{
  "device": {
    "jax": "0.11.2",
    "backend": "gpu",
    "devices": [
      "cuda:0"
    ],
    "python": "3.14.8",
    "device_kind": "NVIDIA GeForce RTX 3090",
    "gpu": "NVIDIA GeForce RTX 3090",
    "driver": "615.78.08",
    "gpu_memory_gb": 24.0,
    "torch": "2.14.1+cu130",
    "torch_cuda": "13.0"
  },
  "config": {
    "runner": "docs/scripts/benchmark_kernel_kmv.py",
    "note": "impl 'default@gpu-benchmarks' rows are the public default rbf_kernel_mv/rbf_kde_density on the base branch (before this lane): Pallas interpret mode on CPU, Pallas with block_o=64 on GPU.",
    "runs": {
      "gpu_a.json": {
        "device": "gpu",
        "output": "/tmp/claude-1003/-home-manuel-src/80785a0e-56be-4da7-a2a8-e1d6effd6199/scratchpad/kmv/gpu_a.json",
        "sections": [
          "grid"
        ],
        "sizes": [
          1024,
          4096,
          16384
        ],
        "dims": [
          2,
          8,
          32
        ],
        "outs": [
          1,
          16
        ],
        "modes": [
          "fwd",
          "fwd_bwd"
        ],
        "impls": [
          "dense",
          "xla",
          "pallas",
          "torch_dense"
        ],
        "repeats": 5,
        "tiles": [
          "64,64,64",
          "64,64,16",
          "128,64,16",
          "64,128,16",
          "128,128,16"
        ],
        "ladder": [
          16384,
          32768,
          49152,
          65536,
          98304,
          131072,
          262144
        ],
        "timeout": 300,
        "child": null
      },
      "gpu_b.json": {
        "device": "gpu",
        "output": "/tmp/claude-1003/-home-manuel-src/80785a0e-56be-4da7-a2a8-e1d6effd6199/scratchpad/kmv/gpu_b.json",
        "sections": [
          "grid"
        ],
        "sizes": [
          65536
        ],
        "dims": [
          2,
          8,
          32
        ],
        "outs": [
          1,
          16
        ],
        "modes": [
          "fwd",
          "fwd_bwd"
        ],
        "impls": [
          "dense",
          "xla",
          "pallas",
          "torch_dense"
        ],
        "repeats": 3,
        "tiles": [
          "64,64,64",
          "64,64,16",
          "128,64,16",
          "64,128,16",
          "128,128,16"
        ],
        "ladder": [
          16384,
          32768,
          49152,
          65536,
          98304,
          131072,
          262144
        ],
        "timeout": 300,
        "child": null
      },
      "gpu_kde.json": {
        "device": "gpu",
        "output": "/tmp/claude-1003/-home-manuel-src/80785a0e-56be-4da7-a2a8-e1d6effd6199/scratchpad/kmv/gpu_kde.json",
        "sections": [
          "kde"
        ],
        "sizes": [
          1024,
          4096,
          16384,
          65536
        ],
        "dims": [
          2,
          8,
          32
        ],
        "outs": [
          1,
          16
        ],
        "modes": [
          "fwd",
          "fwd_bwd"
        ],
        "impls": [
          "dense",
          "xla",
          "pallas"
        ],
        "repeats": 5,
        "tiles": [
          "64,64,64",
          "64,64,16",
          "128,64,16",
          "64,128,16",
          "128,128,16"
        ],
        "ladder": [
          16384,
          32768,
          49152,
          65536,
          98304,
          131072,
          262144
        ],
        "timeout": 300,
        "child": null
      },
      "gpu_peak.json": {
        "device": "gpu",
        "output": "/tmp/claude-1003/-home-manuel-src/80785a0e-56be-4da7-a2a8-e1d6effd6199/scratchpad/kmv/gpu_peak.json",
        "sections": [
          "peak"
        ],
        "sizes": [
          4096,
          16384,
          65536
        ],
        "dims": [
          2,
          8,
          32
        ],
        "outs": [
          1,
          16
        ],
        "modes": [
          "fwd",
          "fwd_bwd"
        ],
        "impls": [
          "dense",
          "xla",
          "pallas"
        ],
        "repeats": 5,
        "tiles": [
          "64,64,64",
          "64,64,16",
          "128,64,16",
          "64,128,16",
          "128,128,16"
        ],
        "ladder": [
          16384,
          32768,
          49152,
          65536,
          98304,
          131072,
          262144
        ],
        "timeout": 60,
        "child": null
      },
      "gpu_maxn.json": {
        "device": "gpu",
        "output": "/tmp/claude-1003/-home-manuel-src/80785a0e-56be-4da7-a2a8-e1d6effd6199/scratchpad/kmv/gpu_maxn.json",
        "sections": [
          "max_n"
        ],
        "sizes": [
          1024,
          4096,
          16384,
          65536
        ],
        "dims": [
          2,
          8,
          32
        ],
        "outs": [
          1,
          16
        ],
        "modes": [
          "fwd",
          "fwd_bwd"
        ],
        "impls": [
          "dense",
          "xla",
          "pallas"
        ],
        "repeats": 5,
        "tiles": [
          "64,64,64",
          "64,64,16",
          "128,64,16",
          "64,128,16",
          "128,128,16"
        ],
        "ladder": [
          16384,
          32768,
          49152,
          65536,
          131072,
          262144,
          524288,
          1048576
        ],
        "timeout": 90,
        "child": null
      },
      "tiles1.json": {
        "device": "gpu",
        "output": "/tmp/claude-1003/-home-manuel-src/80785a0e-56be-4da7-a2a8-e1d6effd6199/scratchpad/kmv/tiles1.json",
        "sections": [
          "tiles"
        ],
        "sizes": [
          4096,
          65536
        ],
        "dims": [
          8
        ],
        "outs": [
          1,
          16
        ],
        "modes": [
          "fwd",
          "fwd_bwd"
        ],
        "impls": [
          "dense",
          "xla",
          "pallas"
        ],
        "repeats": 3,
        "tiles": [
          "64,64,64",
          "64,64,16",
          "128,64,16",
          "64,128,16",
          "128,128,16",
          "32,64,16"
        ],
        "ladder": [
          16384,
          32768,
          49152,
          65536,
          98304,
          131072,
          262144
        ],
        "timeout": 300,
        "child": null
      },
      "gpu_cross.json": {
        "script": "crossover probe, 3 repeats x 5 samples"
      },
      "before_gpu.json": {
        "device": "gpu",
        "output": "/tmp/claude-1003/-home-manuel-src/80785a0e-56be-4da7-a2a8-e1d6effd6199/scratchpad/kmv/before_gpu.json",
        "sections": [
          "grid",
          "kde"
        ],
        "sizes": [
          1024,
          4096,
          16384,
          65536
        ],
        "dims": [
          8,
          32
        ],
        "outs": [
          1,
          16
        ],
        "modes": [
          "fwd",
          "fwd_bwd"
        ],
        "impls": [
          "default"
        ],
        "repeats": 5,
        "tiles": [
          "64,64,64",
          "64,64,16",
          "128,64,16",
          "64,128,16",
          "128,128,16"
        ],
        "ladder": [
          16384,
          32768,
          49152,
          65536,
          98304,
          131072,
          262144
        ],
        "timeout": 300,
        "child": null
      }
    }
  },
  "results": [
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 1,
      "xla_temp_mb": 4.0078125,
      "xla_arg_mb": 0.019535064697265625,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.24368860600225162,
      "inner": 200,
      "median_ms": 0.04969269000866916,
      "q1_ms": 0.049661084995022975,
      "q3_ms": 0.04989302498870529,
      "rel_err": 0.00010546792541596541
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 1,
      "xla_temp_mb": 4.0078125,
      "xla_arg_mb": 0.019535064697265625,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.11108206500648521,
      "inner": 200,
      "median_ms": 0.0504195200119284,
      "q1_ms": 0.04985830997611629,
      "q3_ms": 0.05046030000812607,
      "rel_err": 0.00010546792541596541
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 1,
      "xla_temp_mb": 0.125,
      "xla_arg_mb": 0.019535064697265625,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.15792199100542348,
      "inner": 200,
      "median_ms": 0.05289515498589026,
      "q1_ms": 0.05280332999973325,
      "q3_ms": 0.05318630999681773,
      "rel_err": 0.00018006080068534358
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 1,
      "compile_s": 0.11819429400202353,
      "inner": 146,
      "median_ms": 0.167602404112025,
      "q1_ms": 0.16712670549168251,
      "q3_ms": 0.17291436303560048,
      "peak_mb": 20.1484375,
      "rel_err": 1.407450305762738e-07
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 1,
      "xla_temp_mb": 12.007843017578125,
      "xla_arg_mb": 0.019535064697265625,
      "xla_out_mb": 0.019565582275390625,
      "compile_s": 0.5147947999939788,
      "inner": 200,
      "median_ms": 0.0647210899842321,
      "q1_ms": 0.06466152000939474,
      "q3_ms": 0.0647528449917445,
      "rel_err": {
        "dq": 0.00038299019758912615,
        "dk": 7.282564191425377e-05,
        "dv": 5.110109886884081e-05,
        "dlengthscale": 5.411950001908059e-06
      }
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 1,
      "xla_temp_mb": 12.007843017578125,
      "xla_arg_mb": 0.019535064697265625,
      "xla_out_mb": 0.019565582275390625,
      "compile_s": 0.5330670710027334,
      "inner": 200,
      "median_ms": 0.06638178001594497,
      "q1_ms": 0.06638012502662605,
      "q3_ms": 0.06657594502030406,
      "rel_err": {
        "dq": 0.0002882644987763921,
        "dk": 7.175986028063647e-05,
        "dv": 5.110109886884081e-05,
        "dlengthscale": 5.411950001908059e-06
      }
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 1,
      "xla_temp_mb": 0.1562652587890625,
      "xla_arg_mb": 0.019535064697265625,
      "xla_out_mb": 0.019565582275390625,
      "compile_s": 0.3573156979982741,
      "inner": 200,
      "median_ms": 0.1138889350113459,
      "q1_ms": 0.11384835997887421,
      "q3_ms": 0.11393522498110542,
      "rel_err": {
        "dq": 0.00046643770530988617,
        "dk": 0.00023386403098183976,
        "dv": 5.6173134713247345e-05,
        "dlengthscale": 0.00018705919493774523
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 1,
      "compile_s": 1.2950596989976475,
      "inner": 39,
      "median_ms": 0.7382795128460902,
      "q1_ms": 0.7376992050409078,
      "q3_ms": 0.7456085127509701,
      "peak_mb": 36.35693359375,
      "rel_err": {
        "dq": 6.490959690663732e-07,
        "dk": 2.2361358143113058e-07,
        "dv": 1.1897115652047611e-07
      }
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 16,
      "xla_temp_mb": 4.25,
      "xla_arg_mb": 0.07812881469726562,
      "xla_out_mb": 0.0625,
      "compile_s": 0.45730775500123855,
      "inner": 200,
      "median_ms": 0.053329280017351266,
      "q1_ms": 0.05320989999745507,
      "q3_ms": 0.05771552499936661,
      "rel_err": 0.0003047168515821507
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 16,
      "xla_temp_mb": 4.25,
      "xla_arg_mb": 0.07812881469726562,
      "xla_out_mb": 0.0625,
      "compile_s": 0.1367638489973615,
      "inner": 200,
      "median_ms": 0.05251553997368319,
      "q1_ms": 0.052363100003276486,
      "q3_ms": 0.05254243998933816,
      "rel_err": 0.0003047168515821507
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 16,
      "xla_temp_mb": 0.0625,
      "xla_arg_mb": 0.07812881469726562,
      "xla_out_mb": 0.0625,
      "compile_s": 0.17532535699865548,
      "inner": 200,
      "median_ms": 0.054632094979751855,
      "q1_ms": 0.05038545499701286,
      "q3_ms": 0.054682035006408114,
      "rel_err": 0.000304693496438327
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 16,
      "compile_s": 0.0005376770059228875,
      "inner": 200,
      "median_ms": 0.17152837001049193,
      "q1_ms": 0.1712756450069719,
      "q3_ms": 0.17174978500406723,
      "peak_mb": 28.390625,
      "rel_err": 3.1609323777099566e-07
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 16,
      "xla_temp_mb": 12.007835388183594,
      "xla_arg_mb": 0.07812881469726562,
      "xla_out_mb": 0.07815933227539062,
      "compile_s": 0.5959421640000073,
      "inner": 200,
      "median_ms": 0.08165763498254819,
      "q1_ms": 0.0816189699980896,
      "q3_ms": 0.08168308999302099,
      "rel_err": {
        "dq": 0.00036477505488051383,
        "dk": 0.00028330977911970265,
        "dv": 0.0001935070964281585,
        "dlengthscale": 0.00017721118728083196
      }
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 16,
      "xla_temp_mb": 16.007835388183594,
      "xla_arg_mb": 0.07812881469726562,
      "xla_out_mb": 0.07815933227539062,
      "compile_s": 0.5976646940034698,
      "inner": 200,
      "median_ms": 0.08562941002310254,
      "q1_ms": 0.0856136250149575,
      "q3_ms": 0.08569067002099473,
      "rel_err": {
        "dq": 0.00034290289247295135,
        "dk": 0.00028300117609058873,
        "dv": 0.0001935070964281585,
        "dlengthscale": 0.00017721118728083196
      }
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 16,
      "xla_temp_mb": 0.1252593994140625,
      "xla_arg_mb": 0.07812881469726562,
      "xla_out_mb": 0.07815933227539062,
      "compile_s": 0.34933144700335106,
      "inner": 200,
      "median_ms": 0.11458289998699911,
      "q1_ms": 0.11454603500169469,
      "q3_ms": 0.11463680501037743,
      "rel_err": {
        "dq": 0.00034499740101650287,
        "dk": 0.00028313114077020885,
        "dv": 0.00019692974126520448,
        "dlengthscale": 0.00017887941411873454
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 16,
      "compile_s": 0.001492014001996722,
      "inner": 61,
      "median_ms": 0.6938643114675661,
      "q1_ms": 0.6923097212985921,
      "q3_ms": 0.7025497541123558,
      "peak_mb": 36.64990234375,
      "rel_err": {
        "dq": 4.73421331444038e-07,
        "dk": 2.3010488703443342e-07,
        "dv": 1.707191464872882e-07
      }
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 4.0078125,
      "xla_arg_mb": 0.06641006469726562,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.22640015800425317,
      "inner": 200,
      "median_ms": 0.04592987501382595,
      "q1_ms": 0.04519684000115376,
      "q3_ms": 0.045941144999233074,
      "rel_err": 0.00027073936874651096
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 4.0078125,
      "xla_arg_mb": 0.06641006469726562,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.10700085199641762,
      "inner": 200,
      "median_ms": 0.04502946998400148,
      "q1_ms": 0.044853139988845214,
      "q3_ms": 0.051654505004989915,
      "rel_err": 0.00027073936874651096
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 0.0625,
      "xla_arg_mb": 0.06641006469726562,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.15256412100279704,
      "inner": 200,
      "median_ms": 0.05778601502242964,
      "q1_ms": 0.04969409499608446,
      "q3_ms": 0.05912805499974638,
      "rel_err": 0.0006879182731054002
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "compile_s": 0.000531004996446427,
      "inner": 200,
      "median_ms": 0.17836265498772264,
      "q1_ms": 0.1760753249982372,
      "q3_ms": 0.18004900000960333,
      "peak_mb": 28.3203125,
      "rel_err": 4.455237833836346e-07
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 12.007843017578125,
      "xla_arg_mb": 0.06641006469726562,
      "xla_out_mb": 0.06644058227539062,
      "compile_s": 0.47639038899797015,
      "inner": 200,
      "median_ms": 0.06187549999594921,
      "q1_ms": 0.061371900010271936,
      "q3_ms": 0.061921790002088535,
      "rel_err": {
        "dq": 0.00037866784446584904,
        "dk": 0.00014223582142673747,
        "dv": 9.811054505965919e-05,
        "dlengthscale": 2.1462438704709186e-05
      }
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 12.007843017578125,
      "xla_arg_mb": 0.06641006469726562,
      "xla_out_mb": 0.06644058227539062,
      "compile_s": 0.512858406000305,
      "inner": 200,
      "median_ms": 0.06208930499269627,
      "q1_ms": 0.06157092499051941,
      "q3_ms": 0.0622236600247561,
      "rel_err": {
        "dq": 0.00037867900094720286,
        "dk": 0.00014224008201984813,
        "dv": 9.811054505965919e-05,
        "dlengthscale": 2.1462438704709186e-05
      }
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 0.0625152587890625,
      "xla_arg_mb": 0.06641006469726562,
      "xla_out_mb": 0.06644058227539062,
      "compile_s": 0.3299631730042165,
      "inner": 200,
      "median_ms": 0.10975386503560003,
      "q1_ms": 0.10975090503052343,
      "q3_ms": 0.10981087001709966,
      "rel_err": {
        "dq": 0.0008397870112002144,
        "dk": 0.0009578645329394173,
        "dv": 0.0014792799682467413,
        "dlengthscale": 0.004471605144141046
