258
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.00294348387978971,
      "compile_s": 0.6863278069067746,
      "inner": 2,
      "median_ms": 23.80115247797221,
      "q1_ms": 23.31758849322796,
      "q3_ms": 24.2618415504694
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.00294348387978971,
      "compile_s": 1.2616351549513638,
      "inner": 1,
      "median_ms": 39.90582493133843,
      "q1_ms": 39.83107511885464,
      "q3_ms": 39.93010497651994
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.00294348387978971,
      "compile_s": 0.7367995949462056,
      "inner": 2,
      "median_ms": 20.32242203131318,
      "q1_ms": 20.210096961818635,
      "q3_ms": 20.367261487990618
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.00294348387978971,
      "compile_s": 6.1944347359240055,
      "inner": 1,
      "median_ms": 63.935822108760476,
      "q1_ms": 63.91378189437091,
      "q3_ms": 63.98076191544533
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.00294348387978971,
      "compile_s": 0.8906802400015295,
      "inner": 3,
      "median_ms": 16.80755599712332,
      "q1_ms": 16.763242660090327,
      "q3_ms": 17.179494646067422
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.00294348387978971,
      "compile_s": 2.4228865751065314,
      "inner": 1,
      "median_ms": 53.68180898949504,
      "q1_ms": 53.64187899976969,
      "q3_ms": 53.92532795667648
    },
    {
      "dtype": "bfloat16",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.00294348387978971,
      "compile_s": 0.9067772279959172,
      "inner": 3,
      "median_ms": 14.48132866062224,
      "q1_ms": 14.353588999559483,
      "q3_ms": 14.483948626245061
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "auto",
      "rel_err": 0.0007541059167124331,
      "compile_s": 0.9405503689777106,
      "inner": 73,
      "median_ms": 0.5462602042427209,
      "q1_ms": 0.5457304922021824,
      "q3_ms": 0.5567727948551717
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "auto",
      "rel_err": 0.0007164274575188756,
      "compile_s": 1.0243748989887536,
      "inner": 24,
      "median_ms": 1.9075637489246826,
      "q1_ms": 1.903122873045504,
      "q3_ms": 1.9100454228464514
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0007621336844749749,
      "compile_s": 0.16073727793991566,
      "inner": 60,
      "median_ms": 0.7321075807946423,
      "q1_ms": 0.7188567658886313,
      "q3_ms": 0.7321527499395112
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0007621336844749749,
      "compile_s": 0.1472451421432197,
      "inner": 39,
      "median_ms": 1.1313662577707033,
      "q1_ms": 1.1131678445216937,
      "q3_ms": 1.1323185649533303
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0007621336844749749,
      "compile_s": 0.1717086690478027,
      "inner": 66,
      "median_ms": 0.672473529861732,
      "q1_ms": 0.6635140161961317,
      "q3_ms": 0.6744603478265079
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0007621336844749749,
      "compile_s": 0.15130685199983418,
      "inner": 42,
      "median_ms": 1.0728795480515276,
      "q1_ms": 1.0610648342186497,
      "q3_ms": 1.0733200525421474
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0007621336844749749,
      "compile_s": 0.17914709891192615,
      "inner": 68,
      "median_ms": 0.6664269261390847,
      "q1_ms": 0.6509734290268491,
      "q3_ms": 0.6833386347245644
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0007621336844749749,
      "compile_s": 0.15841515408828855,
      "inner": 42,
      "median_ms": 1.1014878095704177,
      "q1_ms": 1.0713402880355716,
      "q3_ms": 1.1020328106713435
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0007612487534061074,
      "compile_s": 0.1978159500285983,
      "inner": 72,
      "median_ms": 0.6125875548200889,
      "q1_ms": 0.5988345429715184,
      "q3_ms": 0.616614069763778
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0007611855398863554,
      "compile_s": 0.15941978199407458,
      "inner": 53,
      "median_ms": 0.8275215821516402,
      "q1_ms": 0.8096997757916743,
      "q3_ms": 0.8305253610366359
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0007612487534061074,
      "compile_s": 0.21284629101864994,
      "inner": 81,
      "median_ms": 0.5520804078076724,
      "q1_ms": 0.5402009255816171,
      "q3_ms": 0.563372714146052
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0007611855398863554,
      "compile_s": 0.1710787711199373,
      "inner": 57,
      "median_ms": 0.768336753377266,
      "q1_ms": 0.7634001065110951,
      "q3_ms": 0.7762139287303415
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0007612487534061074,
      "compile_s": 0.2238718420267105,
      "inner": 78,
      "median_ms": 0.5700155519522154,
      "q1_ms": 0.5532571425040563,
      "q3_ms": 0.5727402943496903
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0007611855398863554,
      "compile_s": 0.1781758030410856,
      "inner": 57,
      "median_ms": 0.773130421804493,
      "q1_ms": 0.765165526485234,
      "q3_ms": 0.7784581424570397
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0007541059167124331,
      "compile_s": 0.2777938100043684,
      "inner": 71,
      "median_ms": 0.6011064511946809,
      "q1_ms": 0.5975839137558786,
      "q3_ms": 0.6047564519333167
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0007541059167124331,
      "compile_s": 0.19640252413228154,
      "inner": 62,
      "median_ms": 0.7015100958186292,
      "q1_ms": 0.6972083243571462,
      "q3_ms": 0.7082100642184096
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0007541059167124331,
      "compile_s": 0.2918193230871111,
      "inner": 84,
      "median_ms": 0.510512357799425,
      "q1_ms": 0.5053209273942879,
      "q3_ms": 0.5212179163382167
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0007541059167124331,
      "compile_s": 0.21013994794338942,
      "inner": 72,
      "median_ms": 0.6399231788236648,
      "q1_ms": 0.63929053915975,
      "q3_ms": 0.6401795706349529
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0007541059167124331,
      "compile_s": 0.3307447109837085,
      "inner": 75,
      "median_ms": 0.5707961072524389,
      "q1_ms": 0.5685154410700003,
      "q3_ms": 0.5751746116826931
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0007541059167124331,
      "compile_s": 0.22515575913712382,
      "inner": 63,
      "median_ms": 0.6923835857638291,
      "q1_ms": 0.6910680187126946,
      "q3_ms": 0.6935053194562594
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0007621336844749749,
      "compile_s": 0.1868219489697367,
      "inner": 89,
      "median_ms": 0.4298054937566264,
      "q1_ms": 0.4278884264064974,
      "q3_ms": 0.4314090899620833
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0007621336844749749,
      "compile_s": 0.1659860739018768,
      "inner": 67,
      "median_ms": 0.6205900155230245,
      "q1_ms": 0.6201834480431098,
      "q3_ms": 0.6265338795231794
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0007621336844749749,
      "compile_s": 0.20235826796852052,
      "inner": 89,
      "median_ms": 0.44824207839922287,
      "q1_ms": 0.4397487304476875,
      "q3_ms": 0.4522452141378033
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0007621336844749749,
      "compile_s": 0.17678971611894667,
      "inner": 63,
      "median_ms": 0.657355889589304,
      "q1_ms": 0.6570920813828707,
      "q3_ms": 0.6630158731861721
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0007621336844749749,
      "compile_s": 0.21201638295315206,
      "inner": 90,
      "median_ms": 0.44965259989516604,
      "q1_ms": 0.4434173900841011,
      "q3_ms": 0.45356803263227147
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0007621336844749749,
      "compile_s": 0.18744196696206927,
      "inner": 64,
      "median_ms": 0.6569790639332496,
      "q1_ms": 0.6474409528891556,
      "q3_ms": 0.6577960775757674
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0007612487534061074,
      "compile_s": 0.23250932898372412,
      "inner": 101,
      "median_ms": 0.39546717751291716,
      "q1_ms": 0.39355935630305566,
      "q3_ms": 0.3958422284905273
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0007612487534061074,
      "compile_s": 0.20186092983931303,
      "inner": 59,
      "median_ms": 0.7078852543954627,
      "q1_ms": 0.7068478162952904,
      "q3_ms": 0.7191497943032596
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0007612487534061074,
      "compile_s": 0.25242357701063156,
      "inner": 99,
      "median_ms": 0.40361853645681733,
      "q1_ms": 0.3984543033922562,
      "q3_ms": 0.4083449891156921
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0007612487534061074,
      "compile_s": 0.21979357302188873,
      "inner": 59,
      "median_ms": 0.7132342187845605,
      "q1_ms": 0.7129079678048521,
      "q3_ms": 0.7272207975324433
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0007612487534061074,
      "compile_s": 0.2721180049702525,
      "inner": 92,
      "median_ms": 0.41806684685466083,
      "q1_ms": 0.4164656426798066,
      "q3_ms": 0.42033421746252675
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0007612487534061074,
      "compile_s": 0.22775544202886522,
      "inner": 80,
      "median_ms": 0.5477433151099831,
      "q1_ms": 0.5426066869404167,
      "q3_ms": 0.5602415389148518
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0007541059167124331,
      "compile_s": 0.37568062311038375,
      "inner": 94,
      "median_ms": 0.4037609571551389,
      "q1_ms": 0.40047362625123334,
      "q3_ms": 0.40483734065151594
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0007541059167124331,
      "compile_s": 0.29632067191414535,
      "inner": 64,
      "median_ms": 0.6501987663796172,
      "q1_ms": 0.6452358138631098,
      "q3_ms": 0.6586299823538866
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0007541059167124331,
      "compile_s": 0.4456050400622189,
      "inner": 107,
      "median_ms": 0.3818390003153097,
      "q1_ms": 0.37359929791538515,
      "q3_ms": 0.3827218984680198
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0007541059167124331,
      "compile_s": 0.3164052190259099,
      "inner": 67,
      "median_ms": 0.6225519542540632,
      "q1_ms": 0.6144616697261582,
      "q3_ms": 0.6290159549619725
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0007541059167124331,
      "compile_s": 0.44539677002467215,
      "inner": 86,
      "median_ms": 0.45591182627736826,
      "q1_ms": 0.4540046265478744,
      "q3_ms": 0.45593101314680523
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0007541059167124331,
      "compile_s": 0.3382673521991819,
      "inner": 72,
      "median_ms": 0.5879990122694937,
      "q1_ms": 0.586717069381848,