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "compile_s": 0.001661704998696223,
      "inner": 51,
      "median_ms": 0.7674695098459922,
      "q1_ms": 0.7614581372050166,
      "q3_ms": 0.8012475686971846,
      "peak_mb": 36.59130859375,
      "rel_err": {
        "dq": 5.857374636460643e-07,
        "dk": 4.84080138723737e-07,
        "dv": 8.413026568281148e-07
      }
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 4.25,
      "xla_arg_mb": 0.12500381469726562,
      "xla_out_mb": 0.0625,
      "compile_s": 0.13218247899931157,
      "inner": 200,
      "median_ms": 0.04472298998734914,
      "q1_ms": 0.04455752499779919,
      "q3_ms": 0.04485648998524994,
      "rel_err": 0.00031239779468764984
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 4.25,
      "xla_arg_mb": 0.12500381469726562,
      "xla_out_mb": 0.0625,
      "compile_s": 0.13320192800165387,
      "inner": 200,
      "median_ms": 0.044603164969885256,
      "q1_ms": 0.04453639001440024,
      "q3_ms": 0.04479261999222217,
      "rel_err": 0.00031239779468764984
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 0.0,
      "xla_arg_mb": 0.12500381469726562,
      "xla_out_mb": 0.0625,
      "compile_s": 0.16848046299855923,
      "inner": 200,
      "median_ms": 0.034637954995559994,
      "q1_ms": 0.03420298500714125,
      "q3_ms": 0.03571204000763828,
      "rel_err": 0.00031191953973564866
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "compile_s": 0.0005251340044196695,
      "inner": 200,
      "median_ms": 0.17134196998085827,
      "q1_ms": 0.17115401500632288,
      "q3_ms": 0.17156684498331742,
      "peak_mb": 28.4375,
      "rel_err": 3.4154625219882097e-07
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 12.007835388183594,
      "xla_arg_mb": 0.12500381469726562,
      "xla_out_mb": 0.12503433227539062,
      "compile_s": 0.1840185789988027,
      "inner": 200,
      "median_ms": 0.07909659500000998,
      "q1_ms": 0.07900481999968179,
      "q3_ms": 0.07911177002824843,
      "rel_err": {
        "dq": 0.0002579291295952936,
        "dk": 0.00029289060848415057,
        "dv": 0.00018415382458524364,
        "dlengthscale": 0.0001370249375041204
      }
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 16.007835388183594,
      "xla_arg_mb": 0.12500381469726562,
      "xla_out_mb": 0.12503433227539062,
      "compile_s": 0.24136666100093862,
      "inner": 200,
      "median_ms": 0.08227951497246977,
      "q1_ms": 0.08226167999964673,
      "q3_ms": 0.0823059650065261,
      "rel_err": {
        "dq": 0.0002578961144852249,
        "dk": 0.0002928877570315094,
        "dv": 0.00018415382458524364,
        "dlengthscale": 0.00013693089156211345
      }
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 0.0625152587890625,
      "xla_arg_mb": 0.12500381469726562,
      "xla_out_mb": 0.12503433227539062,
      "compile_s": 0.34545547199377324,
      "inner": 200,
      "median_ms": 0.11515849000716116,
      "q1_ms": 0.11507353003253229,
      "q3_ms": 0.11518238501594169,
      "rel_err": {
        "dq": 0.0002565191564148692,
        "dk": 0.0002922680382897579,
        "dv": 0.00017977281646229062,
        "dlengthscale": 0.00013570829431602314
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "compile_s": 0.0014424299952224828,
      "inner": 61,
      "median_ms": 0.6776429507852777,
      "q1_ms": 0.6769986065308808,
      "q3_ms": 0.6790570819322935,
      "peak_mb": 36.88427734375,
      "rel_err": {
        "dq": 2.4814531001026894e-07,
        "dk": 1.301716247437023e-07,
        "dv": 2.1935618727495032e-07
      }
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 4.0078125,
      "xla_arg_mb": 0.2539100646972656,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.26339853800163837,
      "inner": 200,
      "median_ms": 0.04877485498582246,
      "q1_ms": 0.048402699976577424,
      "q3_ms": 0.05073300999356434,
      "rel_err": 6.22279183120454e-05
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 4.0078125,
      "xla_arg_mb": 0.2539100646972656,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.12084883900388377,
      "inner": 200,
      "median_ms": 0.051929679975728504,
      "q1_ms": 0.05189461498957826,
      "q3_ms": 0.05226095501711825,
      "rel_err": 6.22279183120454e-05
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 0.0625,
      "xla_arg_mb": 0.2539100646972656,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.19224621000466868,
      "inner": 200,
      "median_ms": 0.05353120497602504,
      "q1_ms": 0.053347415014286526,
      "q3_ms": 0.05358170499675907,
      "rel_err": 0.0004708603397108438
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 1,
      "compile_s": 0.0005186310008866712,
      "inner": 200,
      "median_ms": 0.16917306500545237,
      "q1_ms": 0.16861390500707785,
      "q3_ms": 0.1697448000049917,
      "peak_mb": 28.5078125,
      "rel_err": 2.477806228288743e-07
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 12.007843017578125,
      "xla_arg_mb": 0.2539100646972656,
      "xla_out_mb": 0.2539405822753906,
      "compile_s": 0.6021091049988172,
      "inner": 200,
      "median_ms": 0.06753371002560016,
      "q1_ms": 0.06743401998392073,
      "q3_ms": 0.06767322000087006,
      "rel_err": {
        "dq": 0.00026755608437059223,
        "dk": 1.4478228058422723e-05,
        "dv": 3.0945707775485552e-06,
        "dlengthscale": 5.963442072525116e-08
      }
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 12.007843017578125,
      "xla_arg_mb": 0.2539100646972656,
      "xla_out_mb": 0.2539405822753906,
      "compile_s": 0.7534952060013893,
      "inner": 200,
      "median_ms": 0.06635918500251137,
      "q1_ms": 0.06622111999604385,
      "q3_ms": 0.06652274500083877,
      "rel_err": {
        "dq": 0.00026758657760623313,
        "dk": 1.4478840183769792e-05,
        "dv": 3.0945707775485552e-06,
        "dlengthscale": 5.963442072525116e-08
      }
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 0.0625152587890625,
      "xla_arg_mb": 0.2539100646972656,
      "xla_out_mb": 0.2539405822753906,
      "compile_s": 0.4480686700044316,
      "inner": 139,
      "median_ms": 0.20134097120972524,
      "q1_ms": 0.20131805037289,
      "q3_ms": 0.20135077700209836,
      "rel_err": {
        "dq": 0.0007377436401438872,
        "dk": 0.00045520296497365777,
        "dv": 0.0003917558290979359,
        "dlengthscale": 0.0007655270588500492
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 1,
      "compile_s": 0.001587946004292462,
      "inner": 49,
      "median_ms": 0.7543359795224149,
      "q1_ms": 0.7524057754733581,
      "q3_ms": 0.7563552856908597,
      "peak_mb": 37.52880859375,
      "rel_err": {
        "dq": 4.0988859049375434e-07,
        "dk": 1.3257902792049906e-07,
        "dv": 2.209103639685566e-07
      }
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 4.25,
      "xla_arg_mb": 0.3125038146972656,
      "xla_out_mb": 0.0625,
      "compile_s": 0.13909817699459381,
      "inner": 200,
      "median_ms": 0.04838416502025211,
      "q1_ms": 0.048339435015805066,
      "q3_ms": 0.048439220008731354,
      "rel_err": 0.0002882423154052157
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 4.25,
      "xla_arg_mb": 0.3125038146972656,
      "xla_out_mb": 0.0625,
      "compile_s": 0.13172263799788198,
      "inner": 200,
      "median_ms": 0.049209829994651955,
      "q1_ms": 0.04909095499897376,
      "q3_ms": 0.04934894499456277,
      "rel_err": 0.0002882423154052157
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 0.0,
      "xla_arg_mb": 0.3125038146972656,
      "xla_out_mb": 0.0625,
      "compile_s": 0.21303584500128636,
      "inner": 200,
      "median_ms": 0.0527777299794252,
      "q1_ms": 0.05276345502352342,
      "q3_ms": 0.052786399974138476,
      "rel_err": 0.0002879177244021982
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "compile_s": 0.0005284399958327413,
      "inner": 200,
      "median_ms": 0.17216161999385804,
      "q1_ms": 0.17071007001504768,
      "q3_ms": 0.17217859996890184,
      "peak_mb": 28.625,
      "rel_err": 3.9624942141264373e-07
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 12.007835388183594,
      "xla_arg_mb": 0.3125038146972656,
      "xla_out_mb": 0.3125343322753906,
      "compile_s": 0.17814197699772194,
      "inner": 200,
      "median_ms": 0.07768712501274422,
      "q1_ms": 0.07767801002046326,
      "q3_ms": 0.07769960000587162,
      "rel_err": {
        "dq": 0.00018788653440159368,
        "dk": 0.00027352753441305556,
        "dv": 0.00018134856306345444,
        "dlengthscale": 5.840121219929934e-05
      }
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 16.007835388183594,
      "xla_arg_mb": 0.3125038146972656,
      "xla_out_mb": 0.3125343322753906,
      "compile_s": 0.23728502700396348,
      "inner": 200,
      "median_ms": 0.07932782998977927,
      "q1_ms": 0.07932437500130618,
      "q3_ms": 0.07939670998894144,
      "rel_err": {
        "dq": 0.00018790962245289608,
        "dk": 0.00027352675253121433,
        "dv": 0.00018134856306345444,
        "dlengthscale": 5.840121219929934e-05
      }
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 0.0625152587890625,
      "xla_arg_mb": 0.3125038146972656,
      "xla_out_mb": 0.3125343322753906,
      "compile_s": 0.4370308599973214,
      "inner": 159,
      "median_ms": 0.21627874211730355,
      "q1_ms": 0.21607263524227324,
      "q3_ms": 0.21633148429850876,
      "rel_err": {
        "dq": 0.00018724864253970186,
        "dk": 0.00027276368815804935,
        "dv": 0.00017874289566915128,
        "dlengthscale": 5.659498914158905e-05
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "compile_s": 0.0014186039988999255,
      "inner": 55,
      "median_ms": 0.6772759453848597,
      "q1_ms": 0.6743571455114182,
      "q3_ms": 0.6798535455759106,
      "peak_mb": 37.82177734375,
      "rel_err": {
        "dq": 2.412307819408992e-07,
        "dk": 1.3190731873246592e-07,
        "dv": 2.5528436328382046e-07
      }
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "out": 1,
      "xla_temp_mb": 64.01562881469727,
      "xla_arg_mb": 0.07812881469726562,
      "xla_out_mb": 0.015625,
      "compile_s": 0.5853647059993818,
      "inner": 112,
      "median_ms": 0.3291486338769443,
      "q1_ms": 0.3290867321733718,
      "q3_ms": 0.32920776785821154,
      "rel_err": 0.0002590105807937189
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "out": 1,
      "xla_temp_mb": 16.024703979492188,
      "xla_arg_mb": 0.07812881469726562,
      "xla_out_mb": 0.015625,
      "compile_s": 0.2867038040058105,
      "inner": 158,
      "median_ms": 0.21025436710053463,
      "q1_ms": 0.21017117090000112,
      "q3_ms": 0.21027560760097613,
      "rel_err": 0.00018502702212646458
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "out": 1,
      "xla_temp_mb": 0.5,
      "xla_arg_mb": 0.07812881469726562,
      "xla_out_mb": 0.015625,
      "compile_s": 0.15515517400490353,
      "inner": 200,
      "median_ms": 0.08094243497907883,
      "q1_ms": 0.08093887998256832,
      "q3_ms": 0.0809632749951561,
      "rel_err": 0.0002590105807937189
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "out": 1,
      "compile_s": 0.001639353002246935,
      "inner": 39,
      "median_ms": 1.1456797435993138,
      "q1_ms": 1.1456011539289297,
      "q3_ms": 1.145966718001709,
      "peak_mb": 208.34375,
      "rel_err": 2.929174054718256e-07
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "out": 1,
      "xla_temp_mb": 128.0468978881836,
      "xla_arg_mb": 0.07812881469726562,
      "xla_out_mb": 0.07815933227539062,
      "compile_s": 0.9654937090017484,
      "inner": 55,
      "median_ms": 0.8094110909785905,
      "q1_ms": 0.8093759454575113,
      "q3_ms": 0.8096189453763972,
      "rel_err": {
        "dq": 0.0004632229829970414,
        "dk": 8.415851637510185e-05,
        "dv": 7.120975872089204e-05,
        "dlengthscale": 0.00032343226761799587
      }
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "out": 1,
      "xla_temp_mb": 32.05647277832031,
      "xla_arg_mb": 0.07812881469726562,
      "xla_out_mb": 0.07815933227539062,
      "compile_s": 0.6019585800022469,
      "inner": 45,
      "median_ms": 1.0156726222097254,
      "q1_ms": 1.015445977746923,
      "q3_ms": 1.0166794221731834,
      "rel_err": {
        "dq": 0.00036946381862851957,
        "dk": 7.884354330673672e-05,
        "dv": 8.136063343341971e-05,
        "dlengthscale": 0.0002305298404322348
      }
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "out": 1,
      "xla_temp_mb": 0.6250152587890625,
      "xla_arg_mb": 0.07812881469726562,
      "xla_out_mb": 0.07815933227539062,
      "compile_s": 0.36026637200120604,
      "inner": 101,
      "median_ms": 0.3555202574266287,
      "q1_ms": 0.3551894356266107,
      "q3_ms": 0.35568502969926885,
      "rel_err": {
        "dq": 0.0004859052557212752,
        "dk": 0.00021814586607706596,
        "dv": 7.120975872089204e-05,
        "dlengthscale": 0.00035685830983179167
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "out": 1,
      "compile_s": 0.003922413998225238,
      "inner": 14,
      "median_ms": 3.357360357248191,
      "q1_ms": 3.3567835712996645,
      "q3_ms": 3.3585904285635997,
      "peak_mb": 336.5166015625,
      "rel_err": {
        "dq": 6.562110428728143e-07,
        "dk": 1.6535156436359836e-07,
        "dv": 9.129716957741982e-08
      }
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "out": 16,
      "xla_temp_mb": 65.0,
      "xla_arg_mb": 0.3125038146972656,
      "xla_out_mb": 0.25,
      "compile_s": 0.5093320320011117,
      "inner": 116,
      "median_ms": 0.33149106030609327,
      "q1_ms": 0.33089881899600027,
      "q3_ms": 0.331740155161242,
      "rel_err": 0.00030104418666990825
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "out": 16,
      "xla_temp_mb": 16.500534057617188,
      "xla_arg_mb": 0.3125038146972656,
      "xla_out_mb": 0.25,
      "compile_s": 0.4821885019991896,
      "inner": 102,
      "median_ms": 0.3710868039371574,
      "q1_ms": 0.37059990198958626,
      "q3_ms": 0.3711260000201554,
      "rel_err": 0.0003007163309717138
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "out": 16,
      "xla_temp_mb": 0.25,
      "xla_arg_mb": 0.3125038146972656,
      "xla_out_mb": 0.25,
      "compile_s": 0.17393330299819354,
      "inner": 200,
      "median_ms": 0.07844315499824006,
      "q1_ms": 0.07844054998713546,
      "q3_ms": 0.0784465650212951,
      "rel_err": 0.0003014644629170566
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "out": 16,
      "compile_s": 0.0017047769943019375,
      "inner": 39,
      "median_ms": 1.1739982308846755,
      "q1_ms": 1.1739833332681788,
      "q3_ms": 1.174612717967158,
      "peak_mb": 208.8125,
      "rel_err": 4.714485867469372e-07
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "out": 16,
      "xla_temp_mb": 192.25025939941406,
      "xla_arg_mb": 0.3125038146972656,
      "xla_out_mb": 0.3125343322753906,
      "compile_s": 0.672171588994388,
      "inner": 32,
      "median_ms": 1.4208915936251287,
      "q1_ms": 1.4208846562269173,
      "q3_ms": 1.4211216873718513,
      "rel_err": {
        "dq": 0.0003428601598039181,
        "dk": 0.0002822232345377269,
        "dv": 0.00017200964034192712,
        "dlengthscale": 3.314382635972548e-05
      }
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "out": 16,
      "xla_temp_mb": 48.35334777832031,
      "xla_arg_mb": 0.3125038146972656,
      "xla_out_mb": 0.3125343322753906,
      "compile_s": 0.7319968899973901,
      "inner": 32,
      "median_ms": 1.4340120312681393,
      "q1_ms": 1.4334403124394157,
      "q3_ms": 1.4351209999858838,
      "rel_err": {
        "dq": 0.0003152936113569983,
        "dk": 0.0002821563816181716,
        "dv": 0.0001714735929637892,
        "dlengthscale": 3.511391394054832e-05
      }
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "out": 16,
      "xla_temp_mb": 0.5002593994140625,
      "xla_arg_mb": 0.3125038146972656,
      "xla_out_mb": 0.3125343322753906,
      "compile_s": 0.3507348440034548,
      "inner": 93,
      "median_ms": 0.393568172062602,
      "q1_ms": 0.3935635268722,
      "q3_ms": 0.393578193559792,
      "rel_err": {
        "dq": 0.0003151890730058485,
        "dk": 0.00028238571941706196,
        "dv": 0.00018819956854199238,
        "dlengthscale": 2.5958801064959813e-05
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "out": 16,
      "compile_s": 0.0038986479994491674,
      "inner": 14,
      "median_ms": 3.4101974285606826,
      "q1_ms": 3.4095569286104626,
      "q3_ms": 3.4105037858742953,
      "peak_mb": 337.6884765625,
      "rel_err": {
        "dq": 5.613641077326309e-07,
        "dk": 1.6656831379630699e-07,
        "dv": 2.7087049400516964e-07
      }
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 64.01562881469727,
      "xla_arg_mb": 0.2656288146972656,
      "xla_out_mb": 0.015625,