      "q3_ms": 0.5880187220302306
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0007621336844749749,
      "compile_s": 0.27904496691189706,
      "inner": 105,
      "median_ms": 0.3594321043541034,
      "q1_ms": 0.35933696088336764,
      "q3_ms": 0.36029085694324403
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0007621336844749749,
      "compile_s": 0.18552057188935578,
      "inner": 76,
      "median_ms": 0.5453273954213058,
      "q1_ms": 0.5390125290049534,
      "q3_ms": 0.5460358040995503
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0007621336844749749,
      "compile_s": 0.29538082401268184,
      "inner": 106,
      "median_ms": 0.34097976376355255,
      "q1_ms": 0.336410339376977,
      "q3_ms": 0.3440314420621912
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0007621336844749749,
      "compile_s": 0.19528566696681082,
      "inner": 77,
      "median_ms": 0.5268934273535942,
      "q1_ms": 0.5208400664730118,
      "q3_ms": 0.5270761419571452
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0007621336844749749,
      "compile_s": 0.3087326989043504,
      "inner": 109,
      "median_ms": 0.3372263590540361,
      "q1_ms": 0.3324275699640633,
      "q3_ms": 0.34060608168397477
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0007621336844749749,
      "compile_s": 0.2040105639025569,
      "inner": 98,
      "median_ms": 0.3994036541909588,
      "q1_ms": 0.39900354898477697,
      "q3_ms": 0.39952619200838463
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0007612487534061074,
      "compile_s": 0.39663884812034667,
      "inner": 123,
      "median_ms": 0.3018844077681623,
      "q1_ms": 0.2974901950304828,
      "q3_ms": 0.30329717111175625
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0007612487534061074,
      "compile_s": 0.22281202394515276,
      "inner": 92,
      "median_ms": 0.40785501123932394,
      "q1_ms": 0.4018618694871016,
      "q3_ms": 0.4081665437020685
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0007612487534061074,
      "compile_s": 0.42471592500805855,
      "inner": 125,
      "median_ms": 0.2998414561152458,
      "q1_ms": 0.29846249520778656,
      "q3_ms": 0.30556160025298595
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0007612487534061074,
      "compile_s": 0.24174804496578872,
      "inner": 92,
      "median_ms": 0.40865262069374975,
      "q1_ms": 0.4083063129497611,
      "q3_ms": 0.41347184805604426
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0007612487534061074,
      "compile_s": 0.4403921430930495,
      "inner": 130,
      "median_ms": 0.2991311395397553,
      "q1_ms": 0.2986439844020284,
      "q3_ms": 0.30118259243094003
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0007612487534061074,
      "compile_s": 0.2502151830121875,
      "inner": 105,
      "median_ms": 0.3492502185205618,
      "q1_ms": 0.34648299084178036,
      "q3_ms": 0.34988117182538625
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0007541059167124331,
      "compile_s": 0.9398732611443847,
      "inner": 106,
      "median_ms": 0.383783896225241,
      "q1_ms": 0.38097079245830484,
      "q3_ms": 0.38913151969746596
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0007541059167124331,
      "compile_s": 0.33780530211515725,
      "inner": 102,
      "median_ms": 0.37499244197034365,
      "q1_ms": 0.37247931365581116,
      "q3_ms": 0.3754035295809017
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0007541059167124331,
      "compile_s": 0.9389349229168147,
      "inner": 75,
      "median_ms": 0.5464507049570481,
      "q1_ms": 0.5463510906944673,
      "q3_ms": 0.556474132463336
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0007541059167124331,
      "compile_s": 0.3567413231357932,
      "inner": 106,
      "median_ms": 0.34825983984431,
      "q1_ms": 0.34806399193223353,
      "q3_ms": 0.3488780469370057
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0007541059167124331,
      "compile_s": 0.9540076442062855,
      "inner": 80,
      "median_ms": 0.5063331627752632,
      "q1_ms": 0.5062714248197153,
      "q3_ms": 0.5127512733452022
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0007541059167124331,
      "compile_s": 0.3792308440897614,
      "inner": 115,
      "median_ms": 0.32296046088247193,
      "q1_ms": 0.3178943819163934,
      "q3_ms": 0.32400827976348606
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0007164274575188756,
      "compile_s": 1.0174297781195492,
      "inner": 17,
      "median_ms": 2.9044570675229324,
      "q1_ms": 2.895977649399463,
      "q3_ms": 2.930757589638233
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0007164274575188756,
      "compile_s": 0.39866841305047274,
      "inner": 10,
      "median_ms": 4.676867718808353,
      "q1_ms": 4.66751370113343,
      "q3_ms": 4.700010712258518
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0007164274575188756,
      "compile_s": 1.0113635929301381,
      "inner": 16,
      "median_ms": 2.9976527439430356,
      "q1_ms": 2.9849652491975576,
      "q3_ms": 2.9990314942551777
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0007164274575188756,
      "compile_s": 0.41960222786292434,
      "inner": 10,
      "median_ms": 4.5885908883064985,
      "q1_ms": 4.573307000100613,
      "q3_ms": 4.650201788172126
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0007164274575188756,
      "compile_s": 1.0297749151941389,
      "inner": 20,
      "median_ms": 2.518718398641795,
      "q1_ms": 2.4548940593376756,
      "q3_ms": 2.5235319044440985
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0007164274575188756,
      "compile_s": 0.39820561395026743,
      "inner": 14,
      "median_ms": 3.5400521509083256,
      "q1_ms": 3.530299274383911,
      "q3_ms": 3.558560723571905
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0007164274575188756,
      "compile_s": 1.0124196310061961,
      "inner": 19,
      "median_ms": 2.5191912672629484,
      "q1_ms": 2.5158017922781015,
      "q3_ms": 2.5685332641985856
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0007164274575188756,
      "compile_s": 0.4183829608373344,
      "inner": 14,
      "median_ms": 3.4564337859462415,
      "q1_ms": 3.401433932594955,
      "q3_ms": 3.4967957929308926
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0007164274575188756,
      "compile_s": 1.0380412051454186,
      "inner": 24,
      "median_ms": 2.124521958952149,
      "q1_ms": 2.0953232500081262,
      "q3_ms": 2.173116420938944
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0007164274575188756,
      "compile_s": 0.3991902880370617,
      "inner": 14,
      "median_ms": 3.342621211361672,
      "q1_ms": 3.342381296014147,
      "q3_ms": 3.3498662025002495
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0007164274575188756,
      "compile_s": 1.0136516229249537,
      "inner": 22,
      "median_ms": 2.216649686240337,
      "q1_ms": 2.1675734103403306,
      "q3_ms": 2.227266001599756
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0007164274575188756,
      "compile_s": 0.4268824860919267,
      "inner": 15,
      "median_ms": 3.0752120073884726,
      "q1_ms": 3.0719139457990727,
      "q3_ms": 3.0789725948125124
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0007164274575188756,
      "compile_s": 1.0206907850224525,
      "inner": 19,
      "median_ms": 2.589079048974734,
      "q1_ms": 2.5801874795242363,
      "q3_ms": 2.607275842149791
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0007164274575188756,
      "compile_s": 0.5616257041692734,
      "inner": 18,
      "median_ms": 2.63714922281603,
      "q1_ms": 2.6291081707717647,
      "q3_ms": 2.6460475055500865
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0007164274575188756,
      "compile_s": 1.0105473508592695,
      "inner": 19,
      "median_ms": 2.5366149401586306,
      "q1_ms": 2.536524418055227,
      "q3_ms": 2.5797522087630473
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0007164274575188756,
      "compile_s": 0.5749260578304529,
      "inner": 19,
      "median_ms": 2.438087788361468,
      "q1_ms": 2.4257015780006586,
      "q3_ms": 2.44394199628579
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0007164274575188756,
      "compile_s": 1.0146980208810419,
      "inner": 21,
      "median_ms": 2.1601758143376735,
      "q1_ms": 2.1488758114477,
      "q3_ms": 2.1684667139890648
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0007164274575188756,
      "compile_s": 0.40133556188084185,
      "inner": 11,
      "median_ms": 4.031079440292987,
      "q1_ms": 4.023053090680729,
      "q3_ms": 4.049476722932674
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0007164274575188756,
      "compile_s": 1.0118959778919816,
      "inner": 14,
      "median_ms": 3.1971852128793086,
      "q1_ms": 3.1502917907865986,
      "q3_ms": 3.2141873546476876
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0007164274575188756,
      "compile_s": 0.4232499450445175,
      "inner": 9,
      "median_ms": 4.8901850047210855,
      "q1_ms": 4.854281769237584,
      "q3_ms": 4.906743764877319
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0007164274575188756,
      "compile_s": 1.0159268479328603,
      "inner": 25,
      "median_ms": 1.8801170773804188,
      "q1_ms": 1.8773103225976229,
      "q3_ms": 1.9286773540079594
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0007164274575188756,
      "compile_s": 0.39878815901465714,
      "inner": 15,
      "median_ms": 3.145268466323614,
      "q1_ms": 3.125215802962581,
      "q3_ms": 3.1554684663812322
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0007164274575188756,
      "compile_s": 1.0099489530548453,
      "inner": 15,
      "median_ms": 3.0146961410840354,
      "q1_ms": 2.9892015270888805,
      "q3_ms": 3.035965406646331
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0007164274575188756,
      "compile_s": 0.42460050084628165,
      "inner": 9,
      "median_ms": 4.919762671407725,
      "q1_ms": 4.858635107262267,
      "q3_ms": 4.925171560090448
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0007164274575188756,
      "compile_s": 1.015737138921395,
      "inner": 30,
      "median_ms": 1.5797975317885478,
      "q1_ms": 1.5687372690687578,
      "q3_ms": 1.5835492017989357
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0007164274575188756,
      "compile_s": 0.5075081449467689,
      "inner": 19,
      "median_ms": 2.387970060992398,
      "q1_ms": 2.3856548420888815,
      "q3_ms": 2.38812904755928
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0007164274575188756,
      "compile_s": 1.0092315250076354,
      "inner": 24,
      "median_ms": 1.9063675038826962,
      "q1_ms": 1.90310879649284,
      "q3_ms": 1.9081475038547069
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0007164274575188756,
      "compile_s": 0.5543176331557333,
      "inner": 17,
      "median_ms": 2.7082047069116553,
      "q1_ms": 2.693895360126215,
      "q3_ms": 2.7294446412912188
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0007164274575188756,
      "compile_s": 1.2764292480424047,
      "inner": 18,
      "median_ms": 2.632680944063597,
      "q1_ms": 2.6201626104820104,
      "q3_ms": 2.638351443844537
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0007164274575188756,
      "compile_s": 0.718907903181389,
      "inner": 22,
      "median_ms": 2.0628727718510413,
      "q1_ms": 2.056919186460701,
      "q3_ms": 2.0632169061255725
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0007164274575188756,
      "compile_s": 1.4159436428453773,
      "inner": 15,
      "median_ms": 3.056234664594134,
      "q1_ms": 3.0521773267537355,
      "q3_ms": 3.0635966764142117
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0007164274575188756,
      "compile_s": 0.7971869690809399,
      "inner": 21,
      "median_ms": 2.1363649007287764,
      "q1_ms": 2.13406394634928,
      "q3_ms": 2.1403382332729444
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0007164274575188756,
      "compile_s": 1.0181353918742388,
      "inner": 28,
      "median_ms": 1.6174736083485186,
      "q1_ms": 1.6070440345044648,
      "q3_ms": 1.6743495361879468
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0007164274575188756,
      "compile_s": 0.3966450940351933,
      "inner": 20,
      "median_ms": 2.3000719491392374,
      "q1_ms": 2.274832548573613,
      "q3_ms": 2.300448005553335
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0007164274575188756,
      "compile_s": 1.0131858449894935,
      "inner": 20,
      "median_ms": 2.2879775031469762,
      "q1_ms": 2.284842054359615,
      "q3_ms": 2.3174854577519
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0007164274575188756,
      "compile_s": 0.4220068180002272,
      "inner": 14,
      "median_ms": 3.299134999646672,
      "q1_ms": 3.298030700534582,
      "q3_ms": 3.2997663531984602
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0007164274575188756,
      "compile_s": 1.0173643538728356,
      "inner": 31,
      "median_ms": 1.549145004021064,
      "q1_ms": 1.5439717775030481,
      "q3_ms": 1.5536946424793814
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0007164274575188756,
      "compile_s": 0.4498252458870411,
      "inner": 25,
      "median_ms": 1.8222272396087646,
      "q1_ms": 1.8214348424226046,
      "q3_ms": 1.8381507974117994
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0007164274575188756,
      "compile_s": 1.0125425059814006,
      "inner": 21,
      "median_ms": 2.111986856020632,
      "q1_ms": 2.1027568949475173,
      "q3_ms": 2.117303055932834
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0007164274575188756,
      "compile_s": 0.48225584998726845,
      "inner": 17,
      "median_ms": 2.758276347509202,
      "q1_ms": 2.7521292898146545,
      "q3_ms": 2.7781915850937366
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0007164274575188756,
      "compile_s": 1.3908519479446113,
      "inner": 23,
      "median_ms": 1.9867843543381796,
      "q1_ms": 1.9854230846723786,
      "q3_ms": 1.9973826128989458
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0007164274575188756,
      "compile_s": 0.7404407658614218,
      "inner": 26,
      "median_ms": 1.72742242172647,
      "q1_ms": 1.7251512662579234,
      "q3_ms": 1.7586423424430764
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0007164274575188756,
      "compile_s": 1.470433710841462,
      "inner": 20,
      "median_ms": 2.2388271405361593,
      "q1_ms": 2.2349736536853015,
      "q3_ms": 2.2653760970570147
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0007164274575188756,
      "compile_s": 0.7750031361356378,
      "inner": 25,
      "median_ms": 1.7834145668894053,
      "q1_ms": 1.7768021672964096,
      "q3_ms": 1.7900241166353226
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0007164274575188756,
      "compile_s": 3.5067234239540994,
      "inner": 17,
      "median_ms": 2.6590007028597245,
      "q1_ms": 2.6544683530707567,
      "q3_ms": 2.712901120128877
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0007164274575188756,
      "compile_s": 1.0392552460543811,
      "inner": 28,
      "median_ms": 1.6159186488948762,
      "q1_ms": 1.6143914960723902,
      "q3_ms": 1.6187264949881606
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 263168, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 263168, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "auto",
      "rel_err": 0.0005889300373382866,
      "compile_s": 0.9772503189742565,
      "inner": 5,
      "median_ms": 8.15611481666565,
      "q1_ms": 8.103910833597183,
      "q3_ms": 8.282118383795023
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "auto",
      "rel_err": 0.001028962666168809,
      "compile_s": 1.0380498899612576,
      "inner": 1,
      "median_ms": 29.963342007249594,
      "q1_ms": 29.849071986973286,
      "q3_ms": 30.129991937428713
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0005950236809439957,
      "compile_s": 0.17086833389475942,
      "inner": 4,
      "median_ms": 10.95563400303945,
      "q1_ms": 10.840981733053923,
      "q3_ms": 11.051808716729283
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0005950236809439957,
      "compile_s": 0.1608653301373124,
      "inner": 2,
      "median_ms": 17.335994984023273,
      "q1_ms": 17.24645495414734,
      "q3_ms": 17.61719840578735
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0005950236809439957,
      "compile_s": 0.18019655998796225,
      "inner": 5,
      "median_ms": 10.27248720638454,
      "q1_ms": 10.158047638833523,
      "q3_ms": 10.288584977388382
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0005950236809439957,
      "compile_s": 0.16704163397662342,
      "inner": 3,
      "median_ms": 16.697206301614642,
      "q1_ms": 16.63101666296522,
      "q3_ms": 16.728049687420327
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0005950236809439957,
      "compile_s": 0.18899945705197752,
      "inner": 5,
      "median_ms": 10.174181591719389,
      "q1_ms": 10.140025429427624,
      "q3_ms": 10.23432919755578
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0005950236809439957,
      "compile_s": 0.1726956891361624,
      "inner": 3,
      "median_ms": 16.926722368225455,
      "q1_ms": 16.884505981579423,
      "q3_ms": 16.970288939774036
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0005930532934144139,
      "compile_s": 0.20846937596797943,
      "inner": 5,
      "median_ms": 9.575723018497229,
      "q1_ms": 9.07705221325159,
      "q3_ms": 9.690580563619733
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0005930167972110212,
      "compile_s": 0.1712902330327779,
      "inner": 4,
      "median_ms": 12.604784744326025,
      "q1_ms": 12.490852503105998,
      "q3_ms": 12.69279676489532
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0005930532934144139,
      "compile_s": 0.22246643900871277,
      "inner": 6,
      "median_ms": 8.445477850424746,
      "q1_ms": 8.141153686059019,
      "q3_ms": 8.658230653963983
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0005930167972110212,
      "compile_s": 0.1822736340109259,
      "inner": 4,
      "median_ms": 12.00887601589784,
      "q1_ms": 11.803911766037345,
      "q3_ms": 12.075133505277336
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0005930532934144139,
      "compile_s": 0.23178641498088837,
      "inner": 6,
      "median_ms": 8.467501145787537,
      "q1_ms": 8.465294687387845,
      "q3_ms": 8.773447014391422
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0005930167972110212,
      "compile_s": 0.1881184889934957,
      "inner": 4,
      "median_ms": 11.949481500778347,
      "q1_ms": 11.80968648986891,
      "q3_ms": 12.00892124325037
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0005889300373382866,
      "compile_s": 0.28258793195709586,
      "inner": 5,
      "median_ms": 9.137851977720857,
      "q1_ms": 9.114264184609056,
      "q3_ms": 9.143074182793498
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0005889300373382866,
      "compile_s": 0.20604871213436127,
      "inner": 4,
      "median_ms": 10.870331549085677,
      "q1_ms": 10.781161778140813,
      "q3_ms": 10.942439024802297
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0005889300373382866,
      "compile_s": 0.2984280111268163,
      "inner": 6,
      "median_ms": 7.838312846918901,
      "q1_ms": 7.812974664072195,
      "q3_ms": 8.127888664603233
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0005889300373382866,
      "compile_s": 0.22003992507234216,
      "inner": 5,
      "median_ms": 9.789646416902542,
      "q1_ms": 9.732122393324971,
      "q3_ms": 9.9598060362041
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0005889300373382866,
      "compile_s": 0.33881274494342506,
      "inner": 5,
      "median_ms": 8.736565196886659,
      "q1_ms": 8.558415807783604,
      "q3_ms": 8.955646585673094
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0005889300373382866,
      "compile_s": 0.23378504998981953,
      "inner": 4,
      "median_ms": 10.677717276848853,
      "q1_ms": 10.605819756165147,
      "q3_ms": 10.691182280424982
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0005950236809439957,
      "compile_s": 0.1922398980241269,
      "inner": 7,
      "median_ms": 6.527495836572988,
      "q1_ms": 6.477848859503865,
      "q3_ms": 6.572334162358727
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0005950236809439957,
      "compile_s": 0.17674601799808443,
      "inner": 5,
      "median_ms": 9.621732821688056,
      "q1_ms": 9.49166719801724,
      "q3_ms": 9.626794792711735
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0005950236809439957,
      "compile_s": 0.20838672597892582,
      "inner": 7,
      "median_ms": 6.807730865797827,
      "q1_ms": 6.765096581407955,
      "q3_ms": 6.9047918410173486
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0005950236809439957,
      "compile_s": 0.18858464783988893,
      "inner": 4,
      "median_ms": 10.360287968069315,
      "q1_ms": 10.256838286295533,
      "q3_ms": 10.39967272663489
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0005950236809439957,
      "compile_s": 0.21880479902029037,
      "inner": 7,
      "median_ms": 6.710222556388804,
      "q1_ms": 6.687369563483766,
      "q3_ms": 6.784455285274556
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0005950236809439957,
      "compile_s": 0.1944784119259566,
      "inner": 4,
      "median_ms": 10.21072076400742,
      "q1_ms": 10.19764324882999,
      "q3_ms": 10.25981327984482
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0005930167972110212,