      "compile_s": 0.37001605400291737,
      "inner": 110,
      "median_ms": 0.32874079095728864,
      "q1_ms": 0.3286195545198015,
      "q3_ms": 0.3287683000301265,
      "rel_err": 0.00037412705947097294
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 16.036178588867188,
      "xla_arg_mb": 0.2656288146972656,
      "xla_out_mb": 0.015625,
      "compile_s": 0.2838426590024028,
      "inner": 161,
      "median_ms": 0.2098205031249354,
      "q1_ms": 0.20968465837512618,
      "q3_ms": 0.20982672668984795,
      "rel_err": 0.00014036383379156865
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 0.25,
      "xla_arg_mb": 0.2656288146972656,
      "xla_out_mb": 0.015625,
      "compile_s": 0.15874849400279345,
      "inner": 200,
      "median_ms": 0.07978474503033794,
      "q1_ms": 0.07976390497788088,
      "q3_ms": 0.07978614998137346,
      "rel_err": 0.00037412705947097294
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 1,
      "compile_s": 0.0016949170021689497,
      "inner": 40,
      "median_ms": 1.1519601250256528,
      "q1_ms": 1.1518536499352194,
      "q3_ms": 1.151972125080647,
      "peak_mb": 208.53125,
      "rel_err": 5.050403268278143e-07
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 128.0468978881836,
      "xla_arg_mb": 0.2656288146972656,
      "xla_out_mb": 0.2656593322753906,
      "compile_s": 0.6357380179979373,
      "inner": 55,
      "median_ms": 0.8089487817911009,
      "q1_ms": 0.808842563657725,
      "q3_ms": 0.8090136182462712,
      "rel_err": {
        "dq": 0.00046952302376242764,
        "dk": 5.7814652771456496e-05,
        "dv": 8.087016180027248e-05,
        "dlengthscale": 5.8788622395836936e-05
      }
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 32.07991027832031,
      "xla_arg_mb": 0.2656288146972656,
      "xla_out_mb": 0.2656593322753906,
      "compile_s": 0.5770079809954041,
      "inner": 45,
      "median_ms": 1.0206177555826594,
      "q1_ms": 1.0202000887754064,
      "q3_ms": 1.0208644443385615,
      "rel_err": {
        "dq": 0.0003001454424129967,
        "dk": 4.743566699939701e-05,
        "dv": 3.2824994429411154e-05,
        "dlengthscale": 2.7665234068629145e-06
      }
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 0.2500152587890625,
      "xla_arg_mb": 0.2656288146972656,
      "xla_out_mb": 0.2656593322753906,
      "compile_s": 0.3375155750036356,
      "inner": 107,
      "median_ms": 0.35182789717985874,
      "q1_ms": 0.35180486918200743,
      "q3_ms": 0.3518885794143567,
      "rel_err": {
        "dq": 0.0005296932922023401,
        "dk": 0.0002180159319189403,
        "dv": 8.087016180027248e-05,
        "dlengthscale": 3.172856532245905e-05
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 1,
      "compile_s": 0.003869152002152987,
      "inner": 14,
      "median_ms": 3.3535115716014325,
      "q1_ms": 3.352395857031557,
      "q3_ms": 3.3540691433050336,
      "peak_mb": 337.0791015625,
      "rel_err": {
        "dq": 7.19485544769799e-07,
        "dk": 1.8660833876187787e-07,
        "dv": 2.604557481643309e-07
      }
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 65.0,
      "xla_arg_mb": 0.5000038146972656,
      "xla_out_mb": 0.25,
      "compile_s": 0.1390860729952692,
      "inner": 116,
      "median_ms": 0.33006154312872615,
      "q1_ms": 0.33003520692200883,
      "q3_ms": 0.3300885775764794,
      "rel_err": 0.00028298069255160044
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 16.500534057617188,
      "xla_arg_mb": 0.5000038146972656,
      "xla_out_mb": 0.25,
      "compile_s": 0.14865596500021638,
      "inner": 106,
      "median_ms": 0.36951799998894547,
      "q1_ms": 0.3695151698473569,
      "q3_ms": 0.3695310377499379,
      "rel_err": 0.00028310534038354223
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 0.0,
      "xla_arg_mb": 0.5000038146972656,
      "xla_out_mb": 0.25,
      "compile_s": 0.17064767500414746,
      "inner": 200,
      "median_ms": 0.07456797000486404,
      "q1_ms": 0.07456496998202056,
      "q3_ms": 0.07457167499524076,
      "rel_err": 0.00028294419170473274
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 16,
      "compile_s": 0.0017115289956564084,
      "inner": 39,
      "median_ms": 1.1796173333632163,
      "q1_ms": 1.1792229999674675,
      "q3_ms": 1.1796234871442908,
      "peak_mb": 209.0,
      "rel_err": 5.085526881563552e-07
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 192.25025939941406,
      "xla_arg_mb": 0.5000038146972656,
      "xla_out_mb": 0.5000343322753906,
      "compile_s": 0.19816640400676988,
      "inner": 33,
      "median_ms": 1.4174740909683432,
      "q1_ms": 1.4170490606127053,
      "q3_ms": 1.4185248788139275,
      "rel_err": {
        "dq": 0.0001754205151439904,
        "dk": 0.0002829781629505229,
        "dv": 0.00018128340765101352,
        "dlengthscale": 1.6351136797921206e-05
      }
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 48.37678527832031,
      "xla_arg_mb": 0.5000038146972656,
      "xla_out_mb": 0.5000343322753906,
      "compile_s": 0.2990375449953717,
      "inner": 32,
      "median_ms": 1.4381106875589467,
      "q1_ms": 1.4380600000549748,
      "q3_ms": 1.438343374957185,
      "rel_err": {
        "dq": 0.00017545912925746792,
        "dk": 0.0002831511839715999,
        "dv": 0.0001815002818534178,
        "dlengthscale": 1.7168693637817268e-05
      }
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 0.2500152587890625,
      "xla_arg_mb": 0.5000038146972656,
      "xla_out_mb": 0.5000343322753906,
      "compile_s": 0.35309167399827857,
      "inner": 95,
      "median_ms": 0.3898884947901886,
      "q1_ms": 0.38963717892520916,
      "q3_ms": 0.3899075895069951,
      "rel_err": {
        "dq": 0.00017343429261538034,
        "dk": 0.00028272303624950624,
        "dv": 0.00018670455853559217,
        "dlengthscale": 8.770155191612283e-06
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 16,
      "compile_s": 0.0039121439986047335,
      "inner": 14,
      "median_ms": 3.409918285927103,
      "q1_ms": 3.409395928299221,
      "q3_ms": 3.4101158572801586,
      "peak_mb": 338.2509765625,
      "rel_err": {
        "dq": 2.7800179322721735e-07,
        "dk": 1.2350770025441923e-07,
        "dv": 3.0409490359345544e-07
      }
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 64.01562881469727,
      "xla_arg_mb": 1.0156288146972656,
      "xla_out_mb": 0.015625,
      "compile_s": 0.38819443700049305,
      "inner": 113,
      "median_ms": 0.333912999962098,
      "q1_ms": 0.33390261948858674,
      "q3_ms": 0.3340011327352325,
      "rel_err": 0.00044360173494841176
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 16.129928588867188,
      "xla_arg_mb": 1.0156288146972656,
      "xla_out_mb": 0.015625,
      "compile_s": 0.2991326349947485,
      "inner": 157,
      "median_ms": 0.21879457963918258,
      "q1_ms": 0.21857870063280307,
      "q3_ms": 0.21893988537882414,
      "rel_err": 8.428934282176707e-05
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 0.25,
      "xla_arg_mb": 1.0156288146972656,
      "xla_out_mb": 0.015625,
      "compile_s": 0.1933497889986029,
      "inner": 159,
      "median_ms": 0.1674013396006798,
      "q1_ms": 0.16736094970746646,
      "q3_ms": 0.16743637109431583,
      "rel_err": 0.00044360173494841176
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 32,
      "out": 1,
      "compile_s": 0.001732038006593939,
      "inner": 39,
      "median_ms": 1.1919015896870777,
      "q1_ms": 1.191312025557869,
      "q3_ms": 1.1920154102904819,
      "peak_mb": 209.28125,
      "rel_err": 6.568929175293356e-07
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 128.0468978881836,
      "xla_arg_mb": 1.0156288146972656,
      "xla_out_mb": 1.0156593322753906,
      "compile_s": 0.711977052000293,
      "inner": 54,
      "median_ms": 0.8266314444308928,
      "q1_ms": 0.8261527777735695,
      "q3_ms": 0.826768944399966,
      "rel_err": {
        "dq": 0.0005342567585036303,
        "dk": 8.609477089888771e-05,
        "dv": 0.000135451451765352,
        "dlengthscale": 8.026168392459138e-05
      }
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 32.17366027832031,
      "xla_arg_mb": 1.0156288146972656,
      "xla_out_mb": 1.0156593322753906,
      "compile_s": 0.7605559589937911,
      "inner": 43,
      "median_ms": 1.0648436047716154,
      "q1_ms": 1.06475879062579,
      "q3_ms": 1.0655165115590641,
      "rel_err": {
        "dq": 0.0002970492229162735,
        "dk": 1.3681849377617507e-05,
        "dv": 3.904596295451718e-06,
        "dlengthscale": 6.348381779346211e-07
      }
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 0.2500152587890625,
      "xla_arg_mb": 1.0156288146972656,
      "xla_out_mb": 1.0156593322753906,
      "compile_s": 0.45492951400228776,
      "inner": 60,
      "median_ms": 0.6726278000617943,
      "q1_ms": 0.6724791832918224,
      "q3_ms": 0.6727042667383406,
      "rel_err": {
        "dq": 0.0006182782429062878,
        "dk": 0.000227334093564408,
        "dv": 0.000135451451765352,
        "dlengthscale": 9.304913865156017e-05
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 32,
      "out": 1,
      "compile_s": 0.003844405997369904,
      "inner": 14,
      "median_ms": 3.409483928408008,
      "q1_ms": 3.40918549980935,
      "q3_ms": 3.4107850000769497,
      "peak_mb": 339.3291015625,
      "rel_err": {
        "dq": 1.0066622262580948e-06,
        "dk": 2.010426879739166e-07,
        "dv": 3.0253640038331234e-07
      }
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 65.0,
      "xla_arg_mb": 1.2500038146972656,
      "xla_out_mb": 0.25,
      "compile_s": 0.14308615399932023,
      "inner": 118,
      "median_ms": 0.3356571017366921,
      "q1_ms": 0.3354105339000807,
      "q3_ms": 0.33570633894345586,
      "rel_err": 0.000360801542626367
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 16.500534057617188,
      "xla_arg_mb": 1.2500038146972656,
      "xla_out_mb": 0.25,
      "compile_s": 0.1562415910011623,
      "inner": 102,
      "median_ms": 0.3808830784406632,
      "q1_ms": 0.38077169608418354,
      "q3_ms": 0.3809968235573786,
      "rel_err": 0.00036069859310041845
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 0.0,
      "xla_arg_mb": 1.2500038146972656,
      "xla_out_mb": 0.25,
      "compile_s": 0.22180950999609195,
      "inner": 155,
      "median_ms": 0.17562404513404134,
      "q1_ms": 0.1755881741935868,
      "q3_ms": 0.17563394193852983,
      "rel_err": 0.0003580547507577877
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 32,
      "out": 16,
      "compile_s": 0.0017273490011575632,
      "inner": 38,
      "median_ms": 1.2207321053306497,
      "q1_ms": 1.220460552614991,
      "q3_ms": 1.2215460000764946,
      "peak_mb": 209.75,
      "rel_err": 6.590295767594524e-07
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 192.03150939941406,
      "xla_arg_mb": 1.2500038146972656,
      "xla_out_mb": 1.2500343322753906,
      "compile_s": 0.20141381999565056,
      "inner": 32,
      "median_ms": 1.4343263749196922,
      "q1_ms": 1.4327646563288,
      "q3_ms": 1.434675156360754,
      "rel_err": {
        "dq": 0.00026256473930110543,
        "dk": 0.00031662674391290526,
        "dv": 0.00023579319231552333,
        "dlengthscale": 0.00014146333901057722
      }
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 48.47053527832031,
      "xla_arg_mb": 1.2500038146972656,
      "xla_out_mb": 1.2500343322753906,
      "compile_s": 0.30181492200063076,
      "inner": 32,
      "median_ms": 1.4831562812105403,
      "q1_ms": 1.4828456874056428,
      "q3_ms": 1.4833566563083878,
      "rel_err": {
        "dq": 0.0002631310277945088,
        "dk": 0.00031652934714532216,
        "dv": 0.00023559531523559241,
        "dlengthscale": 0.00014273296746650588
      }
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 0.2500152587890625,
      "xla_arg_mb": 1.2500038146972656,
      "xla_out_mb": 1.2500343322753906,
      "compile_s": 0.4376212380011566,
      "inner": 58,
      "median_ms": 0.7336158103436023,
      "q1_ms": 0.733544620743487,
      "q3_ms": 0.7336417068566742,
      "rel_err": {
        "dq": 0.00025811965440153834,
        "dk": 0.0003128882428661512,
        "dv": 0.00022581444962974112,
        "dlengthscale": 0.00013230864961782848
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 32,
      "out": 16,
      "compile_s": 0.003900531999533996,
      "inner": 14,
      "median_ms": 3.4681707859687907,
      "q1_ms": 3.4678880001592916,
      "q3_ms": 3.4683110716287047,
      "peak_mb": 340.5009765625,
      "rel_err": {
        "dq": 4.100003325554601e-07,
        "dk": 1.360501672913456e-07,
        "dv": 3.389066555343746e-07
      }
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 2,
      "out": 1,
      "xla_temp_mb": 2048.062759399414,
      "xla_arg_mb": 0.3125038146972656,
      "xla_out_mb": 0.0625,
      "compile_s": 0.7461940990033327,
      "inner": 9,
      "median_ms": 5.000456777149036,
      "q1_ms": 4.989806777707094,
      "q3_ms": 5.000822999540509,
      "rel_err": 0.0005553607623100564
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 2,
      "out": 1,
      "xla_temp_mb": 16.065719604492188,
      "xla_arg_mb": 0.3125038146972656,
      "xla_out_mb": 0.0625,
      "compile_s": 0.23775491600099485,
      "inner": 13,
      "median_ms": 3.604815384628287,
      "q1_ms": 3.602534769295464,
      "q3_ms": 3.606417538181771,
      "rel_err": 0.0004505296568916202
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 2,
      "out": 1,
      "xla_temp_mb": 2.0,
      "xla_arg_mb": 0.3125038146972656,
      "xla_out_mb": 0.0625,
      "compile_s": 0.1567179319972638,
      "inner": 66,
      "median_ms": 0.6846508787764998,
      "q1_ms": 0.6817342879058385,
      "q3_ms": 0.6847118938546346,
      "rel_err": 0.0006284109967216741
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 2,
      "out": 1,
      "compile_s": 0.019587139002396725,
      "inner": 2,
      "median_ms": 17.70817350188736,
      "q1_ms": 17.7034240005014,
      "q3_ms": 17.709179999656044,
      "peak_mb": 3088.625,
      "rel_err": 8.867788113758722e-07
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 2,
      "out": 1,
      "xla_temp_mb": 3072.1252670288086,
      "xla_arg_mb": 0.3125038146972656,
      "xla_out_mb": 0.3125343322753906,
      "compile_s": 1.2514062649934203,
      "inner": 3,
      "median_ms": 13.694152999960352,
      "q1_ms": 13.686288332489008,
      "q3_ms": 13.776969666650984,
      "rel_err": {
        "dq": 0.00112602327449691,
        "dk": 0.00039423905141291766,
        "dv": 0.00044845500124132134,
        "dlengthscale": 0.0005148451034368835
      }
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 2,
      "out": 1,
      "xla_temp_mb": 32.13850402832031,
      "xla_arg_mb": 0.3125038146972656,
      "xla_out_mb": 0.3125343322753906,
      "compile_s": 0.5719507800022257,
      "inner": 3,
      "median_ms": 16.73699799963894,
      "q1_ms": 16.71318400137049,
      "q3_ms": 16.743169665763464,
      "rel_err": {
        "dq": 0.0009727232675825654,
        "dk": 0.0002923790276807472,
        "dv": 0.00011949662597939464,
        "dlengthscale": 0.003230017554002867
      }
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 2,
      "out": 1,
      "xla_temp_mb": 2.5000152587890625,
      "xla_arg_mb": 0.3125038146972656,
      "xla_out_mb": 0.3125343322753906,
      "compile_s": 0.37144679199991515,
      "inner": 13,
      "median_ms": 3.664748692244757,
      "q1_ms": 3.651061230742086,
      "q3_ms": 3.666784077023084,
      "rel_err": {
        "dq": 0.0011505220492793423,
        "dk": 0.0004243705972941857,
        "dv": 0.0005028783183954194,
        "dlengthscale": 0.0018368188675733968
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 2,
      "out": 1,
      "compile_s": 0.05274188500334276,
      "inner": 1,
      "median_ms": 51.426254998659715,
      "q1_ms": 51.4194019997376,
      "q3_ms": 51.43223699997179,
      "peak_mb": 5137.3134765625,
      "rel_err": {
        "dq": 1.5979774503019445e-06,
        "dk": 1.8717990039782921e-07,
        "dv": 2.0954606582593787e-07
      }
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 2,
      "out": 16,
      "xla_temp_mb": 2048.062759399414,
      "xla_arg_mb": 1.2500038146972656,
      "xla_out_mb": 1.0,
      "compile_s": 0.6771714920032537,
      "inner": 9,
      "median_ms": 5.026307999893713,
      "q1_ms": 5.026017333471423,
      "q3_ms": 5.02674877740598,
      "rel_err": 0.0002817557364900757
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 2,
      "out": 16,
      "xla_temp_mb": 16.501022338867188,
      "xla_arg_mb": 1.2500038146972656,
      "xla_out_mb": 1.0,
      "compile_s": 0.4806384590046946,
      "inner": 8,
      "median_ms": 6.120865000411868,
      "q1_ms": 6.107764124863024,
      "q3_ms": 6.126142625362263,