      "compile_s": 0.23680853098630905,
      "inner": 8,
      "median_ms": 5.986485630273819,
      "q1_ms": 5.927129503106698,
      "q3_ms": 5.999088141834363
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0005930167972110212,
      "compile_s": 0.2130650330800563,
      "inner": 4,
      "median_ms": 9.989706508349627,
      "q1_ms": 9.906631719786674,
      "q3_ms": 10.11977851158008
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0005930167972110212,
      "compile_s": 0.25864558620378375,
      "inner": 8,
      "median_ms": 6.0872266185469925,
      "q1_ms": 6.056201760657132,
      "q3_ms": 6.094969110563397
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0005930167972110212,
      "compile_s": 0.22937489207834005,
      "inner": 4,
      "median_ms": 10.274000756908208,
      "q1_ms": 10.269008227624,
      "q3_ms": 10.27631328906864
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0005930167972110212,
      "compile_s": 0.2793157820124179,
      "inner": 7,
      "median_ms": 6.398446126175778,
      "q1_ms": 6.274745001324585,
      "q3_ms": 6.555654274831925
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0005930167972110212,
      "compile_s": 0.23705159104429185,
      "inner": 6,
      "median_ms": 8.629622519947588,
      "q1_ms": 8.617389326294264,
      "q3_ms": 8.653947501443326
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0005889300373382866,
      "compile_s": 0.3814369149040431,
      "inner": 8,
      "median_ms": 6.14215899258852,
      "q1_ms": 5.9671131311915815,
      "q3_ms": 6.144826504169032
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0005889300373382866,
      "compile_s": 0.3047747351229191,
      "inner": 5,
      "median_ms": 9.40892742946744,
      "q1_ms": 9.400757635012269,
      "q3_ms": 9.42153544165194
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0005889300373382866,
      "compile_s": 0.4515245819929987,
      "inner": 8,
      "median_ms": 5.793986143544316,
      "q1_ms": 5.712778744054958,
      "q3_ms": 5.864958482561633
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0005889300373382866,
      "compile_s": 0.3240007939748466,
      "inner": 5,
      "median_ms": 9.097440401092172,
      "q1_ms": 9.087052196264267,
      "q3_ms": 9.105230402201414
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0005889300373382866,
      "compile_s": 0.45204628095962107,
      "inner": 7,
      "median_ms": 6.850297729085598,
      "q1_ms": 6.804613695879068,
      "q3_ms": 6.872809259220958
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0005889300373382866,
      "compile_s": 0.3453502089250833,
      "inner": 5,
      "median_ms": 8.843806805089116,
      "q1_ms": 8.840403007343411,
      "q3_ms": 8.860946772620082
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0005950236809439957,
      "compile_s": 0.2836146999616176,
      "inner": 9,
      "median_ms": 5.266449543543988,
      "q1_ms": 5.25865067417423,
      "q3_ms": 5.26768843539887
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0005950236809439957,
      "compile_s": 0.19813336292281747,
      "inner": 6,
      "median_ms": 7.018005009740591,
      "q1_ms": 6.99161016382277,
      "q3_ms": 7.038903345043461
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0005950236809439957,
      "compile_s": 0.30093691498041153,
      "inner": 9,
      "median_ms": 5.143546551052068,
      "q1_ms": 5.08562998018331,
      "q3_ms": 5.276873992342089
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0005950236809439957,
      "compile_s": 0.2036302292253822,
      "inner": 6,
      "median_ms": 7.129212996611993,
      "q1_ms": 7.0615816706170635,
      "q3_ms": 7.1599211466188235
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0005950236809439957,
      "compile_s": 0.31398382084444165,
      "inner": 10,
      "median_ms": 5.080983694642782,
      "q1_ms": 5.028947908431292,
      "q3_ms": 5.1138466922566295
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0005950236809439957,
      "compile_s": 0.21158177801407874,
      "inner": 7,
      "median_ms": 6.246367968352778,
      "q1_ms": 6.227006570303014,
      "q3_ms": 6.249225132965615
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0005930167972110212,
      "compile_s": 0.4001038169953972,
      "inner": 11,
      "median_ms": 4.5075619127601385,
      "q1_ms": 4.437040283598684,
      "q3_ms": 4.52863363633779
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0005930167972110212,
      "compile_s": 0.22874193312600255,
      "inner": 8,
      "median_ms": 5.644296499667689,
      "q1_ms": 5.636421614326537,
      "q3_ms": 5.656650231685489
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0005930167972110212,
      "compile_s": 0.42879446083679795,
      "inner": 11,
      "median_ms": 4.545394437048922,
      "q1_ms": 4.473316546698863,
      "q3_ms": 4.592271627519618
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0005930167972110212,
      "compile_s": 0.24652200704440475,
      "inner": 8,
      "median_ms": 5.686388874892145,
      "q1_ms": 5.6552464957349,
      "q3_ms": 5.747422517742962
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0005930167972110212,
      "compile_s": 0.44651813502423465,
      "inner": 11,
      "median_ms": 4.540344552052292,
      "q1_ms": 4.444457548247142,
      "q3_ms": 4.60439526729963
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0005930167972110212,
      "compile_s": 0.255359044065699,
      "inner": 9,
      "median_ms": 5.22926299729281,
      "q1_ms": 5.201294216223889,
      "q3_ms": 5.303439425511493
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0005889300373382866,
      "compile_s": 0.944245127029717,
      "inner": 9,
      "median_ms": 5.667461884311503,
      "q1_ms": 5.621916433382366,
      "q3_ms": 5.761206098314788
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0005889300373382866,
      "compile_s": 0.34183415793813765,
      "inner": 9,
      "median_ms": 5.12952000523607,
      "q1_ms": 5.129146570753719,
      "q3_ms": 5.1526365439511
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0005889300373382866,
      "compile_s": 0.9488276049960405,
      "inner": 5,
      "median_ms": 8.162220614030957,
      "q1_ms": 8.070761011913419,
      "q3_ms": 8.164240838959813
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0005889300373382866,
      "compile_s": 0.3618660259526223,
      "inner": 9,
      "median_ms": 5.032980116084218,
      "q1_ms": 5.004253548880418,
      "q3_ms": 5.034430237073038
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0005889300373382866,
      "compile_s": 0.9687259329948574,
      "inner": 6,
      "median_ms": 7.742609828710556,
      "q1_ms": 7.736236477891604,
      "q3_ms": 7.743648175771038
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0005889300373382866,
      "compile_s": 0.3868953709024936,
      "inner": 10,
      "median_ms": 4.824106395244598,
      "q1_ms": 4.739289707504213,
      "q3_ms": 4.84551228582859
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.001028962666168809,
      "compile_s": 1.0586386481299996,
      "inner": 1,
      "median_ms": 50.21331901662052,
      "q1_ms": 49.806110095232725,
      "q3_ms": 50.39564915932715
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.001028962666168809,
      "compile_s": 0.46922757709398866,
      "inner": 1,
      "median_ms": 73.3137580100447,
      "q1_ms": 73.23823892511427,
      "q3_ms": 73.3757191337645
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.001028962666168809,
      "compile_s": 1.0585714988410473,
      "inner": 1,
      "median_ms": 52.82528209500015,
      "q1_ms": 51.9019749481231,
      "q3_ms": 53.51387104019523
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.001028962666168809,
      "compile_s": 0.4887270948383957,
      "inner": 1,
      "median_ms": 72.68124003894627,
      "q1_ms": 72.57354096509516,
      "q3_ms": 72.70441087894142
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.001028962666168809,
      "compile_s": 1.0535027410369366,
      "inner": 1,
      "median_ms": 41.875990107655525,
      "q1_ms": 41.175963124260306,
      "q3_ms": 41.96065105497837
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.001028962666168809,
      "compile_s": 0.4489378489088267,
      "inner": 1,
      "median_ms": 55.292845936492085,
      "q1_ms": 55.08552608080208,
      "q3_ms": 55.38145615719259
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.001028962666168809,
      "compile_s": 1.0460407109931111,
      "inner": 1,
      "median_ms": 43.294986942783,
      "q1_ms": 43.181316927075386,
      "q3_ms": 43.71236590668559
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.001028962666168809,
      "compile_s": 0.4684508480131626,
      "inner": 1,
      "median_ms": 54.70505706034601,
      "q1_ms": 54.04388904571533,
      "q3_ms": 54.749547969549894
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.001028962666168809,
      "compile_s": 1.0411112140864134,
      "inner": 1,
      "median_ms": 34.184970892965794,
      "q1_ms": 33.79019210115075,
      "q3_ms": 34.88082904368639
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.001028962666168809,
      "compile_s": 0.450213034870103,
      "inner": 1,
      "median_ms": 51.2730460613966,
      "q1_ms": 51.10053694806993,
      "q3_ms": 51.466005854308605
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.001028962666168809,
      "compile_s": 1.0425209309905767,
      "inner": 1,
      "median_ms": 36.87107400037348,
      "q1_ms": 36.083556013181806,
      "q3_ms": 36.95669397711754
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.001028962666168809,
      "compile_s": 0.4729769660625607,
      "inner": 1,
      "median_ms": 48.261265037581325,
      "q1_ms": 48.114625038579106,
      "q3_ms": 48.27771405689418
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.001028962666168809,
      "compile_s": 1.0668274168856442,
      "inner": 1,
      "median_ms": 40.49742384813726,
      "q1_ms": 40.29604489915073,
      "q3_ms": 45.25668197311461
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.001028962666168809,
      "compile_s": 0.5843351560179144,
      "inner": 1,
      "median_ms": 40.27645499445498,
      "q1_ms": 40.23207491263747,
      "q3_ms": 40.30296509154141
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.001028962666168809,
      "compile_s": 1.0492823631502688,
      "inner": 1,
      "median_ms": 41.40990204177797,
      "q1_ms": 40.97092314623296,
      "q3_ms": 41.454471880570054
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.001028962666168809,
      "compile_s": 0.6120956740342081,
      "inner": 1,
      "median_ms": 38.052490912377834,
      "q1_ms": 37.87866095080972,
      "q3_ms": 38.22099999524653
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.001028962666168809,