      "rel_err": 0.00033392880401812393
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 2,
      "out": 16,
      "xla_temp_mb": 1.0,
      "xla_arg_mb": 1.2500038146972656,
      "xla_out_mb": 1.0,
      "compile_s": 0.17789053199521732,
      "inner": 59,
      "median_ms": 0.8015242033791965,
      "q1_ms": 0.794333135595711,
      "q3_ms": 0.806299000001765,
      "rel_err": 0.00034285109955084286
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 2,
      "out": 16,
      "compile_s": 0.0199456080008531,
      "inner": 2,
      "median_ms": 17.966301998967538,
      "q1_ms": 17.966251500183716,
      "q3_ms": 17.968690997804515,
      "peak_mb": 3090.5,
      "rel_err": 7.482365641932637e-07
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 2,
      "out": 16,
      "xla_temp_mb": 4096.125511169434,
      "xla_arg_mb": 1.2500038146972656,
      "xla_out_mb": 1.2500343322753906,
      "compile_s": 0.7376500489990576,
      "inner": 3,
      "median_ms": 16.18829033153209,
      "q1_ms": 16.187371332004357,
      "q3_ms": 16.189442333901145,
      "rel_err": {
        "dq": 0.00019364809559040326,
        "dk": 0.000213353883253073,
        "dv": 0.00023716034369457206,
        "dlengthscale": 7.931323426579474e-05
      }
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 2,
      "out": 16,
      "xla_temp_mb": 49.09211730957031,
      "xla_arg_mb": 1.2500038146972656,
      "xla_out_mb": 1.2500343322753906,
      "compile_s": 0.6283590640014154,
      "inner": 2,
      "median_ms": 23.39789049801766,
      "q1_ms": 23.397545497573446,
      "q3_ms": 23.40893149812473,
      "rel_err": {
        "dq": 0.00041392118247846185,
        "dk": 0.0003150854606536342,
        "dv": 0.0002226212085125437,
        "dlengthscale": 0.0001293465321494908
      }
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 2,
      "out": 16,
      "xla_temp_mb": 2.0009918212890625,
      "xla_arg_mb": 1.2500038146972656,
      "xla_out_mb": 1.2500343322753906,
      "compile_s": 0.3556595429981826,
      "inner": 11,
      "median_ms": 4.37395136395935,
      "q1_ms": 4.341887091207106,
      "q3_ms": 4.394467545683834,
      "rel_err": {
        "dq": 0.0004417165952309446,
        "dk": 0.0003271235963299346,
        "dv": 0.0003870729992756963,
        "dlengthscale": 0.0001640159073711516
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 2,
      "out": 16,
      "compile_s": 0.05238310599816032,
      "inner": 1,
      "median_ms": 51.825941998686176,
      "q1_ms": 51.76563799614087,
      "q3_ms": 51.89474199869437,
      "peak_mb": 5142.0009765625,
      "rel_err": {
        "dq": 6.47221110880174e-07,
        "dk": 1.9274627347923984e-07,
        "dv": 2.914659586136965e-07
      }
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 2048.062759399414,
      "xla_arg_mb": 1.0625038146972656,
      "xla_out_mb": 0.0625,
      "compile_s": 0.44021937499928754,
      "inner": 9,
      "median_ms": 4.9951677776536805,
      "q1_ms": 4.991694555630804,
      "q3_ms": 5.002600000201103,
      "rel_err": 0.0006915842610169338
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 16.065719604492188,
      "xla_arg_mb": 1.0625038146972656,
      "xla_out_mb": 0.0625,
      "compile_s": 0.2421629189993837,
      "inner": 13,
      "median_ms": 3.6409891539593584,
      "q1_ms": 3.6384574615602525,
      "q3_ms": 3.6456941536296372,
      "rel_err": 0.0003344410330196829
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 1.0,
      "xla_arg_mb": 1.0625038146972656,
      "xla_out_mb": 0.0625,
      "compile_s": 0.15034729500621324,
      "inner": 62,
      "median_ms": 0.6898090323375072,
      "q1_ms": 0.6897791290905462,
      "q3_ms": 0.6898416612520739,
      "rel_err": 0.000834137764951186
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 8,
      "out": 1,
      "compile_s": 0.019731011001567822,
      "inner": 2,
      "median_ms": 17.84342900282354,
      "q1_ms": 17.838369501987472,
      "q3_ms": 17.84441099880496,
      "peak_mb": 3089.375,
      "rel_err": 1.3919330702381483e-06
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 3072.1252670288086,
      "xla_arg_mb": 1.0625038146972656,
      "xla_out_mb": 1.0625343322753906,
      "compile_s": 0.8858174699998926,
      "inner": 3,
      "median_ms": 13.717186666326597,
      "q1_ms": 13.708012668454709,
      "q3_ms": 13.73083933382683,
      "rel_err": {
        "dq": 0.000881775549351918,
        "dk": 0.0001385138857005559,
        "dv": 0.00014329507462852186,
        "dlengthscale": 0.0010748536545408817
      }
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 32.14436340332031,
      "xla_arg_mb": 1.0625038146972656,
      "xla_out_mb": 1.0625343322753906,
      "compile_s": 0.5730989429939655,
      "inner": 2,
      "median_ms": 16.85548900059075,
      "q1_ms": 16.851972999575082,
      "q3_ms": 16.863629502040567,
      "rel_err": {
        "dq": 0.0005785145995253749,
        "dk": 0.00015562294933290705,
        "dv": 5.431522132134613e-05,
        "dlengthscale": 0.0004717947376908426
      }
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 1.0000152587890625,
      "xla_arg_mb": 1.0625038146972656,
      "xla_out_mb": 1.0625343322753906,
      "compile_s": 0.33429656400403474,
      "inner": 13,
      "median_ms": 3.724933384756486,
      "q1_ms": 3.6799012310579062,
      "q3_ms": 3.7313886921485673,
      "rel_err": {
        "dq": 0.0011106504600731121,
        "dk": 0.00039586208048699546,
        "dv": 0.0003980356960548107,
        "dlengthscale": 0.0043201810215070715
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 8,
      "out": 1,
      "compile_s": 0.052002556003571954,
      "inner": 1,
      "median_ms": 51.619641002616845,
      "q1_ms": 51.58514600043418,
      "q3_ms": 51.62599399773171,
      "peak_mb": 5139.5634765625,
      "rel_err": {
        "dq": 1.8468217627838418e-06,
        "dk": 2.1724794194235426e-07,
        "dv": 3.050837555826251e-07
      }
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 2048.062759399414,
      "xla_arg_mb": 2.0000038146972656,
      "xla_out_mb": 1.0,
      "compile_s": 0.17662105899944436,
      "inner": 9,
      "median_ms": 5.022413888784488,
      "q1_ms": 5.009413999990405,
      "q3_ms": 5.03518477813082,
      "rel_err": 0.0004538603736511576
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 16.501022338867188,
      "xla_arg_mb": 2.0000038146972656,
      "xla_out_mb": 1.0,
      "compile_s": 0.1491862180046155,
      "inner": 8,
      "median_ms": 6.139012124549481,
      "q1_ms": 6.129405000137922,
      "q3_ms": 6.143926249933429,
      "rel_err": 0.0004813642515449593
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 0.0,
      "xla_arg_mb": 2.0000038146972656,
      "xla_out_mb": 1.0,
      "compile_s": 0.1676394509995589,
      "inner": 59,
      "median_ms": 0.8050174067848583,
      "q1_ms": 0.797480949170106,
      "q3_ms": 0.8094176101020806,
      "rel_err": 0.00047503605149104854
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 8,
      "out": 16,
      "compile_s": 0.019989360000181478,
      "inner": 2,
      "median_ms": 18.1278524978552,
      "q1_ms": 18.12314850030816,
      "q3_ms": 18.134820500563364,
      "peak_mb": 3091.25,
      "rel_err": 1.2911350156843375e-06
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 4096.125511169434,
      "xla_arg_mb": 2.0000038146972656,
      "xla_out_mb": 2.0000343322753906,
      "compile_s": 0.22202508799819043,
      "inner": 3,
      "median_ms": 16.134541998326313,
      "q1_ms": 16.117883334421396,
      "q3_ms": 16.144687666383106,
      "rel_err": {
        "dq": 0.00030098864471493766,
        "dk": 0.0002820107013537091,
        "dv": 0.00027116851682839905,
        "dlengthscale": 2.992222793666894e-05
      }
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 49.09797668457031,
      "xla_arg_mb": 2.0000038146972656,
      "xla_out_mb": 2.0000343322753906,
      "compile_s": 0.2365845710010035,
      "inner": 2,
      "median_ms": 23.572517002321547,
      "q1_ms": 23.571915000502486,
      "q3_ms": 23.618172999704257,
      "rel_err": {
        "dq": 0.00038613404717888883,
        "dk": 0.0003638338627618456,
        "dv": 0.0002865382974232944,
        "dlengthscale": 7.156558337719227e-05
      }
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 1.0000152587890625,
      "xla_arg_mb": 2.0000038146972656,
      "xla_out_mb": 2.0000343322753906,
      "compile_s": 0.3520861500001047,
      "inner": 11,
      "median_ms": 4.397654363525693,
      "q1_ms": 4.3860297267780295,
      "q3_ms": 4.403215090860613,
      "rel_err": {
        "dq": 0.000361359420633864,
        "dk": 0.000353624949340216,
        "dv": 0.0003283392108207461,
        "dlengthscale": 1.0577602869332014e-05
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 8,
      "out": 16,
      "compile_s": 0.05326136799703818,
      "inner": 1,
      "median_ms": 51.933836002717726,
      "q1_ms": 51.9032980009797,
      "q3_ms": 51.937142001406755,
      "peak_mb": 5144.2509765625,
      "rel_err": {
        "dq": 9.07025359667594e-07,
        "dk": 1.6769009723252595e-07,
        "dv": 3.152454209925606e-07
      }
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 2048.062759399414,
      "xla_arg_mb": 4.062503814697266,
      "xla_out_mb": 0.0625,
      "compile_s": 0.46634075000474695,
      "inner": 9,
      "median_ms": 5.146441999790517,
      "q1_ms": 5.146286000025511,
      "q3_ms": 5.148706332773954,
      "rel_err": 0.0017719683416703642
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 16.065719604492188,
      "xla_arg_mb": 4.062503814697266,
      "xla_out_mb": 0.0625,
      "compile_s": 0.2826484680044814,
      "inner": 12,
      "median_ms": 4.134320000351484,
      "q1_ms": 4.121791416764609,
      "q3_ms": 4.13544058331657,
      "rel_err": 0.0002326226292687556
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 1.0,
      "xla_arg_mb": 4.062503814697266,
      "xla_out_mb": 0.0625,
      "compile_s": 0.19735726399812847,
      "inner": 24,
      "median_ms": 1.859463624896307,
      "q1_ms": 1.859242416685447,
      "q3_ms": 1.860227208150415,
      "rel_err": 0.001641141025805016
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 32,
      "out": 1,
      "compile_s": 0.02036287700320827,
      "inner": 2,
      "median_ms": 18.486150500393705,
      "q1_ms": 18.48384600089048,
      "q3_ms": 18.49042350295349,
      "peak_mb": 3092.375,
      "rel_err": 2.2879022782085846e-06
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 3072.1252670288086,
      "xla_arg_mb": 4.062503814697266,
      "xla_out_mb": 4.062534332275391,
      "compile_s": 0.908824320002168,
      "inner": 3,
      "median_ms": 13.964074333974471,
      "q1_ms": 13.956355665868614,
      "q3_ms": 13.98697066664075,
      "rel_err": {
        "dq": 0.0019310152404039462,
        "dk": 0.002302823752107087,
        "dv": 0.002401135615270241,
        "dlengthscale": 0.0016808642779408646
      }
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 32.16780090332031,
      "xla_arg_mb": 4.062503814697266,
      "xla_out_mb": 4.062534332275391,
      "compile_s": 0.7790191210006014,
      "inner": 2,
      "median_ms": 18.601323001348646,
      "q1_ms": 18.592932497995207,
      "q3_ms": 18.602115498651983,
      "rel_err": {
        "dq": 0.00039525478717798414,
        "dk": 4.89257286399017e-05,
        "dv": 1.2141832304175453e-05,
        "dlengthscale": 1.2681570329931077e-05
      }
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 1.0000152587890625,
      "xla_arg_mb": 4.062503814697266,
      "xla_out_mb": 4.062534332275391,
      "compile_s": 0.4536894979974022,
      "inner": 6,
      "median_ms": 7.711256999755278,
      "q1_ms": 7.682168665875603,
      "q3_ms": 7.721701666014269,
      "rel_err": {
        "dq": 0.0018701120835248014,
        "dk": 0.0020332929262989446,
        "dv": 0.0022132119645081895,
        "dlengthscale": 0.003649511208895165
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 32,
      "out": 1,
      "compile_s": 0.05355421299464069,
      "inner": 1,
      "median_ms": 52.31958600052167,
      "q1_ms": 52.31321500468766,
      "q3_ms": 52.32862300181296,
      "peak_mb": 5148.5634765625,
      "rel_err": {
        "dq": 2.588400080852777e-06,
        "dk": 1.6979100581562603e-07,
        "dv": 2.075918922141816e-07
      }
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 2048.062759399414,
      "xla_arg_mb": 5.000003814697266,
      "xla_out_mb": 1.0,
      "compile_s": 0.1827417329986929,
      "inner": 9,
      "median_ms": 5.169386555886983,
      "q1_ms": 5.168235444418517,
      "q3_ms": 5.170180222194176,
      "rel_err": 0.000340854453551966
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 16.501022338867188,
      "xla_arg_mb": 5.000003814697266,
      "xla_out_mb": 1.0,
      "compile_s": 0.15134340000076918,
      "inner": 7,
      "median_ms": 6.581093571315121,
      "q1_ms": 6.564770999830216,
      "q3_ms": 6.59838185778686,
      "rel_err": 0.0003399429657787489
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 0.0,
      "xla_arg_mb": 5.000003814697266,
      "xla_out_mb": 1.0,
      "compile_s": 0.21756980599457165,
      "inner": 22,
      "median_ms": 2.0041128634395795,
      "q1_ms": 2.0034361365436446,
      "q3_ms": 2.005034590679729,
      "rel_err": 0.0003568455289269294
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 32,
      "out": 16,
      "compile_s": 0.020649138998123817,
      "inner": 2,
      "median_ms": 18.77765700191958,
      "q1_ms": 18.766956996842055,
      "q3_ms": 18.780202000925783,
      "peak_mb": 3094.25,
      "rel_err": 9.547489433766405e-07
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 4096.125511169434,
      "xla_arg_mb": 5.000003814697266,
      "xla_out_mb": 5.000034332275391,
      "compile_s": 0.2195562649940257,
      "inner": 3,
      "median_ms": 16.537163666119643,
      "q1_ms": 16.44202366636212,
      "q3_ms": 16.54700500027199,
      "rel_err": {
        "dq": 0.00018455903609195217,
        "dk": 0.0002136116213783805,
        "dv": 0.00023691181396161824,
        "dlengthscale": 5.8730275232267006e-05
      }
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 49.12141418457031,
      "xla_arg_mb": 5.000003814697266,
      "xla_out_mb": 5.000034332275391,
      "compile_s": 0.24378251400048612,
      "inner": 1,
      "median_ms": 25.276145999669097,
      "q1_ms": 25.2751640000497,
      "q3_ms": 25.27970200026175,
      "rel_err": {
        "dq": 0.00023526544209147532,
        "dk": 0.00029560430579712486,
        "dv": 0.00021314125706740244,
        "dlengthscale": 0.00011831937338694062
      }
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 1.0000152587890625,
      "xla_arg_mb": 5.000003814697266,
      "xla_out_mb": 5.000034332275391,
      "compile_s": 0.4385870559999603,
      "inner": 6,
      "median_ms": 8.342408166451301,
      "q1_ms": 8.324228667333955,
      "q3_ms": 8.43080733223663,
      "rel_err": {
        "dq": 0.00026773156116133637,
        "dk": 0.00031452442857737135,
        "dv": 0.00039385195841988016,
        "dlengthscale": 0.00017255734410200386
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 32,
      "out": 16,
      "compile_s": 0.05404019199340837,
      "inner": 1,
      "median_ms": 52.66582100011874,
      "q1_ms": 52.66171300172573,
      "q3_ms": 52.707361006469,
      "peak_mb": 5153.2509765625,
      "rel_err": {
        "dq": 5.035940937197607e-07,
        "dk": 1.6123763685663227e-07,
        "dv": 3.300889265443891e-07
      }
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 2,
      "out": 1,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Out of memory while trying to allocate 32.00GiB with allocator GPU_0_bfc on device 0. [executable_name='jit__lambda'] [tf-allocator-allocation-error='']"
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 2,
      "out": 1,
      "xla_temp_mb": 16.314010620117188,
      "xla_arg_mb": 1.2500038146972656,
      "xla_out_mb": 0.25,
      "compile_s": 0.2365952229956747,
      "inner": 1,
      "median_ms": 52.991898999607656,
      "q1_ms": 52.991073000157485,
      "q3_ms": 53.14519449893851,
      "rel_err": 7.570602797027379e-05
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 2,
      "out": 1,
      "xla_temp_mb": 8.0,
      "xla_arg_mb": 1.2500038146972656,
      "xla_out_mb": 0.25,
      "compile_s": 0.16327213699696586,
      "inner": 6,
      "median_ms": 8.56127116639982,
      "q1_ms": 8.319126166801045,
      "q3_ms": 8.569862416455482,
      "rel_err": 0.0001332208820502045
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 2,
      "out": 1,
      "error": "OutOfMemoryError: CUDA out of memory. Tried to allocate 16.00 GiB. GPU 0 has a total capacity of 23.61 GiB of which 5.54 GiB is free. Including non-PyTorch memory, this process has 18.03 GiB memory in use. Of the allocated memory 3.25 MiB is allocated by PyTorch, and 2.75 MiB is reserved by PyTorch but unallocated. I"