      "compile_s": 1.0459636619780213,
      "inner": 1,
      "median_ms": 31.829946907237172,
      "q1_ms": 31.36181808076799,
      "q3_ms": 32.82814519479871
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.001028962666168809,
      "compile_s": 0.43793730810284615,
      "inner": 1,
      "median_ms": 41.632480919361115,
      "q1_ms": 41.53107199817896,
      "q3_ms": 41.71905107796192
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.001028962666168809,
      "compile_s": 1.062438759021461,
      "inner": 1,
      "median_ms": 49.12325204350054,
      "q1_ms": 49.06944208778441,
      "q3_ms": 49.714110093191266
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.001028962666168809,
      "compile_s": 0.49372008186765015,
      "inner": 1,
      "median_ms": 75.350133003667,
      "q1_ms": 75.33091306686401,
      "q3_ms": 75.76428283937275
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.001028962666168809,
      "compile_s": 1.0600937351118773,
      "inner": 1,
      "median_ms": 29.470903100445867,
      "q1_ms": 29.446663102135062,
      "q3_ms": 29.518963070586324
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.001028962666168809,
      "compile_s": 0.44591148709878325,
      "inner": 1,
      "median_ms": 47.82478604465723,
      "q1_ms": 47.781425062566996,
      "q3_ms": 48.18712500855327
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.001028962666168809,
      "compile_s": 1.0592228469904512,
      "inner": 1,
      "median_ms": 46.461507910862565,
      "q1_ms": 46.4602189604193,
      "q3_ms": 46.48108803667128
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.001028962666168809,
      "compile_s": 0.4954816079698503,
      "inner": 1,
      "median_ms": 75.86677209474146,
      "q1_ms": 75.81141195259988,
      "q3_ms": 75.87541197426617
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.001028962666168809,
      "compile_s": 1.0380128431133926,
      "inner": 2,
      "median_ms": 24.729775614105165,
      "q1_ms": 24.666485493071377,
      "q3_ms": 24.751700460910797
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.001028962666168809,
      "compile_s": 0.5441965409554541,
      "inner": 1,
      "median_ms": 37.00598399154842,
      "q1_ms": 36.917534191161394,
      "q3_ms": 37.04470302909613
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.001028962666168809,
      "compile_s": 1.0392245498951524,
      "inner": 1,
      "median_ms": 29.759792843833566,
      "q1_ms": 29.50719278305769,
      "q3_ms": 30.003861989825964
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.001028962666168809,
      "compile_s": 0.5951949381269515,
      "inner": 1,
      "median_ms": 42.247400153428316,
      "q1_ms": 42.21585998311639,
      "q3_ms": 42.288299882784486
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.001028962666168809,
      "compile_s": 1.3256418318487704,
      "inner": 1,
      "median_ms": 39.85739592462778,
      "q1_ms": 39.618836948648095,
      "q3_ms": 40.420104982331395
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.001028962666168809,
      "compile_s": 0.750517952023074,
      "inner": 1,
      "median_ms": 31.574836932122707,
      "q1_ms": 31.42929798923433,
      "q3_ms": 31.95287613198161
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.001028962666168809,
      "compile_s": 1.4579729968681931,
      "inner": 1,
      "median_ms": 46.1661999579519,
      "q1_ms": 46.16131982766092,
      "q3_ms": 46.2390489410609
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.001028962666168809,
      "compile_s": 0.8301874750759453,
      "inner": 1,
      "median_ms": 32.76974498294294,
      "q1_ms": 32.55855501629412,
      "q3_ms": 32.80083416029811
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.001028962666168809,
      "compile_s": 1.0380227218847722,
      "inner": 2,
      "median_ms": 25.25208902079612,
      "q1_ms": 25.13514447491616,
      "q3_ms": 25.99758200813085
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.001028962666168809,
      "compile_s": 0.43146250513382256,
      "inner": 1,
      "median_ms": 34.656648989766836,
      "q1_ms": 34.61035015061498,
      "q3_ms": 35.12049908749759
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.001028962666168809,
      "compile_s": 1.0509354488458484,
      "inner": 1,
      "median_ms": 35.16406915150583,
      "q1_ms": 34.70531897619367,
      "q3_ms": 35.263147903606296
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.001028962666168809,
      "compile_s": 0.47567812888883054,
      "inner": 1,
      "median_ms": 50.465169129893184,
      "q1_ms": 50.45849783346057,
      "q3_ms": 50.92049692757428
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.001028962666168809,
      "compile_s": 1.0356796369887888,
      "inner": 2,
      "median_ms": 23.49831396713853,
      "q1_ms": 23.422624450176954,
      "q3_ms": 23.57297344133258
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.001028962666168809,
      "compile_s": 0.4788014388177544,
      "inner": 1,
      "median_ms": 27.790107997134328,
      "q1_ms": 27.458808850497007,
      "q3_ms": 27.873018058016896
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.001028962666168809,
      "compile_s": 1.0440521640703082,
      "inner": 1,
      "median_ms": 32.89957414381206,
      "q1_ms": 32.79937501065433,
      "q3_ms": 33.01729494705796
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.001028962666168809,
      "compile_s": 0.5251114089041948,
      "inner": 1,
      "median_ms": 42.40144998766482,
      "q1_ms": 42.28653013706207,
      "q3_ms": 42.5180490128696
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.001028962666168809,
      "compile_s": 1.4062037849798799,
      "inner": 1,
      "median_ms": 31.058990163728595,
      "q1_ms": 31.05454915203154,
      "q3_ms": 31.22039884328842
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.001028962666168809,
      "compile_s": 0.7484809700399637,
      "inner": 1,
      "median_ms": 26.408682111650705,
      "q1_ms": 26.378260925412178,
      "q3_ms": 26.56420087441802
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.001028962666168809,
      "compile_s": 1.4982783650048077,
      "inner": 1,
      "median_ms": 34.12368195131421,
      "q1_ms": 34.098261035978794,
      "q3_ms": 34.12582096643746
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.001028962666168809,
      "compile_s": 0.7991495181340724,
      "inner": 1,
      "median_ms": 27.377568883821368,
      "q1_ms": 27.363409055396914,
      "q3_ms": 27.3818199057132
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.001028962666168809,
      "compile_s": 3.5203097930643708,
      "inner": 1,
      "median_ms": 43.03605784662068,
      "q1_ms": 42.83650894649327,
      "q3_ms": 43.42654813081026
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.001028962666168809,
      "compile_s": 1.0616978278849274,
      "inner": 2,
      "median_ms": 24.219321901910007,
      "q1_ms": 24.218446924351156,
      "q3_ms": 24.261222453787923
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 263168, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 64,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 263168, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "auto",
      "rel_err": 0.0006167562096379697,
      "compile_s": 0.5390457131434232,
      "inner": 56,
      "median_ms": 0.7196699277431305,
      "q1_ms": 0.719108124030754,
      "q3_ms": 0.7361236410880727
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "auto",
      "rel_err": 0.0008851464372128248,
      "compile_s": 1.5039671012200415,
      "inner": 7,
      "median_ms": 6.08415126667491,
      "q1_ms": 6.058284281087773,
      "q3_ms": 6.145116873085499
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0006184946396388113,
      "compile_s": 0.22191631514579058,
      "inner": 40,
      "median_ms": 1.2380015512462705,
      "q1_ms": 1.2024678755551577,
      "q3_ms": 1.2600495014339685
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0006184946396388113,
      "compile_s": 0.17647430207580328,
      "inner": 25,
      "median_ms": 1.8369112443178892,
      "q1_ms": 1.8075840827077627,
      "q3_ms": 1.8469403963536024
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0006184946396388113,
      "compile_s": 0.2289956861641258,
      "inner": 42,
      "median_ms": 1.1056049760164959,
      "q1_ms": 1.100053833354087,
      "q3_ms": 1.1462565467116377
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0006184946396388113,
      "compile_s": 0.18618135689757764,
      "inner": 28,
      "median_ms": 1.7089623558734144,
      "q1_ms": 1.7039095317678792,
      "q3_ms": 1.7346479996506656
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0006184946396388113,
      "compile_s": 0.2432646998204291,
      "inner": 42,
      "median_ms": 1.1230696912943607,
      "q1_ms": 1.0941569260986788,
      "q3_ms": 1.1573196222473467
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0006184946396388113,
      "compile_s": 0.19514510384760797,
      "inner": 28,
      "median_ms": 1.6862292125421976,
      "q1_ms": 1.6660885303281248,
      "q3_ms": 1.6897691779636912
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0006201686919666827,
      "compile_s": 0.2768164621666074,
      "inner": 44,
      "median_ms": 1.0829053618098523,
      "q1_ms": 1.030972793655978,
      "q3_ms": 1.0900812710381367
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0006201686919666827,
      "compile_s": 0.2007280399557203,
      "inner": 34,
      "median_ms": 1.3341044678407556,
      "q1_ms": 1.33074978914331,
      "q3_ms": 1.34148561697015
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0006201686919666827,
      "compile_s": 0.2982457869220525,
      "inner": 50,
      "median_ms": 0.9695251006633043,
      "q1_ms": 0.937803378328681,
      "q3_ms": 0.9747918602079153
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0006201686919666827,
      "compile_s": 0.2174710959661752,
      "inner": 39,
      "median_ms": 1.2167835106643348,
      "q1_ms": 1.2031574136553667,
      "q3_ms": 1.2213096428567018
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0006201686919666827,
      "compile_s": 0.3374374450650066,
      "inner": 43,
      "median_ms": 1.0702451420298151,
      "q1_ms": 1.0676049075067737,
      "q3_ms": 1.0709293013395265
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0006201686919666827,
      "compile_s": 0.23070326214656234,
      "inner": 36,
      "median_ms": 1.2885722228222425,
      "q1_ms": 1.2853327724668715,
      "q3_ms": 1.2977385841723945
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0006167562096379697,
      "compile_s": 0.42494230903685093,
      "inner": 46,
      "median_ms": 1.0192766937467717,
      "q1_ms": 1.0030761097922272,
      "q3_ms": 1.023063431089015
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0006167562096379697,
      "compile_s": 0.2666357890702784,
      "inner": 37,
      "median_ms": 1.210088759811746,
      "q1_ms": 1.2093163284196242,
      "q3_ms": 1.22723332606256
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0006167562096379697,
      "compile_s": 0.47155598690733314,
      "inner": 46,
      "median_ms": 0.9953521961427252,
      "q1_ms": 0.9806083221717374,
      "q3_ms": 0.9976917644962668
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0006167562096379697,
      "compile_s": 0.2863338668830693,
      "inner": 39,
      "median_ms": 1.1274640042430315,
      "q1_ms": 1.1166604355168648,
      "q3_ms": 1.1281573321097171
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 344064, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 344064, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0006184946396388113,
      "compile_s": 0.27621287386864424,
      "inner": 39,
      "median_ms": 1.1164191572998579,
      "q1_ms": 1.0945692013662596,
      "q3_ms": 1.1164212050155187
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0006184946396388113,
      "compile_s": 0.21805642382241786,
      "inner": 26,
      "median_ms": 1.6651444960958683,
      "q1_ms": 1.6618787687128553,
      "q3_ms": 1.70895942630103
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0006184946396388113,
      "compile_s": 0.2995389939751476,
      "inner": 39,
      "median_ms": 1.099620235319703,
      "q1_ms": 1.0959579227253413,
      "q3_ms": 1.1043684342159676
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0006184946396388113,
      "compile_s": 0.23390407301485538,
      "inner": 25,
      "median_ms": 1.7500046826899052,
      "q1_ms": 1.7273703217506409,
      "q3_ms": 1.7640946339815855
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0006184946396388113,
      "compile_s": 0.318297334946692,
      "inner": 41,
      "median_ms": 1.0592236081365405,
      "q1_ms": 1.0554601960792773,
      "q3_ms": 1.062263609705175
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0006184946396388113,
      "compile_s": 0.24643643200397491,
      "inner": 29,
      "median_ms": 1.5441628959800664,
      "q1_ms": 1.5429715176337753,
      "q3_ms": 1.5817390000126486
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.000620233069639653,
      "compile_s": 0.3827411679085344,
      "inner": 44,
      "median_ms": 1.002768747804856,
      "q1_ms": 1.0020942268469795,
      "q3_ms": 1.0117855642668225
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.000620233069639653,
      "compile_s": 0.2786394681315869,
      "inner": 31,
      "median_ms": 1.4089837429984924,
      "q1_ms": 1.4072650622936986,
      "q3_ms": 1.4294295106083155
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.000620233069639653,
      "compile_s": 0.4545201410073787,
      "inner": 36,
      "median_ms": 1.170248362339205,
      "q1_ms": 1.1679241951141093,
      "q3_ms": 1.2069957511913445
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.000620233069639653,
      "compile_s": 0.3022695870604366,
      "inner": 31,
      "median_ms": 1.4068631290067588,
      "q1_ms": 1.405092166556466,
      "q3_ms": 1.4262666114636005
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.000620233069639653,
      "compile_s": 0.45752824400551617,
      "inner": 38,
      "median_ms": 1.1212302641452927,
      "q1_ms": 1.1192802401063473,
      "q3_ms": 1.1486191314792162
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.000620233069639653,
      "compile_s": 0.32087753689847887,
      "inner": 34,
      "median_ms": 1.319284235839458,
      "q1_ms": 1.294363968019538,
      "q3_ms": 1.3193909955375334
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0006167562096379697,
      "compile_s": 0.7382798360195011,
      "inner": 41,
      "median_ms": 1.0573667837534009,
      "q1_ms": 1.055115072919828,
      "q3_ms": 1.0600801750381545
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0006167562096379697,
      "compile_s": 0.42225725506432354,
      "inner": 33,
      "median_ms": 1.3558346651155841,
      "q1_ms": 1.3260138503304033,
      "q3_ms": 1.359369217728575
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0006167562096379697,
      "compile_s": 0.8350183549337089,
      "inner": 36,
      "median_ms": 1.249047028573437,
      "q1_ms": 1.243565108678821,
      "q3_ms": 1.2607264410083492
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0006167562096379697,
      "compile_s": 0.4775522218551487,
      "inner": 34,
      "median_ms": 1.281107563580222,
      "q1_ms": 1.2650514459785294,
      "q3_ms": 1.283776672447429
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 360448, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 360448, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0006184946396388113,
      "compile_s": 0.5889145138207823,
      "inner": 50,
      "median_ms": 0.8909336989745498,
      "q1_ms": 0.8824945008382201,
      "q3_ms": 0.894360882230103
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0006184946396388113,
      "compile_s": 0.25969031685963273,
      "inner": 46,
      "median_ms": 0.9067663467367706,
      "q1_ms": 0.901961567528222,
      "q3_ms": 0.9079422174102586
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0006184946396388113,
      "compile_s": 0.6151125549804419,
      "inner": 52,
      "median_ms": 0.8612446925746133,
      "q1_ms": 0.8495597106118041,
      "q3_ms": 0.8651981524263437
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0006184946396388113,
      "compile_s": 0.28101042099297047,
      "inner": 46,
      "median_ms": 0.9241563055421348,
      "q1_ms": 0.9135502420932702,
      "q3_ms": 0.931151935061359
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0006184946396388113,
      "compile_s": 0.6306102050002664,
      "inner": 39,
      "median_ms": 1.1282947678596544,
      "q1_ms": 1.122714002401783,
      "q3_ms": 1.1329322050397213
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0006184946396388113,
      "compile_s": 0.28734961501322687,
      "inner": 54,
      "median_ms": 0.7971508894115686,
      "q1_ms": 0.7889542418221632,
      "q3_ms": 0.7983784803568765
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.000620233069639653,
      "compile_s": 0.9066027100197971,
      "inner": 56,
      "median_ms": 0.823495356598869,
      "q1_ms": 0.812240389807682,
      "q3_ms": 0.8247380361094007
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.000620233069639653,
      "compile_s": 0.3361011389642954,
      "inner": 54,
      "median_ms": 0.7902451670142236,
      "q1_ms": 0.7856025929666228,
      "q3_ms": 0.7933090587526008
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.000620233069639653,
      "compile_s": 0.9588207239285111,
      "inner": 42,
      "median_ms": 1.0234154483658215,
      "q1_ms": 1.0223749781116134,
      "q3_ms": 1.040278978291012
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.000620233069639653,
      "compile_s": 0.3588984489906579,
      "inner": 53,
      "median_ms": 0.7726680004160922,
      "q1_ms": 0.7689331157378992,
      "q3_ms": 0.7946771912206456
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.000620233069639653,
      "compile_s": 0.995421099010855,
      "inner": 44,
      "median_ms": 0.9714738381179897,
      "q1_ms": 0.9706297529522668,
      "q3_ms": 0.9877594791098752
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.000620233069639653,
      "compile_s": 0.3816517509985715,
      "inner": 57,
      "median_ms": 0.7207804040885285,
      "q1_ms": 0.7198184744961429,
      "q3_ms": 0.7403992982417869
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0006167562096379697,
      "compile_s": 1.995964135043323,
      "inner": 37,
      "median_ms": 1.1819437007746987,
      "q1_ms": 1.1795020813273418,
      "q3_ms": 1.1925068935631093
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0006167562096379697,
      "compile_s": 0.5426974731963128,
      "inner": 57,
      "median_ms": 0.7281100516321889,
      "q1_ms": 0.7199818072350401,
      "q3_ms": 0.7364815958824596
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 393216, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 393216, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0008851464372128248,
      "compile_s": 2.1005086249206215,
      "inner": 8,
      "median_ms": 6.009814504068345,
      "q1_ms": 5.975390755338594,
      "q3_ms": 6.019036867655814
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0008851464372128248,
      "compile_s": 0.6145124770700932,
      "inner": 4,
      "median_ms": 10.238618473522365,
      "q1_ms": 10.207748506218195,
      "q3_ms": 10.296840802766383
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0008851464372128248,
      "compile_s": 2.0742729920893908,
      "inner": 10,
      "median_ms": 4.909899295307696,
      "q1_ms": 4.8306084936484694,
      "q3_ms": 4.962365096434951
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0008851464372128248,
      "compile_s": 0.6190034050960094,
      "inner": 6,
      "median_ms": 7.229116163216531,
      "q1_ms": 7.207862994012733,
      "q3_ms": 7.334749330766499
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0008851464372128248,
      "compile_s": 2.088086560834199,
      "inner": 9,
      "median_ms": 5.103326868265867,
      "q1_ms": 5.102004674780701,
      "q3_ms": 5.164047780757149
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0008851464372128248,
      "compile_s": 0.6153788201045245,
      "inner": 9,
      "median_ms": 5.45108822795252,
      "q1_ms": 5.423514896796809,
      "q3_ms": 5.45351476305061
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0008851464372128248,
      "compile_s": 2.1106402850709856,
      "inner": 4,
      "median_ms": 11.148583784233779,
      "q1_ms": 10.996456781867892,
      "q3_ms": 11.161986272782087
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0008851464372128248,
      "compile_s": 0.8889282520394772,
      "inner": 10,
      "median_ms": 4.919167305342853,
      "q1_ms": 4.850220517255366,
      "q3_ms": 4.923762218095362
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0008851464372128248,
      "compile_s": 2.0871846261434257,
      "inner": 9,
      "median_ms": 5.154941101661986,
      "q1_ms": 5.1416977722611685,
      "q3_ms": 5.202275442166461
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0008851464372128248,
      "compile_s": 0.6168643049895763,
      "inner": 6,
      "median_ms": 7.288589491508901,
      "q1_ms": 7.228486356325448,
      "q3_ms": 7.3156995155538125
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0008851464372128248,
      "compile_s": 2.090059149079025,
      "inner": 8,
      "median_ms": 6.0419006331358105,
      "q1_ms": 6.039828120265156,