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 2,
      "out": 1,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Out of memory while trying to allocate 32.06GiB with allocator GPU_0_bfc on device 0. [executable_name='jit_loss'] [tf-allocator-allocation-error='']"
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 2,
      "out": 1,
      "xla_temp_mb": 33.01081848144531,
      "xla_arg_mb": 1.2500038146972656,
      "xla_out_mb": 1.2500343322753906,
      "compile_s": 0.5692150830000173,
      "inner": 1,
      "median_ms": 244.43382400204428,
      "q1_ms": 244.3613665018347,
      "q3_ms": 244.78542400174774,
      "rel_err": {
        "dq": 0.0002049347131189955,
        "dk": 6.4787251052357e-05,
        "dv": 4.1788874628903204e-05,
        "dlengthscale": 2.686807202306878e-05
      }
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 2,
      "out": 1,
      "xla_temp_mb": 10.000015258789062,
      "xla_arg_mb": 1.2500038146972656,
      "xla_out_mb": 1.2500343322753906,
      "compile_s": 0.41993899600493023,
      "inner": 1,
      "median_ms": 49.08369500481058,
      "q1_ms": 49.0666025034443,
      "q3_ms": 49.12173750199145,
      "rel_err": {
        "dq": 0.00031756754342912856,
        "dk": 0.00022097469643961128,
        "dv": 0.0009325765608562457,
        "dlengthscale": 9.895636354442286e-05
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 2,
      "out": 1,
      "error": "OutOfMemoryError: CUDA out of memory. Tried to allocate 16.00 GiB. GPU 0 has a total capacity of 23.61 GiB of which 5.54 GiB is free. Including non-PyTorch memory, this process has 18.03 GiB memory in use. Of the allocated memory 4.50 MiB is allocated by PyTorch, and 1.50 MiB is reserved by PyTorch but unallocated. I"
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 2,
      "out": 16,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Out of memory while trying to allocate 32.00GiB with allocator GPU_0_bfc on device 0. [tf-allocator-allocation-error=''] [executable_name='jit__lambda']"
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 2,
      "out": 16,
      "xla_temp_mb": 16.251510620117188,
      "xla_arg_mb": 5.000003814697266,
      "xla_out_mb": 4.0,
      "compile_s": 0.6226726940003573,
      "inner": 1,
      "median_ms": 100.10267200414091,
      "q1_ms": 100.09693600295577,
      "q3_ms": 100.24520650404156,
      "rel_err": 0.0002720896624216077
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 2,
      "out": 16,
      "xla_temp_mb": 4.0,
      "xla_arg_mb": 5.000003814697266,
      "xla_out_mb": 4.0,
      "compile_s": 0.1940030600017053,
      "inner": 5,
      "median_ms": 10.48745419975603,
      "q1_ms": 10.258097199402982,
      "q3_ms": 10.518126100214431,
      "rel_err": 0.00036931391428800654
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 2,
      "out": 16,
      "error": "OutOfMemoryError: CUDA out of memory. Tried to allocate 16.00 GiB. GPU 0 has a total capacity of 23.61 GiB of which 5.52 GiB is free. Including non-PyTorch memory, this process has 18.05 GiB memory in use. Of the allocated memory 7.00 MiB is allocated by PyTorch, and 19.00 MiB is reserved by PyTorch but unallocated. "
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 2,
      "out": 16,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Out of memory while trying to allocate 48.06GiB with allocator GPU_0_bfc on device 0. [tf-allocator-allocation-error=''] [executable_name='jit_loss']"
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 2,
      "out": 16,
      "xla_temp_mb": 52.51472473144531,
      "xla_arg_mb": 5.000003814697266,
      "xla_out_mb": 5.000034332275391,
      "compile_s": 0.6760045070041087,
      "inner": 1,
      "median_ms": 446.95149299514014,
      "q1_ms": 433.1019299970649,
      "q3_ms": 449.7755159973167,
      "rel_err": {
        "dq": 0.0002667452703422564,
        "dk": 0.00030297283314736396,
        "dv": 0.00021342322275084017,
        "dlengthscale": 1.3604932547804552e-05
      }
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 2,
      "out": 16,
      "xla_temp_mb": 8.003921508789062,
      "xla_arg_mb": 5.000003814697266,
      "xla_out_mb": 5.000034332275391,
      "compile_s": 0.39436010099598207,
      "inner": 1,
      "median_ms": 62.31137099530315,
      "q1_ms": 59.95440199694713,
      "q3_ms": 62.86612649637391,
      "rel_err": {
        "dq": 0.00030422492198437385,
        "dk": 0.00039275777198325063,
        "dv": 0.0011023992298342296,
        "dlengthscale": 0.00019214758630827858
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 2,
      "out": 16,
      "error": "OutOfMemoryError: CUDA out of memory. Tried to allocate 16.00 GiB. GPU 0 has a total capacity of 23.61 GiB of which 4.45 GiB is free. Including non-PyTorch memory, this process has 18.05 GiB memory in use. Process 1398290 has 266.00 MiB memory in use. Process 1398435 has 410.00 MiB memory in use. Process 1398438 has "
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 1,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Out of memory while trying to allocate 32.00GiB with allocator GPU_0_bfc on device 0. [tf-allocator-allocation-error=''] [executable_name='jit__lambda']"
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 16.314010620117188,
      "xla_arg_mb": 4.250003814697266,
      "xla_out_mb": 0.25,
      "compile_s": 0.3046001580005395,
      "inner": 1,
      "median_ms": 58.176067999738734,
      "q1_ms": 58.09035600032075,
      "q3_ms": 58.58033299955423,
      "rel_err": 0.0001339336907187664
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 4.0,
      "xla_arg_mb": 4.250003814697266,
      "xla_out_mb": 0.25,
      "compile_s": 0.19694530600099824,
      "inner": 6,
      "median_ms": 9.890561166685075,
      "q1_ms": 9.685877249770176,
      "q3_ms": 10.047856249608838,
      "rel_err": 0.0004258536975935094
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 1,
      "error": "OutOfMemoryError: CUDA out of memory. Tried to allocate 16.00 GiB. GPU 0 has a total capacity of 23.61 GiB of which 4.20 GiB is free. Including non-PyTorch memory, this process has 18.04 GiB memory in use. Process 1398290 has 266.00 MiB memory in use. Process 1398435 has 540.00 MiB memory in use. Process 1398438 has "
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 1,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Out of memory while trying to allocate 32.06GiB with allocator GPU_0_bfc on device 0. [executable_name='jit_loss'] [tf-allocator-allocation-error='']"
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 33.01228332519531,
      "xla_arg_mb": 4.250003814697266,
      "xla_out_mb": 4.250034332275391,
      "compile_s": 0.5970431129971985,
      "inner": 1,
      "median_ms": 258.9217650020146,
      "q1_ms": 258.8943380033015,
      "q3_ms": 265.4351380006119,
      "rel_err": {
        "dq": 0.0003427874576850463,
        "dk": 4.7188125382821614e-05,
        "dv": 2.3535916709716275e-05,
        "dlengthscale": 2.14969233768624e-06
      }
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 4.0000152587890625,
      "xla_arg_mb": 4.250003814697266,
      "xla_out_mb": 4.250034332275391,
      "compile_s": 0.39100900199991884,
      "inner": 1,
      "median_ms": 49.87342999811517,
      "q1_ms": 49.699274997692555,
      "q3_ms": 52.05412099894602,
      "rel_err": {
        "dq": 0.0005792850186256146,
        "dk": 0.00037737415382191434,
        "dv": 0.0011879417297920721,
        "dlengthscale": 0.00025961669001287665
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 1,
      "error": "OutOfMemoryError: CUDA out of memory. Tried to allocate 16.00 GiB. GPU 0 has a total capacity of 23.61 GiB of which 4.19 GiB is free. Including non-PyTorch memory, this process has 18.04 GiB memory in use. Process 1398290 has 266.00 MiB memory in use. Process 1398435 has 542.00 MiB memory in use. Process 1398438 has "
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 16,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Out of memory while trying to allocate 32.00GiB with allocator GPU_0_bfc on device 0. [tf-allocator-allocation-error=''] [executable_name='jit__lambda']"
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 16.251510620117188,
      "xla_arg_mb": 8.000003814697266,
      "xla_out_mb": 4.0,
      "compile_s": 0.6242642750003142,
      "inner": 1,
      "median_ms": 104.85053499724017,
      "q1_ms": 104.42442349813064,
      "q3_ms": 107.75169900080073,
      "rel_err": 0.0002566603840240124
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 0.0,
      "xla_arg_mb": 8.000003814697266,
      "xla_out_mb": 4.0,
      "compile_s": 0.22268543599784607,
      "inner": 5,
      "median_ms": 10.615823000262026,
      "q1_ms": 10.432895900157746,
      "q3_ms": 10.648515800130554,
      "rel_err": 0.0003292965508110889
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 16,
      "error": "OutOfMemoryError: CUDA out of memory. Tried to allocate 16.00 GiB. GPU 0 has a total capacity of 23.61 GiB of which 4.20 GiB is free. Including non-PyTorch memory, this process has 18.04 GiB memory in use. Process 1398290 has 266.00 MiB memory in use. Process 1398435 has 542.00 MiB memory in use. Process 1398438 has "
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 16,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Out of memory while trying to allocate 48.06GiB with allocator GPU_0_bfc on device 0. [executable_name='jit_loss'] [tf-allocator-allocation-error='']"
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 52.51618957519531,
      "xla_arg_mb": 8.000003814697266,
      "xla_out_mb": 8.00003433227539,
      "compile_s": 0.5417805980032426,
      "inner": 1,
      "median_ms": 398.67910299653886,
      "q1_ms": 398.5993319984118,
      "q3_ms": 400.9853014977125,
      "rel_err": {
        "dq": 0.00019049642703273314,
        "dk": 0.00027913899471866183,
        "dv": 0.00018824836324637744,
        "dlengthscale": 0.00014128377977437923
      }
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 4.0000152587890625,
      "xla_arg_mb": 8.000003814697266,
      "xla_out_mb": 8.00003433227539,
      "compile_s": 0.3724540250041173,
      "inner": 1,
      "median_ms": 58.72082900168607,
      "q1_ms": 58.09851150115719,
      "q3_ms": 62.053764500888065,
      "rel_err": {
        "dq": 0.0003211574092597976,
        "dk": 0.0003460264140785003,
        "dv": 0.0011243937670311255,
        "dlengthscale": 0.0003029466744495477
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 16,
      "error": "OutOfMemoryError: CUDA out of memory. Tried to allocate 16.00 GiB. GPU 0 has a total capacity of 23.61 GiB of which 3.74 GiB is free. Including non-PyTorch memory, this process has 18.06 GiB memory in use. Process 1398290 has 266.00 MiB memory in use. Process 1398435 has 728.00 MiB memory in use. Process 1398438 has "
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 32,
      "out": 1,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Out of memory while trying to allocate 32.00GiB with allocator GPU_0_bfc on device 0. [executable_name='jit__lambda'] [tf-allocator-allocation-error='']"
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 16.314010620117188,
      "xla_arg_mb": 16.250003814697266,
      "xla_out_mb": 0.25,
      "compile_s": 0.7478080979999504,
      "inner": 1,
      "median_ms": 72.55690699821571,
      "q1_ms": 72.55572950089118,
      "q3_ms": 72.55793349759188,
      "rel_err": 3.75113714985198e-05
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 4.0,
      "xla_arg_mb": 16.250003814697266,
      "xla_out_mb": 0.25,
      "compile_s": 0.5445247800016659,
      "inner": 2,
      "median_ms": 23.72510750137735,
      "q1_ms": 23.590866001541144,
      "q3_ms": 24.543601251934888,
      "rel_err": 0.0001588064487459717
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 32,
      "out": 1,
      "error": "OutOfMemoryError: CUDA out of memory. Tried to allocate 16.00 GiB. GPU 0 has a total capacity of 23.61 GiB of which 3.70 GiB is free. Including non-PyTorch memory, this process has 18.08 GiB memory in use. Process 1398290 has 266.00 MiB memory in use. Process 1398435 has 748.00 MiB memory in use. Process 1398438 has "
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 32,
      "out": 1,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Out of memory while trying to allocate 32.06GiB with allocator GPU_0_bfc on device 0. [executable_name='jit_loss'] [tf-allocator-allocation-error='']"
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 33.01814270019531,
      "xla_arg_mb": 16.250003814697266,
      "xla_out_mb": 16.25003433227539,
      "compile_s": 1.639937659005227,
      "inner": 1,
      "median_ms": 353.6672409973107,
      "q1_ms": 351.12190050131176,
      "q3_ms": 357.54801799703273,
      "rel_err": {
        "dq": 0.00018792139390235314,
        "dk": 8.08565538365244e-06,
        "dv": 3.1343467024518383e-06,
        "dlengthscale": 1.0107265653857393e-06
      }
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 4.0000152587890625,
      "xla_arg_mb": 16.250003814697266,
      "xla_out_mb": 16.25003433227539,
      "compile_s": 0.9185581740021007,
      "inner": 1,
      "median_ms": 105.82950700336369,
      "q1_ms": 104.91833800188033,
      "q3_ms": 106.55989900260465,
      "rel_err": {
        "dq": 0.0004077921565150324,
        "dk": 0.0002098497317836943,
        "dv": 0.0008561674742966704,
        "dlengthscale": 0.0002607674538695207
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 32,
      "out": 1,
      "error": "OutOfMemoryError: CUDA out of memory. Tried to allocate 16.00 GiB. GPU 0 has a total capacity of 23.61 GiB of which 2.60 GiB is free. Including non-PyTorch memory, this process has 18.10 GiB memory in use. Process 1398290 has 266.00 MiB memory in use. Process 1398435 has 1.48 GiB memory in use. Process 1398438 has 79"
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 32,
      "out": 16,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Out of memory while trying to allocate 32.00GiB with allocator GPU_0_bfc on device 0. [tf-allocator-allocation-error=''] [executable_name='jit__lambda']"
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 16.251510620117188,
      "xla_arg_mb": 20.000003814697266,
      "xla_out_mb": 4.0,
      "compile_s": 0.6099694000004092,
      "inner": 1,
      "median_ms": 122.01040099898819,
      "q1_ms": 121.61912099691108,
      "q3_ms": 122.77293900115183,
      "rel_err": 0.0004076308986506655
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 0.0,
      "xla_arg_mb": 20.000003814697266,
      "xla_out_mb": 4.0,
      "compile_s": 0.3961015969980508,
      "inner": 1,
      "median_ms": 25.128201996267308,
      "q1_ms": 25.11588400011533,
      "q3_ms": 25.138551496638684,
      "rel_err": 0.0003999355875769339
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 32,
      "out": 16,
      "error": "OutOfMemoryError: CUDA out of memory. Tried to allocate 16.00 GiB. GPU 0 has a total capacity of 23.61 GiB of which 2.95 GiB is free. Including non-PyTorch memory, this process has 18.08 GiB memory in use. Process 1398290 has 266.00 MiB memory in use. Process 1398435 has 1.48 GiB memory in use. Process 1398438 has 80"
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 32,
      "out": 16,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Out of memory while trying to allocate 48.06GiB with allocator GPU_0_bfc on device 0. [executable_name='jit_loss'] [tf-allocator-allocation-error='']"
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 52.52204895019531,
      "xla_arg_mb": 20.000003814697266,
      "xla_out_mb": 20.00003433227539,
      "compile_s": 0.29779620499903103,
      "inner": 1,
      "median_ms": 482.51511899434263,
      "q1_ms": 482.31387799751246,
      "q3_ms": 482.76067349434015,
      "rel_err": {
        "dq": 0.00034501094956540935,
        "dk": 0.0003811953315083937,
        "dv": 0.00032063782807241465,
        "dlengthscale": 0.00026487939515569537
      }
    },
    {
      "family": "kernel_mv",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 4.0000152587890625,
      "xla_arg_mb": 20.000003814697266,
      "xla_out_mb": 20.00003433227539,
      "compile_s": 0.4357475149954553,
      "inner": 1,
      "median_ms": 112.32389499491546,
      "q1_ms": 111.8968019982276,
      "q3_ms": 113.12964449462015,
      "rel_err": {
        "dq": 0.00020710062773278596,
        "dk": 0.0003689897128854479,
        "dv": 0.0010307062159115308,
        "dlengthscale": 4.929135057561207e-06
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 32,
      "out": 16,
      "error": "OutOfMemoryError: CUDA out of memory. Tried to allocate 16.00 GiB. GPU 0 has a total capacity of 23.61 GiB of which 5.47 GiB is free. Including non-PyTorch memory, this process has 18.10 GiB memory in use. Of the allocated memory 57.00 MiB is allocated by PyTorch, and 23.00 MiB is reserved by PyTorch but unallocated."