      "q3_ms": 6.043274392141029
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0008851464372128248,
      "compile_s": 0.626910189865157,
      "inner": 8,
      "median_ms": 6.0205670015420765,
      "q1_ms": 5.946285877143964,
      "q3_ms": 6.052358134184033
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0008851464372128248,
      "compile_s": 2.1022078269161284,
      "inner": 8,
      "median_ms": 5.77833887655288,
      "q1_ms": 5.76317249215208,
      "q3_ms": 5.80294374958612
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0008851464372128248,
      "compile_s": 0.8515802789479494,
      "inner": 9,
      "median_ms": 5.297909671854641,
      "q1_ms": 5.294967339270645,
      "q3_ms": 5.327107462411125
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0008851464372128248,
      "compile_s": 3.580690657021478,
      "inner": 6,
      "median_ms": 7.134614861570299,
      "q1_ms": 7.080040018384655,
      "q3_ms": 7.1400965098291636
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0008851464372128248,
      "compile_s": 1.2997925609815866,
      "inner": 6,
      "median_ms": 6.965253502130508,
      "q1_ms": 6.953115497405331,
      "q3_ms": 6.98229515304168
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0008851464372128248,
      "compile_s": 2.0875596951227635,
      "inner": 6,
      "median_ms": 7.270131182546417,
      "q1_ms": 7.1653882041573524,
      "q3_ms": 7.289822795428336
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0008851464372128248,
      "compile_s": 0.6095081050880253,
      "inner": 11,
      "median_ms": 4.22420998273248,
      "q1_ms": 4.213933621279217,
      "q3_ms": 4.243097268044949
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0008851464372128248,
      "compile_s": 2.0820061799604446,
      "inner": 8,
      "median_ms": 5.863271129783243,
      "q1_ms": 5.842476122779772,
      "q3_ms": 5.908624763833359
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0008851464372128248,
      "compile_s": 0.7655473318882287,
      "inner": 10,
      "median_ms": 4.422081611119211,
      "q1_ms": 4.3992456048727036,
      "q3_ms": 4.480072413571179
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0008851464372128248,
      "compile_s": 3.2751988258678466,
      "inner": 8,
      "median_ms": 5.4052009945735335,
      "q1_ms": 5.394227366195992,
      "q3_ms": 5.482475855387747
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0008851464372128248,
      "compile_s": 1.5999067751690745,
      "inner": 11,
      "median_ms": 4.232639087025414,
      "q1_ms": 4.210348177531904,
      "q3_ms": 4.251033639196645
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 327680, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 327680, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 2048,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "auto",
      "rel_err": 0.0005071765626780689,
      "compile_s": 0.5499883689917624,
      "inner": 4,
      "median_ms": 10.87711175205186,
      "q1_ms": 10.763697267975658,
      "q3_ms": 11.030091729480773
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "auto",
      "rel_err": 0.0007406402728520334,
      "compile_s": 1.612728272099048,
      "inner": 1,
      "median_ms": 92.71684102714062,
      "q1_ms": 92.52406097948551,
      "q3_ms": 92.73307002149522
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0005035605281591415,
      "compile_s": 0.23705023690126836,
      "inner": 2,
      "median_ms": 18.845326034352183,
      "q1_ms": 18.389967503026128,
      "q3_ms": 19.138755975291133
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0005035605281591415,
      "compile_s": 0.2013882000464946,
      "inner": 1,
      "median_ms": 28.120307018980384,
      "q1_ms": 27.79672807082534,
      "q3_ms": 28.64432497881353
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0005035605281591415,
      "compile_s": 0.24362373002804816,
      "inner": 3,
      "median_ms": 17.309361991162103,
      "q1_ms": 16.933832705641787,
      "q3_ms": 17.635387601330876
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0005035605281591415,
      "compile_s": 0.20879482990130782,
      "inner": 1,
      "median_ms": 27.23276987671852,
      "q1_ms": 26.863690000027418,
      "q3_ms": 27.43112901225686
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0005035605281591415,
      "compile_s": 0.2561164880171418,
      "inner": 3,
      "median_ms": 17.892996935794752,
      "q1_ms": 17.846190293009084,
      "q3_ms": 18.092330001915496
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0005035605281591415,
      "compile_s": 0.2213704080786556,
      "inner": 1,
      "median_ms": 26.933490065857768,
      "q1_ms": 26.59825119189918,
      "q3_ms": 27.05551008693874
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0005050370818935335,
      "compile_s": 0.2908705079462379,
      "inner": 3,
      "median_ms": 16.525200645749766,
      "q1_ms": 16.171574670200545,
      "q3_ms": 16.79425671075781
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.000504976836964488,
      "compile_s": 0.21920336317270994,
      "inner": 2,
      "median_ms": 21.051155519671738,
      "q1_ms": 20.823576487600803,
      "q3_ms": 21.104570478200912
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0005050370818935335,
      "compile_s": 0.3129436180461198,
      "inner": 3,
      "median_ms": 14.597231987863779,
      "q1_ms": 14.485545689240098,
      "q3_ms": 15.075641063352426
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.000504976836964488,
      "compile_s": 0.23468460119329393,
      "inner": 2,
      "median_ms": 18.967025447636843,
      "q1_ms": 18.891426036134362,
      "q3_ms": 18.994466052390635
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0005050370818935335,
      "compile_s": 0.3550034989602864,
      "inner": 3,
      "median_ms": 16.442804054046672,
      "q1_ms": 16.345720971003175,
      "q3_ms": 16.559490313132603
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.000504976836964488,
      "compile_s": 0.24555883300490677,
      "inner": 2,
      "median_ms": 20.005712984129786,
      "q1_ms": 19.981003017164767,
      "q3_ms": 20.311227533966303
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0005071765626780689,
      "compile_s": 0.4388106109108776,
      "inner": 3,
      "median_ms": 15.568006318062544,
      "q1_ms": 15.534729696810246,
      "q3_ms": 15.619189711287618
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0005071765626780689,
      "compile_s": 0.2831520759500563,
      "inner": 2,
      "median_ms": 18.551866989582777,
      "q1_ms": 18.50787689909339,
      "q3_ms": 18.821680918335915
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0005071765626780689,
      "compile_s": 0.4868525171186775,
      "inner": 3,
      "median_ms": 15.395410048464933,
      "q1_ms": 15.17223403789103,
      "q3_ms": 15.419916637862721
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0005071765626780689,
      "compile_s": 0.301057409029454,
      "inner": 3,
      "median_ms": 17.288375335435074,
      "q1_ms": 17.257952053720754,
      "q3_ms": 17.30763167142868
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 344064, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 32,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 344064, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0005035605281591415,
      "compile_s": 0.29347242903895676,
      "inner": 2,
      "median_ms": 17.357604927383363,
      "q1_ms": 17.34689506702125,
      "q3_ms": 17.618963960558176
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0005035605281591415,
      "compile_s": 0.24310888908803463,
      "inner": 1,
      "median_ms": 25.067054899409413,
      "q1_ms": 25.042515015229583,
      "q3_ms": 25.06752498447895
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0005035605281591415,
      "compile_s": 0.31805903487838805,
      "inner": 2,
      "median_ms": 17.023840453475714,
      "q1_ms": 17.0236510457471,
      "q3_ms": 17.215000465512276
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0005035605281591415,
      "compile_s": 0.26113412296399474,
      "inner": 1,
      "median_ms": 27.291968930512667,
      "q1_ms": 26.901850011199713,
      "q3_ms": 27.391859097406268
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0005035605281591415,
      "compile_s": 0.3347204620949924,
      "inner": 3,
      "median_ms": 16.34485103810827,
      "q1_ms": 16.324917708213132,
      "q3_ms": 16.513280648117263
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0005035605281591415,
      "compile_s": 0.2725054428447038,
      "inner": 2,
      "median_ms": 24.11425206810236,
      "q1_ms": 23.8486435264349,
      "q3_ms": 24.19013192411512
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.000504976836964488,
      "compile_s": 0.3988891248591244,
      "inner": 3,
      "median_ms": 15.195520672326287,
      "q1_ms": 15.173450655614337,
      "q3_ms": 15.270720313613614
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.000504976836964488,
      "compile_s": 0.3003675208892673,
      "inner": 2,
      "median_ms": 20.639640977606177,
      "q1_ms": 20.3551824670285,
      "q3_ms": 20.690495963208377
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.000504976836964488,
      "compile_s": 0.47708192211575806,
      "inner": 2,
      "median_ms": 18.226943095214665,
      "q1_ms": 17.820963519625366,
      "q3_ms": 18.25007749721408
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.000504976836964488,
      "compile_s": 0.321032838197425,
      "inner": 2,
      "median_ms": 21.404559491202235,
      "q1_ms": 21.35893958620727,
      "q3_ms": 21.94627351127565
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.000504976836964488,
      "compile_s": 0.4739667109679431,
      "inner": 2,
      "median_ms": 17.236390034668148,
      "q1_ms": 17.202060553245246,
      "q3_ms": 17.526779556646943
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.000504976836964488,
      "compile_s": 0.34140903502702713,
      "inner": 2,
      "median_ms": 20.266057457774878,
      "q1_ms": 20.157602499239147,
      "q3_ms": 20.3228120226413
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0005071765626780689,
      "compile_s": 0.7548474431969225,
      "inner": 3,
      "median_ms": 16.26866132331391,
      "q1_ms": 16.254214725146692,
      "q3_ms": 16.352834335217874
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0005071765626780689,
      "compile_s": 0.4430165111552924,
      "inner": 2,
      "median_ms": 19.23915510997176,
      "q1_ms": 18.912110943347216,
      "q3_ms": 19.313324941322207
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0005071765626780689,
      "compile_s": 0.8548405629117042,
      "inner": 2,
      "median_ms": 19.17044003494084,
      "q1_ms": 19.059715559706092,