    },
    {
      "family": "kde_density",
      "impl": "dense",
      "mode": "fwd",
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "xla_temp_mb": 4.0078125,
      "xla_arg_mb": 0.015628814697265625,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.0007323360041482374,
      "inner": 200,
      "median_ms": 0.044207190003362484,
      "q1_ms": 0.043158055013918784,
      "q3_ms": 0.05126223499246407
    },
    {
      "family": "kde_density",
      "impl": "xla",
      "mode": "fwd",
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "xla_temp_mb": 4.0078125,
      "xla_arg_mb": 0.015628814697265625,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.0006862279988126829,
      "inner": 200,
      "median_ms": 0.04242947001330322,
      "q1_ms": 0.042373169999336824,
      "q3_ms": 0.04258502500306349
    },
    {
      "family": "kde_density",
      "impl": "pallas",
      "mode": "fwd",
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "xla_temp_mb": 0.125,
      "xla_arg_mb": 0.015628814697265625,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.0006982609993428923,
      "inner": 200,
      "median_ms": 0.03803971500019543,
      "q1_ms": 0.037817800002812874,
      "q3_ms": 0.038201474999368656
    },
    {
      "family": "kde_density",
      "impl": "dense",
      "mode": "fwd_bwd",
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "xla_temp_mb": 8.016136169433594,
      "xla_arg_mb": 0.015628814697265625,
      "xla_out_mb": 0.007831573486328125,
      "compile_s": 0.0009094300039578229,
      "inner": 200,
      "median_ms": 0.06088338999688858,
      "q1_ms": 0.06080519000533968,
      "q3_ms": 0.06102060000557685
    },
    {
      "family": "kde_density",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "xla_temp_mb": 8.016136169433594,
      "xla_arg_mb": 0.015628814697265625,
      "xla_out_mb": 0.007831573486328125,
      "compile_s": 0.0007971689992700703,
      "inner": 200,
      "median_ms": 0.051044824976997916,
      "q1_ms": 0.050477855002100114,
      "q3_ms": 0.051362375015742145
    },
    {
      "family": "kde_density",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "xla_temp_mb": 0.1565093994140625,
      "xla_arg_mb": 0.015628814697265625,
      "xla_out_mb": 0.007831573486328125,
      "compile_s": 0.0008720300029381178,
      "inner": 200,
      "median_ms": 0.05946550500084413,
      "q1_ms": 0.05941150502621895,
      "q3_ms": 0.059640034996846225
    },
    {
      "family": "kde_density",
      "impl": "dense",
      "mode": "fwd",
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "xla_temp_mb": 4.0078125,
      "xla_arg_mb": 0.06250381469726562,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.0006654290045844391,
      "inner": 200,
      "median_ms": 0.04308922500058543,
      "q1_ms": 0.04202125001029344,
      "q3_ms": 0.045785524998791516
    },
    {
      "family": "kde_density",
      "impl": "xla",
      "mode": "fwd",
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "xla_temp_mb": 4.0078125,
      "xla_arg_mb": 0.06250381469726562,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.000608670998190064,
      "inner": 200,
      "median_ms": 0.04255981999449432,
      "q1_ms": 0.04213928001263412,
      "q3_ms": 0.04281671001081122
    },
    {
      "family": "kde_density",
      "impl": "pallas",
      "mode": "fwd",
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "xla_temp_mb": 0.0625,
      "xla_arg_mb": 0.06250381469726562,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.0005835839983774349,
      "inner": 200,
      "median_ms": 0.04407889999129111,
      "q1_ms": 0.044044729984307196,
      "q3_ms": 0.04408159999002237
    },
    {
      "family": "kde_density",
      "impl": "dense",
      "mode": "fwd_bwd",
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "xla_temp_mb": 8.016136169433594,
      "xla_arg_mb": 0.06250381469726562,
      "xla_out_mb": 0.031269073486328125,
      "compile_s": 0.0009311519970651716,
      "inner": 200,
      "median_ms": 0.05044113499025116,
      "q1_ms": 0.04997414998797467,
      "q3_ms": 0.050448200017854106
    },
    {
      "family": "kde_density",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "xla_temp_mb": 8.016136169433594,
      "xla_arg_mb": 0.06250381469726562,
      "xla_out_mb": 0.031269073486328125,
      "compile_s": 0.0008433260009041987,
      "inner": 200,
      "median_ms": 0.046820179995847866,
      "q1_ms": 0.04669404501328245,
      "q3_ms": 0.047149205020105
    },
    {
      "family": "kde_density",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "xla_temp_mb": 0.0630035400390625,
      "xla_arg_mb": 0.06250381469726562,
      "xla_out_mb": 0.031269073486328125,
      "compile_s": 0.000906765999388881,
      "inner": 200,
      "median_ms": 0.05658560501615284,
      "q1_ms": 0.056472994983778335,
      "q3_ms": 0.056613665001350455
    },
    {
      "family": "kde_density",
      "impl": "dense",
      "mode": "fwd",
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "xla_temp_mb": 64.01562881469727,
      "xla_arg_mb": 0.06250381469726562,
      "xla_out_mb": 0.015625,
      "compile_s": 0.0010788009967654943,
      "inner": 114,
      "median_ms": 0.32730161400283114,
      "q1_ms": 0.32722849994714026,
      "q3_ms": 0.32737438596086577
    },
    {
      "family": "kde_density",
      "impl": "xla",
      "mode": "fwd",
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "xla_temp_mb": 16.024696350097656,
      "xla_arg_mb": 0.06250381469726562,
      "xla_out_mb": 0.015625,
      "compile_s": 0.0011227739960304461,
      "inner": 155,
      "median_ms": 0.212190690311423,
      "q1_ms": 0.2120993483620095,
      "q3_ms": 0.2122096903246617
    },
    {
      "family": "kde_density",
      "impl": "pallas",
      "mode": "fwd",
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "xla_temp_mb": 0.5,
      "xla_arg_mb": 0.06250381469726562,
      "xla_out_mb": 0.015625,
      "compile_s": 0.00076145999628352,
      "inner": 200,
      "median_ms": 0.08011818001250504,
      "q1_ms": 0.0800966899987543,
      "q3_ms": 0.08013526501599699
    },
    {
      "family": "kde_density",
      "impl": "dense",
      "mode": "fwd_bwd",
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "xla_temp_mb": 128.0627670288086,
      "xla_arg_mb": 0.06250381469726562,
      "xla_out_mb": 0.031269073486328125,
      "compile_s": 0.0015777049993630499,
      "inner": 75,
      "median_ms": 0.5736504000378773,
      "q1_ms": 0.5735259066568688,
      "q3_ms": 0.573724013423392
    },
    {
      "family": "kde_density",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "xla_temp_mb": 16.08428192138672,
      "xla_arg_mb": 0.06250381469726562,
      "xla_out_mb": 0.031269073486328125,
      "compile_s": 0.0019471340056043118,
      "inner": 63,
      "median_ms": 0.6858891110390513,
      "q1_ms": 0.685873698366488,
      "q3_ms": 0.6858894285248843
    },
    {
      "family": "kde_density",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "xla_temp_mb": 0.6252593994140625,
      "xla_arg_mb": 0.06250381469726562,
      "xla_out_mb": 0.031269073486328125,
      "compile_s": 0.0011043189952033572,
      "inner": 153,
      "median_ms": 0.17160250984316425,
      "q1_ms": 0.17159517643877362,
      "q3_ms": 0.1716253660037943
    },
    {
      "family": "kde_density",
      "impl": "dense",
      "mode": "fwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "xla_temp_mb": 64.01562881469727,
      "xla_arg_mb": 0.2500038146972656,
      "xla_out_mb": 0.015625,
      "compile_s": 0.001023295997583773,
      "inner": 118,
      "median_ms": 0.32613483902315565,
      "q1_ms": 0.3261192118693774,
      "q3_ms": 0.32617270335322246
    },
    {
      "family": "kde_density",
      "impl": "xla",
      "mode": "fwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "xla_temp_mb": 16.036170959472656,
      "xla_arg_mb": 0.2500038146972656,
      "xla_out_mb": 0.015625,
      "compile_s": 0.0010666180023690686,
      "inner": 158,
      "median_ms": 0.21013953798351798,
      "q1_ms": 0.21003706961327903,
      "q3_ms": 0.2101730823124865
    },
    {
      "family": "kde_density",
      "impl": "pallas",
      "mode": "fwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "xla_temp_mb": 0.25,
      "xla_arg_mb": 0.2500038146972656,
      "xla_out_mb": 0.015625,
      "compile_s": 0.0006970789982005954,
      "inner": 200,
      "median_ms": 0.07921627500763861,
      "q1_ms": 0.07915556499938248,
      "q3_ms": 0.07922804499685299
    },
    {
      "family": "kde_density",
      "impl": "dense",
      "mode": "fwd_bwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "xla_temp_mb": 128.0627670288086,
      "xla_arg_mb": 0.2500038146972656,
      "xla_out_mb": 0.12501907348632812,
      "compile_s": 0.0014295340006356128,
      "inner": 76,
      "median_ms": 0.5756023157645311,
      "q1_ms": 0.5755230920744667,
      "q3_ms": 0.5756541315924167
    },
    {
      "family": "kde_density",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "xla_temp_mb": 16.27178192138672,
      "xla_arg_mb": 0.2500038146972656,
      "xla_out_mb": 0.12501907348632812,
      "compile_s": 0.0019209950041840784,
      "inner": 63,
      "median_ms": 0.6812713174786907,
      "q1_ms": 0.6808969683045271,
      "q3_ms": 0.6812830951752229
    },
    {
      "family": "kde_density",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "xla_temp_mb": 0.2505035400390625,
      "xla_arg_mb": 0.2500038146972656,
      "xla_out_mb": 0.12501907348632812,
      "compile_s": 0.0009930790038197301,
      "inner": 171,
      "median_ms": 0.16824984794505735,
      "q1_ms": 0.16819916370941587,
      "q3_ms": 0.1682530701730934
    },
    {
      "family": "kde_density",
      "impl": "dense",
      "mode": "fwd",
      "n": 16384,
      "m": 16384,
      "dim": 2,
      "xla_temp_mb": 2048.062759399414,
      "xla_arg_mb": 0.2500038146972656,
      "xla_out_mb": 0.0625,
      "compile_s": 0.058590953005477786,
      "inner": 9,
      "median_ms": 4.98691966721607,
      "q1_ms": 4.985764222510625,
      "q3_ms": 5.000403889021578
    },
    {
      "family": "kde_density",
      "impl": "xla",
      "mode": "fwd",
      "n": 16384,
      "m": 16384,
      "dim": 2,
      "xla_temp_mb": 16.065711975097656,
      "xla_arg_mb": 0.2500038146972656,
      "xla_out_mb": 0.0625,
      "compile_s": 0.004144470993196592,
      "inner": 14,
      "median_ms": 3.351376357126615,
      "q1_ms": 3.3509514287288766,
      "q3_ms": 3.3649107146110118
    },
    {
      "family": "kde_density",
      "impl": "pallas",
      "mode": "fwd",
      "n": 16384,
      "m": 16384,
      "dim": 2,
      "xla_temp_mb": 2.0,
      "xla_arg_mb": 0.2500038146972656,
      "xla_out_mb": 0.0625,
      "compile_s": 0.0012919639993924648,
      "inner": 68,
      "median_ms": 0.6787947058910504,
      "q1_ms": 0.6761383677204482,
      "q3_ms": 0.6790099558951196
    },
    {
      "family": "kde_density",
      "impl": "dense",
      "mode": "fwd_bwd",
      "n": 16384,
      "m": 16384,
      "dim": 2,
      "xla_temp_mb": 2048.3752670288086,
      "xla_arg_mb": 0.2500038146972656,
      "xla_out_mb": 0.12501907348632812,
      "compile_s": 0.00967577700066613,
      "inner": 5,
      "median_ms": 8.752126799663529,
      "q1_ms": 8.746590401278809,
      "q3_ms": 8.77446480008075
    },
    {
      "family": "kde_density",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 16384,
      "m": 16384,
      "dim": 2,
      "xla_temp_mb": 16.13823699951172,
      "xla_arg_mb": 0.2500038146972656,
      "xla_out_mb": 0.12501907348632812,
      "compile_s": 0.012408336995576974,
      "inner": 4,
      "median_ms": 11.056220499085612,
      "q1_ms": 11.043674499887857,
      "q3_ms": 11.08941600068647
    },
    {
      "family": "kde_density",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 16384,
      "m": 16384,
      "dim": 2,
      "xla_temp_mb": 2.5002593994140625,
      "xla_arg_mb": 0.2500038146972656,
      "xla_out_mb": 0.12501907348632812,
      "compile_s": 0.0022723689980921336,
      "inner": 32,
      "median_ms": 1.5500504375722812,
      "q1_ms": 1.5435225000146602,
      "q3_ms": 1.5522721876095602
    },
    {
      "family": "kde_density",
      "impl": "dense",
      "mode": "fwd",
      "n": 16384,
      "m": 16384,
      "dim": 8,
      "xla_temp_mb": 2048.062759399414,
      "xla_arg_mb": 1.0000038146972656,
      "xla_out_mb": 0.0625,
      "compile_s": 0.005720293003832921,
      "inner": 9,
      "median_ms": 4.981327889254317,
      "q1_ms": 4.979681445143393,
      "q3_ms": 4.983120222782923
    },
    {
      "family": "kde_density",
      "impl": "xla",
      "mode": "fwd",
      "n": 16384,
      "m": 16384,
      "dim": 8,
      "xla_temp_mb": 16.065711975097656,
      "xla_arg_mb": 1.0000038146972656,
      "xla_out_mb": 0.0625,
      "compile_s": 0.004180529002042022,
      "inner": 14,
      "median_ms": 3.371538928539459,
      "q1_ms": 3.370232857248214,
      "q3_ms": 3.3718280714570676
    },
    {
      "family": "kde_density",
      "impl": "pallas",
      "mode": "fwd",
      "n": 16384,
      "m": 16384,
      "dim": 8,
      "xla_temp_mb": 1.0,
      "xla_arg_mb": 1.0000038146972656,
      "xla_out_mb": 0.0625,
      "compile_s": 0.001282146004086826,
      "inner": 66,
      "median_ms": 0.6919113485227255,
      "q1_ms": 0.6893322272596627,
      "q3_ms": 0.6932502424619583
    },
    {
      "family": "kde_density",
      "impl": "dense",
      "mode": "fwd_bwd",
      "n": 16384,
      "m": 16384,
      "dim": 8,
      "xla_temp_mb": 2048.2505111694336,
      "xla_arg_mb": 1.0000038146972656,
      "xla_out_mb": 0.5000190734863281,
      "compile_s": 0.009625560996937566,
      "inner": 5,
      "median_ms": 8.778843400068581,
      "q1_ms": 8.768628000689205,
      "q3_ms": 8.82604679936776
    },
    {
      "family": "kde_density",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 16384,
      "m": 16384,
      "dim": 8,
      "xla_temp_mb": 16.31572723388672,
      "xla_arg_mb": 1.0000038146972656,
      "xla_out_mb": 0.5000190734863281,
      "compile_s": 0.012197046999062877,
      "inner": 4,
      "median_ms": 11.086851000072784,
      "q1_ms": 11.080596748797689,
      "q3_ms": 11.094545499872766
    },
    {
      "family": "kde_density",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 16384,
      "m": 16384,
      "dim": 8,
      "xla_temp_mb": 1.0012359619140625,
      "xla_arg_mb": 1.0000038146972656,
      "xla_out_mb": 0.5000190734863281,
      "compile_s": 0.0022703360009472817,
      "inner": 31,
      "median_ms": 1.5763065483394798,
      "q1_ms": 1.5682351290560028,
      "q3_ms": 1.5772382903378457
    },
    {
      "family": "kde_density",
      "impl": "dense",
      "mode": "fwd",
      "n": 65536,
      "m": 65536,
      "dim": 2,
      "error": "JaxRuntimeError: NOT_FOUND: Failed to get configs for: 2 out of 7 instructions. See logs for all failures. Example failure: "
    },
    {
      "family": "kde_density",
      "impl": "xla",
      "mode": "fwd",
      "n": 65536,
      "m": 65536,
      "dim": 2,
      "xla_temp_mb": 16.314002990722656,
      "xla_arg_mb": 1.0000038146972656,
      "xla_out_mb": 0.25,
      "compile_s": 0.053066241001943126,
      "inner": 1,
      "median_ms": 52.69707900151843,
      "q1_ms": 52.69326599955093,
      "q3_ms": 52.759871003217995
    },
    {
      "family": "kde_density",
      "impl": "pallas",
      "mode": "fwd",
      "n": 65536,
      "m": 65536,
      "dim": 2,
      "xla_temp_mb": 8.0,
      "xla_arg_mb": 1.0000038146972656,
      "xla_out_mb": 0.25,
      "compile_s": 0.008391937000851613,
      "inner": 6,
      "median_ms": 8.545377832585169,
      "q1_ms": 8.454094999857867,
      "q3_ms": 8.558243499768045
    },
    {
      "family": "kde_density",
      "impl": "dense",
      "mode": "fwd_bwd",
      "n": 65536,
      "m": 65536,
      "dim": 2,
      "error": "JaxRuntimeError: NOT_FOUND: Failed to get configs for: 3 out of 18 instructions. See logs for all failures. Example failure: "
    },
    {
      "family": "kde_density",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 65536,
      "m": 65536,
      "dim": 2,
      "xla_temp_mb": 16.56645965576172,
      "xla_arg_mb": 1.0000038146972656,
      "xla_out_mb": 0.5000190734863281,
      "compile_s": 0.17759291199763538,
      "inner": 1,
      "median_ms": 176.85545700078364,
      "q1_ms": 176.5676219947636,
      "q3_ms": 176.94000000483356
    },
    {
      "family": "kde_density",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 65536,
      "m": 65536,
      "dim": 2,
      "xla_temp_mb": 10.000503540039062,
      "xla_arg_mb": 1.0000038146972656,
      "xla_out_mb": 0.5000190734863281,
      "compile_s": 0.018913084000814706,
      "inner": 2,
      "median_ms": 20.128973501414293,
      "q1_ms": 19.95579600043129,
      "q3_ms": 20.131217999733053
    },
    {
      "family": "kde_density",
      "impl": "dense",
      "mode": "fwd",
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "error": "JaxRuntimeError: NOT_FOUND: Failed to get configs for: 2 out of 7 instructions. See logs for all failures. Example failure: "
    },
    {
      "family": "kde_density",
      "impl": "xla",
      "mode": "fwd",
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "xla_temp_mb": 16.314002990722656,
      "xla_arg_mb": 4.000003814697266,
      "xla_out_mb": 0.25,
      "compile_s": 0.05526784499670612,
      "inner": 1,
      "median_ms": 55.17832799523603,
      "q1_ms": 55.079630998079665,
      "q3_ms": 55.405945000529755
    },
    {
      "family": "kde_density",
      "impl": "pallas",
      "mode": "fwd",
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "xla_temp_mb": 4.0,
      "xla_arg_mb": 4.000003814697266,
      "xla_out_mb": 0.25,
      "compile_s": 0.00870797599782236,
      "inner": 6,
      "median_ms": 8.710553499743886,
      "q1_ms": 8.633663167226283,
      "q3_ms": 8.728936666254109
    },
    {
      "family": "kde_density",
      "impl": "dense",
      "mode": "fwd_bwd",
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "error": "JaxRuntimeError: RESOURCE_EXHAUSTED: Failed to get configs for: 3 out of 18 instructions. See logs for all failures. Example failure: "
    },
    {
      "family": "kde_density",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "xla_temp_mb": 16.56645965576172,
      "xla_arg_mb": 4.000003814697266,
      "xla_out_mb": 2.000019073486328,
      "compile_s": 0.18417184900317807,
      "inner": 1,
      "median_ms": 184.0264839993324,
      "q1_ms": 183.77770599909127,
      "q3_ms": 184.55276000167942
    },
    {
      "family": "kde_density",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "xla_temp_mb": 4.2505035400390625,
      "xla_arg_mb": 4.000003814697266,
      "xla_out_mb": 2.000019073486328,
      "compile_s": 0.019373374001588672,
      "inner": 2,
      "median_ms": 20.51986300284625,
      "q1_ms": 20.35054799853242,
      "q3_ms": 20.63729549990967
    },
    {
      "family": "kernel_mv_peak",
      "impl": "dense",
      "mode": "fwd",
      "n": 4096,
      "dim": 8,
      "out": 16,
      "run_s": 0.6253617089969339,
      "peak_mb": 321.99951171875
    },
    {
      "family": "kernel_mv_peak",
      "impl": "xla",
      "mode": "fwd",
      "n": 4096,
      "dim": 8,
      "out": 16,
      "run_s": 0.6359039240051061,
      "peak_mb": 148.5009765625
    },
    {
      "family": "kernel_mv_peak",
      "impl": "pallas",
      "mode": "fwd",
      "n": 4096,
      "dim": 8,
      "out": 16,
      "run_s": 0.17337567600043258,
      "peak_mb": 0.750732421875
    },
    {
      "family": "kernel_mv_peak",
      "impl": "dense",
      "mode": "fwd_bwd",
      "n": 4096,
      "dim": 8,
      "out": 16,
      "run_s": 1.5272621260010055,
      "peak_mb": 352.531982421875
    },
    {
      "family": "kernel_mv_peak",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 4096,
      "dim": 8,
      "out": 16,
      "run_s": 1.9810887120038387,
      "peak_mb": 148.5009765625
    },
    {
      "family": "kernel_mv_peak",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 4096,
      "dim": 8,
      "out": 16,
      "run_s": 0.38025483599631116,
      "peak_mb": 64.5009765625
    },
    {
      "family": "kernel_mv_peak",
      "impl": "dense",
      "mode": "fwd",
      "n": 16384,
      "dim": 8,
      "out": 16,
      "run_s": 0.8881765439946321,
      "peak_mb": 2150.0009765625
    },
    {
      "family": "kernel_mv_peak",
      "impl": "xla",
      "mode": "fwd",
      "n": 16384,
      "dim": 8,
      "out": 16,
      "run_s": 0.6617377090005903,
      "peak_mb": 150.0009765625
    },
    {
      "family": "kernel_mv_peak",
      "impl": "pallas",
      "mode": "fwd",
      "n": 16384,
      "dim": 8,
      "out": 16,
      "run_s": 0.30714679400261957,
      "peak_mb": 3.000732421875
    },
    {
      "family": "kernel_mv_peak",
      "impl": "dense",
      "mode": "fwd_bwd",
      "n": 16384,
      "dim": 8,
      "out": 16,
      "run_s": 2.140029948997835,
      "peak_mb": 4100.625732421875
    },
    {
      "family": "kernel_mv_peak",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 16384,
      "dim": 8,
      "out": 16,
      "run_s": 1.3562038050004048,
      "peak_mb": 150.0009765625
    },
    {
      "family": "kernel_mv_peak",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 16384,
      "dim": 8,
      "out": 16,
      "run_s": 0.3758589099961682,
      "peak_mb": 66.0009765625
    },
    {
      "family": "kernel_mv_peak",
      "impl": "dense",
      "mode": "fwd",
      "n": 65536,
      "dim": 8,
      "out": 16,
      "error": "jax.errors.JaxRuntimeError: RESOURCE_EXHAUSTED: Out of memory while trying to allocate 32.00GiB with allocator GPU_0_bfc on device 0. [tf-allocator-allocation-error=''] [executable_name='jit__lambda']"
    },
    {
      "family": "kernel_mv_peak",
      "impl": "xla",
      "mode": "fwd",
      "n": 65536,
      "dim": 8,
      "out": 16,
      "run_s": 0.7229622279992327,
      "peak_mb": 156.0009765625
    },
    {
      "family": "kernel_mv_peak",
      "impl": "pallas",
      "mode": "fwd",
      "n": 65536,
      "dim": 8,
      "out": 16,
      "run_s": 0.17694786399806617,
      "peak_mb": 12.000732421875
    },
    {
      "family": "kernel_mv_peak",
      "impl": "dense",
      "mode": "fwd_bwd",
      "n": 65536,
      "dim": 8,
      "out": 16,
      "error": "jax.errors.JaxRuntimeError: RESOURCE_EXHAUSTED: Out of memory while trying to allocate 48.06GiB with allocator GPU_0_bfc on device 0. [executable_name='jit_loss'] [tf-allocator-allocation-error='']"
    },
    {
      "family": "kernel_mv_peak",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 65536,
      "dim": 8,
      "out": 16,
      "run_s": 1.8435094179949374,
      "peak_mb": 156.0009765625
    },
    {
      "family": "kernel_mv_peak",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 65536,
      "dim": 8,
      "out": 16,
      "run_s": 0.4975039480050327,
      "peak_mb": 72.0009765625
    },
    {
      "family": "kernel_mv_max_n",
      "impl": "dense",
      "mode": "fwd_bwd",
      "n": 16384,
      "dim": 8,
      "out": 16,
      "run_s": 1.8520322430049418,
      "peak_mb": 4100.625732421875
    },
    {
      "family": "kernel_mv_max_n",
      "impl": "dense",
      "mode": "fwd_bwd",
      "n": 32768,
      "dim": 8,
      "out": 16,
      "error": "jax.errors.JaxRuntimeError: RESOURCE_EXHAUSTED: Out of memory while trying to allocate 12.02GiB with allocator GPU_0_bfc on device 0. [executable_name='jit_loss'] [tf-allocator-allocation-error='']"
    },
    {
      "family": "kernel_mv_max_n",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 16384,
      "dim": 8,
      "out": 16,
      "run_s": 1.3216708629988716,
      "peak_mb": 150.0009765625
    },
    {
      "family": "kernel_mv_max_n",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 32768,
      "dim": 8,
      "out": 16,
      "run_s": 1.2851345210001455,
      "peak_mb": 154.0
    },
    {
      "family": "kernel_mv_max_n",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 49152,
      "dim": 8,
      "out": 16,
      "run_s": 1.4895770579969394,
      "peak_mb": 154.5
    },
    {
      "family": "kernel_mv_max_n",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 65536,
      "dim": 8,
      "out": 16,
      "run_s": 1.3865610170032596,
      "peak_mb": 156.0009765625
    },
    {
      "family": "kernel_mv_max_n",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 131072,
      "dim": 8,
      "out": 16,
      "run_s": 2.5915712420028285,
      "peak_mb": 180.0009765625
    },
    {
      "family": "kernel_mv_max_n",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 262144,
      "dim": 8,
      "out": 16,
      "run_s": 9.264941023000574,
      "peak_mb": 212.0009765625
    },
    {
      "family": "kernel_mv_max_n",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 524288,
      "dim": 8,
      "out": 16,
      "run_s": 48.13018161699438,
      "peak_mb": 292.0009765625
    },
    {
      "family": "kernel_mv_max_n",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 1048576,
      "dim": 8,
      "out": 16,
      "error": "timeout>90s"
    },
    {
      "family": "kernel_mv_max_n",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 16384,
      "dim": 8,
      "out": 16,
      "run_s": 0.3846019709963002,
      "peak_mb": 66.0009765625
    },
    {
      "family": "kernel_mv_max_n",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 32768,
      "dim": 8,
      "out": 16,
      "run_s": 0.3904465079976944,
      "peak_mb": 70.0
    },
    {
      "family": "kernel_mv_max_n",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 49152,
      "dim": 8,
      "out": 16,
      "run_s": 0.4127722150005866,
      "peak_mb": 70.5
    },
    {
      "family": "kernel_mv_max_n",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 65536,
      "dim": 8,
      "out": 16,
      "run_s": 0.43678081699908944,
      "peak_mb": 72.0009765625
    },
    {
      "family": "kernel_mv_max_n",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 131072,
      "dim": 8,
      "out": 16,
      "run_s": 0.6083057710056892,
      "peak_mb": 64.001220703125
    },
    {
      "family": "kernel_mv_max_n",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 262144,
      "dim": 8,
      "out": 16,
      "run_s": 1.2779736919983407,
      "peak_mb": 80.00146484375
    },
    {
      "family": "kernel_mv_max_n",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 524288,
      "dim": 8,
      "out": 16,
      "run_s": 4.005967329001578,
      "peak_mb": 160.00146484375
    },
    {
      "family": "kernel_mv_max_n",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 1048576,
      "dim": 8,
      "out": 16,
      "run_s": 15.065956498998275,
      "peak_mb": 320.00146484375
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 1,
      "block_q": 64,
      "block_k": 64,
      "block_o": 64,
      "xla_temp_mb": 0.25,
      "xla_arg_mb": 0.2656288146972656,
      "xla_out_mb": 0.015625,
      "compile_s": 0.243990795002901,
      "inner": 68,
      "median_ms": 0.1831170441565679,
      "q1_ms": 0.13764830146101303,
      "q3_ms": 0.18870415442506783
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 1,
      "block_q": 64,
      "block_k": 64,
      "block_o": 16,
      "xla_temp_mb": 0.25,
      "xla_arg_mb": 0.2656288146972656,
      "xla_out_mb": 0.015625,
      "compile_s": 0.25262843500240706,
      "inner": 174,
      "median_ms": 0.09080505170176963,
      "q1_ms": 0.09077369826230296,
      "q3_ms": 0.09091776722005768
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 1,
      "block_q": 128,
      "block_k": 64,
      "block_o": 16,
      "xla_temp_mb": 0.25,
      "xla_arg_mb": 0.2656288146972656,
      "xla_out_mb": 0.015625,
      "compile_s": 0.3856469000020297,
      "inner": 25,
      "median_ms": 0.12134340009652078,
      "q1_ms": 0.12116646001231857,
      "q3_ms": 0.12211484005092643
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 1,
      "block_q": 64,
      "block_k": 128,
      "block_o": 16,
      "xla_temp_mb": 0.25,
      "xla_arg_mb": 0.2656288146972656,
      "xla_out_mb": 0.015625,
      "compile_s": 0.3512541330055683,
      "inner": 200,
      "median_ms": 0.08238634996814653,
      "q1_ms": 0.07807637748555862,
      "q3_ms": 0.08472569748846581
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 1,
      "block_q": 128,
      "block_k": 128,
      "block_o": 16,
      "xla_temp_mb": 0.25,
      "xla_arg_mb": 0.2656288146972656,
      "xla_out_mb": 0.015625,
      "compile_s": 0.803415185000631,
      "inner": 34,
      "median_ms": 0.1614487352044842,
      "q1_ms": 0.14706154401852364,
      "q3_ms": 0.21026013222153245
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 1,
      "block_q": 32,
      "block_k": 64,
      "block_o": 16,
      "xla_temp_mb": 0.25,
      "xla_arg_mb": 0.2656288146972656,
      "xla_out_mb": 0.015625,
      "compile_s": 0.2807822179966024,
      "inner": 189,
      "median_ms": 0.10623225927471168,
      "q1_ms": 0.10603662434655547,
      "q3_ms": 0.11427020369588506
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 1,
      "block_q": 64,
      "block_k": 64,
      "block_o": 64,
      "xla_temp_mb": 0.2500152587890625,
      "xla_arg_mb": 0.2656288146972656,
      "xla_out_mb": 0.2656593322753906,
      "compile_s": 0.5680364810032188,
      "inner": 92,
      "median_ms": 0.40002863043691433,
      "q1_ms": 0.3988224891303537,
      "q3_ms": 0.4002885217509491
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 1,
      "block_q": 64,
      "block_k": 64,
      "block_o": 16,
      "xla_temp_mb": 0.2500152587890625,
      "xla_arg_mb": 0.2656288146972656,
      "xla_out_mb": 0.2656593322753906,
      "compile_s": 0.5146311310018064,
      "inner": 102,
      "median_ms": 0.3553389117359092,
      "q1_ms": 0.35334268624651216,
      "q3_ms": 0.35670487255522265
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 1,
      "block_q": 128,
      "block_k": 64,
      "block_o": 16,
      "xla_temp_mb": 0.2500152587890625,
      "xla_arg_mb": 0.2656288146972656,
      "xla_out_mb": 0.2656593322753906,
      "compile_s": 0.877353583004151,
      "inner": 88,
      "median_ms": 0.41764955684894545,
      "q1_ms": 0.415844488667393,
      "q3_ms": 0.42718912498623296
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 1,
      "block_q": 64,
      "block_k": 128,
      "block_o": 16,
      "xla_temp_mb": 0.2500152587890625,
      "xla_arg_mb": 0.2656288146972656,
      "xla_out_mb": 0.2656593322753906,
      "compile_s": 0.9239612619930995,
      "inner": 75,
      "median_ms": 0.5416388799979662,
      "q1_ms": 0.5409003533228922,
      "q3_ms": 0.5501194400130771
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 1,
      "block_q": 128,
      "block_k": 128,
      "block_o": 16,
      "xla_temp_mb": 0.2500152587890625,
      "xla_arg_mb": 0.2656288146972656,
      "xla_out_mb": 0.2656593322753906,
      "compile_s": 2.581584212995949,
      "inner": 33,
      "median_ms": 1.0008004544281655,
      "q1_ms": 0.9834842423901358,
      "q3_ms": 1.01237384837022
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 1,
      "block_q": 32,
      "block_k": 64,
      "block_o": 16,
      "xla_temp_mb": 0.2500152587890625,
      "xla_arg_mb": 0.2656288146972656,
      "xla_out_mb": 0.2656593322753906,
      "compile_s": 0.5614197820032132,
      "inner": 60,
      "median_ms": 0.65984658334249,
      "q1_ms": 0.6575860583325266,
      "q3_ms": 0.6810292999944068
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 16,
      "block_q": 64,
      "block_k": 64,
      "block_o": 64,
      "xla_temp_mb": 0.0,
      "xla_arg_mb": 0.5000038146972656,
      "xla_out_mb": 0.25,
      "compile_s": 0.35367298500204924,
      "inner": 200,
      "median_ms": 0.1655940499767894,
      "q1_ms": 0.1423821174830664,
      "q3_ms": 0.18724835998000344
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 16,
      "block_q": 64,
      "block_k": 64,
      "block_o": 16,
      "xla_temp_mb": 0.0,
      "xla_arg_mb": 0.5000038146972656,
      "xla_out_mb": 0.25,
      "compile_s": 0.32944712699827505,
      "inner": 24,
      "median_ms": 0.1972633335753926,
      "q1_ms": 0.19163456257350237,
      "q3_ms": 0.3655656669252494
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 16,
      "block_q": 128,
      "block_k": 64,
      "block_o": 16,
      "xla_temp_mb": 0.0,
      "xla_arg_mb": 0.5000038146972656,
      "xla_out_mb": 0.25,
      "compile_s": 0.7698296730013681,
      "inner": 132,
      "median_ms": 0.1466270454825876,
      "q1_ms": 0.14571782577893994,
      "q3_ms": 0.14901026136640488
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 16,
      "block_q": 64,
      "block_k": 128,
      "block_o": 16,
      "xla_temp_mb": 0.0,
      "xla_arg_mb": 0.5000038146972656,
      "xla_out_mb": 0.25,
      "compile_s": 0.53755132000515,
      "inner": 200,
      "median_ms": 0.08851112503180047,
      "q1_ms": 0.0882044975151075,
      "q3_ms": 0.08953248001489555
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 16,
      "block_q": 128,
      "block_k": 128,
      "block_o": 16,
      "xla_temp_mb": 0.0,
      "xla_arg_mb": 0.5000038146972656,
      "xla_out_mb": 0.25,
      "compile_s": 0.8756937250000192,
      "inner": 148,
      "median_ms": 0.15244508107866137,
      "q1_ms": 0.14216019932260876,
      "q3_ms": 0.15251081419370854
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 16,
      "block_q": 32,
      "block_k": 64,
      "block_o": 16,
      "xla_temp_mb": 0.0,
      "xla_arg_mb": 0.5000038146972656,
      "xla_out_mb": 0.25,
      "compile_s": 0.273659452002903,
      "inner": 13,
      "median_ms": 0.3443571538530075,
      "q1_ms": 0.24179442306586468,
      "q3_ms": 0.38551573071834777
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 16,
      "block_q": 64,
      "block_k": 64,
      "block_o": 64,
      "xla_temp_mb": 0.2500152587890625,
      "xla_arg_mb": 0.5000038146972656,
      "xla_out_mb": 0.5000343322753906,
      "compile_s": 0.5541003040052601,
      "inner": 83,
      "median_ms": 0.46112113253561304,
      "q1_ms": 0.4608521867558887,
      "q3_ms": 0.4703774276829089
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 16,
      "block_q": 64,
      "block_k": 64,
      "block_o": 16,
      "xla_temp_mb": 0.2500152587890625,
      "xla_arg_mb": 0.5000038146972656,
      "xla_out_mb": 0.5000343322753906,
      "compile_s": 0.5530782030036789,
      "inner": 70,
      "median_ms": 0.4554100856434421,
      "q1_ms": 0.4542809499980649,
      "q3_ms": 0.45545617140305694
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 16,
      "block_q": 128,
      "block_k": 64,
      "block_o": 16,
      "xla_temp_mb": 0.2500152587890625,
      "xla_arg_mb": 0.5000038146972656,
      "xla_out_mb": 0.5000343322753906,
      "compile_s": 0.955414613999892,
      "inner": 67,
      "median_ms": 0.6464873880379027,
      "q1_ms": 0.6354356716398268,
      "q3_ms": 0.6466707164324375
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 16,
      "block_q": 64,
      "block_k": 128,
      "block_o": 16,
      "xla_temp_mb": 0.2500152587890625,
      "xla_arg_mb": 0.5000038146972656,
      "xla_out_mb": 0.5000343322753906,
      "compile_s": 0.9517698629933875,
      "inner": 47,
      "median_ms": 0.6972728510748872,
      "q1_ms": 0.6897309361560706,
      "q3_ms": 0.703659382936089
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 16,
      "block_q": 128,
      "block_k": 128,
      "block_o": 16,
      "xla_temp_mb": 0.2500152587890625,
      "xla_arg_mb": 0.5000038146972656,
      "xla_out_mb": 0.5000343322753906,
      "compile_s": 2.658317283996439,
      "inner": 29,
      "median_ms": 1.3417465173727672,
      "q1_ms": 1.305077706939332,
      "q3_ms": 1.34437234486666
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 16,
      "block_q": 32,
      "block_k": 64,
      "block_o": 16,
      "xla_temp_mb": 0.2500152587890625,
      "xla_arg_mb": 0.5000038146972656,
      "xla_out_mb": 0.5000343322753906,
      "compile_s": 0.6683369449965539,
      "inner": 62,
      "median_ms": 0.7148840482958833,
      "q1_ms": 0.693349459623043,
      "q3_ms": 0.7149306612489219
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd",
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 1,
      "block_q": 64,
      "block_k": 64,
      "block_o": 64,
      "xla_temp_mb": 4.0,
      "xla_arg_mb": 4.250003814697266,
      "xla_out_mb": 0.25,
      "compile_s": 0.22082907799631357,
      "inner": 6,
      "median_ms": 9.09193449964126,
      "q1_ms": 8.835856249788776,
      "q3_ms": 9.123530083040048
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd",
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 1,
      "block_q": 64,
      "block_k": 64,
      "block_o": 16,
      "xla_temp_mb": 4.0,
      "xla_arg_mb": 4.250003814697266,
      "xla_out_mb": 0.25,
      "compile_s": 0.3018269930034876,
      "inner": 6,
      "median_ms": 9.68660049935958,
      "q1_ms": 9.396352666347713,
      "q3_ms": 9.994473333184335
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd",
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 1,
      "block_q": 128,
      "block_k": 64,
      "block_o": 16,
      "xla_temp_mb": 4.0,
      "xla_arg_mb": 4.250003814697266,
      "xla_out_mb": 0.25,
      "compile_s": 0.36664484799985075,
      "inner": 5,
      "median_ms": 8.92058379977243,
      "q1_ms": 8.718846600095276,
      "q3_ms": 9.15357240010053
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd",
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 1,
      "block_q": 64,
      "block_k": 128,
      "block_o": 16,
      "xla_temp_mb": 4.0,
      "xla_arg_mb": 4.250003814697266,
      "xla_out_mb": 0.25,
      "compile_s": 0.3211142340005608,
      "inner": 5,
      "median_ms": 10.279679400264286,
      "q1_ms": 10.182616900419816,
      "q3_ms": 10.564805300236912
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd",
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 1,
      "block_q": 128,
      "block_k": 128,
      "block_o": 16,
      "xla_temp_mb": 4.0,
      "xla_arg_mb": 4.250003814697266,
      "xla_out_mb": 0.25,
      "compile_s": 0.7046347739960765,
      "inner": 3,
      "median_ms": 9.338063333416358,
      "q1_ms": 9.155246001076495,
      "q3_ms": 10.008524165944738
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd",
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 1,
      "block_q": 32,
      "block_k": 64,
      "block_o": 16,
      "xla_temp_mb": 4.0,
      "xla_arg_mb": 4.250003814697266,
      "xla_out_mb": 0.25,
      "compile_s": 0.2554624620024697,
      "inner": 3,
      "median_ms": 17.111241334835842,
      "q1_ms": 16.897432167752413,
      "q3_ms": 17.41520783374047
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 1,
      "block_q": 64,
      "block_k": 64,
      "block_o": 64,
      "xla_temp_mb": 4.0000152587890625,
      "xla_arg_mb": 4.250003814697266,
      "xla_out_mb": 4.250034332275391,
      "compile_s": 0.7140253570032655,
      "inner": 1,
      "median_ms": 57.887032999133226,
      "q1_ms": 57.84703199969954,
      "q3_ms": 57.99495699830004
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 1,
      "block_q": 64,
      "block_k": 64,
      "block_o": 16,
      "xla_temp_mb": 4.0000152587890625,
      "xla_arg_mb": 4.250003814697266,
      "xla_out_mb": 4.250034332275391,
      "compile_s": 0.5526111689978279,
      "inner": 1,
      "median_ms": 56.2818789985613,
      "q1_ms": 56.00263599990285,
      "q3_ms": 57.239428999309894
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 1,