      "q3_ms": 19.212205545045435
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0005071765626780689,
      "compile_s": 0.4994680949021131,
      "inner": 2,
      "median_ms": 19.508554483763874,
      "q1_ms": 19.125585560686886,
      "q3_ms": 19.522104528732598
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 360448, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 64,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 360448, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0005035605281591415,
      "compile_s": 0.6086871209554374,
      "inner": 3,
      "median_ms": 13.665151316672564,
      "q1_ms": 13.50128169481953,
      "q3_ms": 13.844140650083622
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0005035605281591415,
      "compile_s": 0.27256723307073116,
      "inner": 3,
      "median_ms": 13.894023994604746,
      "q1_ms": 13.811334346731504,
      "q3_ms": 13.904230669140816
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.0005035605281591415,
      "compile_s": 0.6336784560699016,
      "inner": 3,
      "median_ms": 13.024529286970695,
      "q1_ms": 12.909413004914919,
      "q3_ms": 13.248081939915815
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.0005035605281591415,
      "compile_s": 0.29089984600432217,
      "inner": 3,
      "median_ms": 14.231983028973142,
      "q1_ms": 14.207513304427266,
      "q3_ms": 14.268539690723022
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.0005035605281591415,
      "compile_s": 0.6506376229226589,
      "inner": 2,
      "median_ms": 17.25516002625227,
      "q1_ms": 17.240145010873675,
      "q3_ms": 17.540524480864406
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 32,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.0005035605281591415,
      "compile_s": 0.29877770598977804,
      "inner": 4,
      "median_ms": 11.977978749200702,
      "q1_ms": 11.943641526158899,
      "q3_ms": 12.078428757376969
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.000504976836964488,
      "compile_s": 0.9172794709447771,
      "inner": 4,
      "median_ms": 12.179543497040868,
      "q1_ms": 12.003061536233872,
      "q3_ms": 12.25654548034072
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.000504976836964488,
      "compile_s": 0.3463086220435798,
      "inner": 4,
      "median_ms": 11.891761503648013,
      "q1_ms": 11.881956772413105,
      "q3_ms": 12.243243225384504
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 4,
      "rel_err": 0.000504976836964488,
      "compile_s": 0.9724279891233891,
      "inner": 3,
      "median_ms": 15.754772660632927,
      "q1_ms": 15.672882630800206,
      "q3_ms": 15.761259011924267
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 2,
      "num_warps": 8,
      "rel_err": 0.000504976836964488,
      "compile_s": 0.3711116770282388,
      "inner": 4,
      "median_ms": 11.816842015832663,
      "q1_ms": 11.644947284366935,
      "q3_ms": 11.94233150454238
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 4,
      "rel_err": 0.000504976836964488,
      "compile_s": 1.006020720815286,
      "inner": 3,
      "median_ms": 14.96099098585546,
      "q1_ms": 14.947871367136637,
      "q3_ms": 15.009124375258883
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 64,
      "num_stages": 3,
      "num_warps": 8,
      "rel_err": 0.000504976836964488,
      "compile_s": 0.39333659992553294,
      "inner": 4,
      "median_ms": 11.056646530050784,
      "q1_ms": 10.979804035741836,
      "q3_ms": 11.311337992083281
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0005071765626780689,
      "compile_s": 2.013757527107373,
      "inner": 2,
      "median_ms": 18.486707005649805,
      "q1_ms": 18.27924244571477,
      "q3_ms": 18.524812068790197
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0005071765626780689,
      "compile_s": 0.5508197718299925,
      "inner": 4,
      "median_ms": 10.863951756618917,
      "q1_ms": 10.855974513106048,
      "q3_ms": 11.09381124842912
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 2,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 393216, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd",
      "config": "forced",
      "block_q": 128,
      "block_k": 128,
      "num_stages": 3,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 393216, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0007406402728520334,
      "compile_s": 2.1884828549809754,
      "inner": 1,
      "median_ms": 95.40857304818928,
      "q1_ms": 95.21914296783507,
      "q3_ms": 95.50212207250297
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0007406402728520334,
      "compile_s": 0.7669717711396515,
      "inner": 1,
      "median_ms": 158.71750912629068,
      "q1_ms": 158.4669689182192,
      "q3_ms": 170.8839770872146
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0007406402728520334,
      "compile_s": 2.164115706924349,
      "inner": 1,
      "median_ms": 89.3809178378433,
      "q1_ms": 87.74579293094575,
      "q3_ms": 89.79769703000784
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0007406402728520334,
      "compile_s": 0.7330522688571364,
      "inner": 1,
      "median_ms": 113.07636718265712,
      "q1_ms": 112.90366691537201,
      "q3_ms": 113.22650592774153
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0007406402728520334,
      "compile_s": 2.243413511896506,
      "inner": 1,
      "median_ms": 100.94996797852218,
      "q1_ms": 100.82341800443828,
      "q3_ms": 102.37139509990811
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0007406402728520334,
      "compile_s": 0.6947603181470186,
      "inner": 1,
      "median_ms": 84.3337110709399,
      "q1_ms": 84.27698211744428,
      "q3_ms": 84.4989710021764
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0007406402728520334,
      "compile_s": 2.2741168818902224,
      "inner": 1,
      "median_ms": 188.29826195724308,
      "q1_ms": 186.73709593713284,
      "q3_ms": 198.6676650121808
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0007406402728520334,
      "compile_s": 0.926473347004503,
      "inner": 1,
      "median_ms": 74.48990689590573,
      "q1_ms": 74.33804706670344,
      "q3_ms": 74.63704608380795
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 32,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0007406402728520334,
      "compile_s": 2.169768812833354,
      "inner": 1,
      "median_ms": 79.51859384775162,
      "q1_ms": 79.45791399106383,
      "q3_ms": 79.56840307451785
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0007406402728520334,
      "compile_s": 0.7353655428159982,
      "inner": 1,
      "median_ms": 113.59975510276854,
      "q1_ms": 113.5915950872004,
      "q3_ms": 113.62860607914627
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0007406402728520334,
      "compile_s": 2.2030490338802338,
      "inner": 1,
      "median_ms": 99.68457999639213,
      "q1_ms": 99.28882122039795,
      "q3_ms": 100.75404611416161
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0007406402728520334,
      "compile_s": 0.7134674559347332,
      "inner": 1,
      "median_ms": 93.32987596280873,
      "q1_ms": 93.16140692681074,
      "q3_ms": 93.36084616370499
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0007406402728520334,
      "compile_s": 2.1990354149602354,
      "inner": 1,
      "median_ms": 88.40409899130464,
      "q1_ms": 88.35034887306392,
      "q3_ms": 88.51959882304072
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0007406402728520334,
      "compile_s": 0.9459988079033792,
      "inner": 1,
      "median_ms": 82.4051951058209,
      "q1_ms": 82.3608951177448,
      "q3_ms": 82.49035384505987
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0007406402728520334,
      "compile_s": 3.7400367881637067,
      "inner": 1,
      "median_ms": 125.50837313756347,
      "q1_ms": 125.15344307757914,
      "q3_ms": 125.57668099179864
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0007406402728520334,
      "compile_s": 1.425403225934133,
      "inner": 1,
      "median_ms": 120.09003711864352,
      "q1_ms": 119.7147078346461,
      "q3_ms": 120.89834385551512
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 64,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0007406402728520334,
      "compile_s": 2.206402495969087,
      "inner": 1,
      "median_ms": 110.11675209738314,
      "q1_ms": 109.94800180196762,
      "q3_ms": 110.129632987082
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 16,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0007406402728520334,
      "compile_s": 0.67650948301889,
      "inner": 1,
      "median_ms": 64.56307112239301,
      "q1_ms": 64.38977201469243,
      "q3_ms": 64.78866096585989
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 16,
      "num_stages": 2,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0007406402728520334,
      "compile_s": 2.1998823520261794,
      "inner": 1,
      "median_ms": 91.40074113383889,
      "q1_ms": 91.25829092226923,
      "q3_ms": 91.77245083265007
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 32,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0007406402728520334,
      "compile_s": 0.8426460090558976,
      "inner": 1,
      "median_ms": 68.14975198358297,
      "q1_ms": 68.10979195870459,
      "q3_ms": 68.78617987968028
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 32,
      "num_stages": 2,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 4,
      "rel_err": 0.0007406402728520334,
      "compile_s": 3.3493939889594913,
      "inner": 1,
      "median_ms": 85.41960688307881,
      "q1_ms": 84.91209801286459,
      "q3_ms": 85.52041603252292
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 64,
      "num_stages": 1,
      "num_warps": 8,
      "rel_err": 0.0007406402728520334,
      "compile_s": 1.660530841909349,
      "inner": 1,
      "median_ms": 69.59866802208126,
      "q1_ms": 69.41189803183079,
      "q3_ms": 69.8397580999881
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 64,
      "num_stages": 2,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 327680, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 128,
      "num_stages": 1,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 327680, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 4,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    },
    {
      "dtype": "float32",
      "head_dim": 128,
      "seq": 8192,
      "batch": 4,
      "heads": 8,
      "tag": "h100_nvl",
      "mode": "fwd_bwd",
      "config": "forced",
      "fwd_block_q": 128,
      "fwd_block_k": 128,
      "bwd_outer": 128,
      "bwd_inner": 128,
      "num_stages": 2,
      "num_warps": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Shared memory size limit exceeded: requested 262144, available: 232448, context: []"
    }
  ]
}