      "block_q": 128,
      "block_k": 64,
      "block_o": 16,
      "xla_temp_mb": 4.0000152587890625,
      "xla_arg_mb": 4.250003814697266,
      "xla_out_mb": 4.250034332275391,
      "compile_s": 0.9022822589977295,
      "inner": 1,
      "median_ms": 56.02409200218972,
      "q1_ms": 55.85923000035109,
      "q3_ms": 56.45544700382743
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 1,
      "block_q": 64,
      "block_k": 128,
      "block_o": 16,
      "xla_temp_mb": 4.0000152587890625,
      "xla_arg_mb": 4.250003814697266,
      "xla_out_mb": 4.250034332275391,
      "compile_s": 1.1590162740030792,
      "inner": 1,
      "median_ms": 64.7824789994047,
      "q1_ms": 61.7382689997612,
      "q3_ms": 68.20938750024652
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 1,
      "block_q": 128,
      "block_k": 128,
      "block_o": 16,
      "xla_temp_mb": 4.0000152587890625,
      "xla_arg_mb": 4.250003814697266,
      "xla_out_mb": 4.250034332275391,
      "compile_s": 3.377689870001632,
      "inner": 1,
      "median_ms": 194.26277599995956,
      "q1_ms": 193.07714449678315,
      "q3_ms": 194.9061624982278
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 1,
      "block_q": 32,
      "block_k": 64,
      "block_o": 16,
      "xla_temp_mb": 4.0000152587890625,
      "xla_arg_mb": 4.250003814697266,
      "xla_out_mb": 4.250034332275391,
      "compile_s": 0.718991534995439,
      "inner": 1,
      "median_ms": 100.33669199765427,
      "q1_ms": 99.63772599803633,
      "q3_ms": 103.74242550096824
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd",
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 16,
      "block_q": 64,
      "block_k": 64,
      "block_o": 64,
      "xla_temp_mb": 0.0,
      "xla_arg_mb": 8.000003814697266,
      "xla_out_mb": 4.0,
      "compile_s": 0.24072980000346433,
      "inner": 4,
      "median_ms": 12.036193749736412,
      "q1_ms": 11.56679149971751,
      "q3_ms": 12.041398749715881
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd",
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 16,
      "block_q": 64,
      "block_k": 64,
      "block_o": 16,
      "xla_temp_mb": 0.0,
      "xla_arg_mb": 8.000003814697266,
      "xla_out_mb": 4.0,
      "compile_s": 0.23812341299344553,
      "inner": 4,
      "median_ms": 10.851652001292678,
      "q1_ms": 10.573689000011655,
      "q3_ms": 11.353052375852712
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd",
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 16,
      "block_q": 128,
      "block_k": 64,
      "block_o": 16,
      "xla_temp_mb": 0.0,
      "xla_arg_mb": 8.000003814697266,
      "xla_out_mb": 4.0,
      "compile_s": 0.4151097279973328,
      "inner": 4,
      "median_ms": 10.76340299914591,
      "q1_ms": 10.629857873936999,
      "q3_ms": 10.84290074868477
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd",
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 16,
      "block_q": 64,
      "block_k": 128,
      "block_o": 16,
      "xla_temp_mb": 0.0,
      "xla_arg_mb": 8.000003814697266,
      "xla_out_mb": 4.0,
      "compile_s": 0.4224443050043192,
      "inner": 4,
      "median_ms": 12.514024749179953,
      "q1_ms": 11.74296587396384,
      "q3_ms": 12.928066249514814
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd",
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 16,
      "block_q": 128,
      "block_k": 128,
      "block_o": 16,
      "xla_temp_mb": 0.0,
      "xla_arg_mb": 8.000003814697266,
      "xla_out_mb": 4.0,
      "compile_s": 0.8420908890038845,
      "inner": 4,
      "median_ms": 12.177576250905986,
      "q1_ms": 11.739528124962817,
      "q3_ms": 12.3822206260229
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd",
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 16,
      "block_q": 32,
      "block_k": 64,
      "block_o": 16,
      "xla_temp_mb": 0.0,
      "xla_arg_mb": 8.000003814697266,
      "xla_out_mb": 4.0,
      "compile_s": 0.25955236300069373,
      "inner": 1,
      "median_ms": 20.62446000491036,
      "q1_ms": 20.512563503871206,
      "q3_ms": 21.51910600150586
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 16,
      "block_q": 64,
      "block_k": 64,
      "block_o": 64,
      "xla_temp_mb": 4.0000152587890625,
      "xla_arg_mb": 8.000003814697266,
      "xla_out_mb": 8.00003433227539,
      "compile_s": 0.6224228360006236,
      "inner": 1,
      "median_ms": 59.980391000863165,
      "q1_ms": 58.920457002386684,
      "q3_ms": 60.49141649782541
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 16,
      "block_q": 64,
      "block_k": 64,
      "block_o": 16,
      "xla_temp_mb": 4.0000152587890625,
      "xla_arg_mb": 8.000003814697266,
      "xla_out_mb": 8.00003433227539,
      "compile_s": 0.4942585459939437,
      "inner": 1,
      "median_ms": 60.82332399819279,
      "q1_ms": 60.08553900028346,
      "q3_ms": 60.82420099846786
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 16,
      "block_q": 128,
      "block_k": 64,
      "block_o": 16,
      "xla_temp_mb": 4.0000152587890625,
      "xla_arg_mb": 8.000003814697266,
      "xla_out_mb": 8.00003433227539,
      "compile_s": 1.14768097599881,
      "inner": 1,
      "median_ms": 62.42619400291005,
      "q1_ms": 61.35953250122839,
      "q3_ms": 65.19113649846986
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 16,
      "block_q": 64,
      "block_k": 128,
      "block_o": 16,
      "xla_temp_mb": 4.0000152587890625,
      "xla_arg_mb": 8.000003814697266,
      "xla_out_mb": 8.00003433227539,
      "compile_s": 1.2279782039986458,
      "inner": 1,
      "median_ms": 70.07693700143136,
      "q1_ms": 68.14534600198385,
      "q3_ms": 70.30901100006304
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 16,
      "block_q": 128,
      "block_k": 128,
      "block_o": 16,
      "xla_temp_mb": 4.0000152587890625,
      "xla_arg_mb": 8.000003814697266,
      "xla_out_mb": 8.00003433227539,
      "compile_s": 2.1924542389970156,
      "inner": 1,
      "median_ms": 196.79153699689778,
      "q1_ms": 196.32788550006808,
      "q3_ms": 200.486206496862
    },
    {
      "family": "kernel_mv_tiles",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 16,
      "block_q": 32,
      "block_k": 64,
      "block_o": 16,
      "xla_temp_mb": 4.0000152587890625,
      "xla_arg_mb": 8.000003814697266,
      "xla_out_mb": 8.00003433227539,
      "compile_s": 1.5800272490014322,
      "inner": 1,
      "median_ms": 131.57741200120654,
      "q1_ms": 130.15517799794907,
      "q3_ms": 137.86151699969196
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 0,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0432,
      "q1_ms": 0.0431,
      "q3_ms": 0.0449
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 0,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0426,
      "q1_ms": 0.0426,
      "q3_ms": 0.043
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0583,
      "q1_ms": 0.0572,
      "q3_ms": 0.0583
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0621,
      "q1_ms": 0.0619,
      "q3_ms": 0.0621
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 0,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 16,
      "median_ms": 0.047,
      "q1_ms": 0.0467,
      "q3_ms": 0.047
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 0,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0402,
      "q1_ms": 0.04,
      "q3_ms": 0.0407
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 16,
      "median_ms": 0.056,
      "q1_ms": 0.056,
      "q3_ms": 0.0561
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 16,
      "median_ms": 0.065,
      "q1_ms": 0.0649,
      "q3_ms": 0.0651
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 0,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0487,
      "q1_ms": 0.0485,
      "q3_ms": 0.0532
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 0,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0362,
      "q1_ms": 0.0341,
      "q3_ms": 0.0367
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0546,
      "q1_ms": 0.053,
      "q3_ms": 0.0552
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 32,
      "out": 16,
      "median_ms": 0.1148,
      "q1_ms": 0.1148,
      "q3_ms": 0.1148
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 0,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0451,
      "q1_ms": 0.045,
      "q3_ms": 0.0455
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 0,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0599,
      "q1_ms": 0.0599,
      "q3_ms": 0.0601
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0585,
      "q1_ms": 0.0585,
      "q3_ms": 0.0586
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "median_ms": 0.1028,
      "q1_ms": 0.1028,
      "q3_ms": 0.1028
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 0,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "median_ms": 0.044,
      "q1_ms": 0.0439,
      "q3_ms": 0.044
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 0,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0394,
      "q1_ms": 0.039,
      "q3_ms": 0.0399
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0816,
      "q1_ms": 0.0816,
      "q3_ms": 0.0816
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "median_ms": 0.118,
      "q1_ms": 0.1178,
      "q3_ms": 0.118
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 0,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0471,
      "q1_ms": 0.047,
      "q3_ms": 0.0477
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 0,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0465,
      "q1_ms": 0.0465,
      "q3_ms": 0.0466
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0789,
      "q1_ms": 0.0788,
      "q3_ms": 0.079
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "median_ms": 0.2164,
      "q1_ms": 0.215,
      "q3_ms": 0.2165
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 0,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0534,
      "q1_ms": 0.0463,
      "q3_ms": 0.0547
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 0,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0374,
      "q1_ms": 0.0374,
      "q3_ms": 0.0375
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 8,
      "out": 1,
      "median_ms": 0.1479,
      "q1_ms": 0.1478,
      "q3_ms": 0.1481
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 8,
      "out": 1,
      "median_ms": 0.1372,
      "q1_ms": 0.1372,
      "q3_ms": 0.1373
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 0,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0507,
      "q1_ms": 0.0506,
      "q3_ms": 0.0508
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 0,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0416,
      "q1_ms": 0.0415,
      "q3_ms": 0.042
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 8,
      "out": 16,
      "median_ms": 0.2213,
      "q1_ms": 0.2212,
      "q3_ms": 0.2214
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 8,
      "out": 16,
      "median_ms": 0.1531,
      "q1_ms": 0.153,
      "q3_ms": 0.1531
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 0,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0526,
      "q1_ms": 0.0526,
      "q3_ms": 0.0527
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 0,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0707,
      "q1_ms": 0.0706,
      "q3_ms": 0.0707
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 32,
      "out": 16,
      "median_ms": 0.2284,
      "q1_ms": 0.2281,
      "q3_ms": 0.2285
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 32,
      "out": 16,
      "median_ms": 0.2782,
      "q1_ms": 0.2782,
      "q3_ms": 0.2782
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 0,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0517,
      "q1_ms": 0.0516,
      "q3_ms": 0.0518
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 0,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 8,
      "out": 1,
      "median_ms": 0.045,
      "q1_ms": 0.044,
      "q3_ms": 0.0491
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 8,
      "out": 1,
      "median_ms": 0.2183,
      "q1_ms": 0.2183,
      "q3_ms": 0.2183
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 8,
      "out": 1,
      "median_ms": 0.1815,
      "q1_ms": 0.1815,
      "q3_ms": 0.1816
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 0,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0923,
      "q1_ms": 0.092,
      "q3_ms": 0.0923
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 0,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0394,
      "q1_ms": 0.0394,
      "q3_ms": 0.0394
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 8,
      "out": 16,
      "median_ms": 0.2874,
      "q1_ms": 0.2874,
      "q3_ms": 0.2875
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 8,
      "out": 16,
      "median_ms": 0.199,
      "q1_ms": 0.199,
      "q3_ms": 0.199
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 0,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0953,
      "q1_ms": 0.0951,
      "q3_ms": 0.0953
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 0,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0897,
      "q1_ms": 0.0897,
      "q3_ms": 0.0898
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 32,
      "out": 16,
      "median_ms": 0.297,
      "q1_ms": 0.2968,
      "q3_ms": 0.2971
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 32,
      "out": 16,
      "median_ms": 0.3719,
      "q1_ms": 0.3719,
      "q3_ms": 0.3719
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 0,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0475,
      "q1_ms": 0.0473,
      "q3_ms": 0.0476
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 0,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0434,
      "q1_ms": 0.0433,
      "q3_ms": 0.0439
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0569,
      "q1_ms": 0.0567,
      "q3_ms": 0.0571
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0556,
      "q1_ms": 0.0555,
      "q3_ms": 0.0559
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 0,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0479,
      "q1_ms": 0.0479,
      "q3_ms": 0.0487
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 0,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0442,
      "q1_ms": 0.0416,
      "q3_ms": 0.0474
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0819,
      "q1_ms": 0.0819,
      "q3_ms": 0.0819
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0655,
      "q1_ms": 0.0655,
      "q3_ms": 0.0655
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 0,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0437,
      "q1_ms": 0.0435,
      "q3_ms": 0.0439
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 0,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0344,
      "q1_ms": 0.0338,
      "q3_ms": 0.0344
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0919,
      "q1_ms": 0.0919,
      "q3_ms": 0.0919
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 32,
      "out": 16,
      "median_ms": 0.102,
      "q1_ms": 0.102,
      "q3_ms": 0.102
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 0,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0516,
      "q1_ms": 0.0514,
      "q3_ms": 0.0516
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 0,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0473,
      "q1_ms": 0.047,
      "q3_ms": 0.0474
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "median_ms": 0.1804,
      "q1_ms": 0.1802,
      "q3_ms": 0.1804
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0975,
      "q1_ms": 0.0974,
      "q3_ms": 0.0975
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 0,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0915,
      "q1_ms": 0.0913,
      "q3_ms": 0.0915
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 0,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0371,
      "q1_ms": 0.0368,
      "q3_ms": 0.0378
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "median_ms": 0.2898,
      "q1_ms": 0.288,
      "q3_ms": 0.2905
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "median_ms": 0.1054,
      "q1_ms": 0.1054,
      "q3_ms": 0.1054
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 0,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0948,
      "q1_ms": 0.0944,
      "q3_ms": 0.0948
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 0,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0477,
      "q1_ms": 0.0477,
      "q3_ms": 0.0478
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "median_ms": 0.2927,
      "q1_ms": 0.2926,
      "q3_ms": 0.2927
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 0,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "median_ms": 0.1933,
      "q1_ms": 0.1933,
      "q3_ms": 0.1934
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 1,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0453,
      "q1_ms": 0.0453,
      "q3_ms": 0.0457
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 1,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0435,
      "q1_ms": 0.0433,
      "q3_ms": 0.0435
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0557,
      "q1_ms": 0.0553,
      "q3_ms": 0.0559
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0553,
      "q1_ms": 0.0551,
      "q3_ms": 0.0557
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 1,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0443,
      "q1_ms": 0.0441,
      "q3_ms": 0.0445
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 1,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0387,
      "q1_ms": 0.0387,
      "q3_ms": 0.0387
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0542,
      "q1_ms": 0.0537,
      "q3_ms": 0.0545
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0576,
      "q1_ms": 0.0573,
      "q3_ms": 0.0579
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 1,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0509,
      "q1_ms": 0.0423,
      "q3_ms": 0.0564
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 1,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0397,
      "q1_ms": 0.0396,
      "q3_ms": 0.0398
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0556,
      "q1_ms": 0.0556,
      "q3_ms": 0.0558
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 32,
      "out": 16,
      "median_ms": 0.1148,
      "q1_ms": 0.1148,
      "q3_ms": 0.1148
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 1,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0538,
      "q1_ms": 0.0526,
      "q3_ms": 0.0539
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 1,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0461,
      "q1_ms": 0.0461,
      "q3_ms": 0.0484
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0558,
      "q1_ms": 0.0558,
      "q3_ms": 0.0558
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0962,
      "q1_ms": 0.0962,
      "q3_ms": 0.0962
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 1,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0462,
      "q1_ms": 0.0461,
      "q3_ms": 0.0464
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 1,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0439,
      "q1_ms": 0.0433,
      "q3_ms": 0.0445
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0757,
      "q1_ms": 0.0757,
      "q3_ms": 0.0758
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "median_ms": 0.1034,
      "q1_ms": 0.1034,
      "q3_ms": 0.1035
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 1,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0464,
      "q1_ms": 0.0449,
      "q3_ms": 0.0499
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 1,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0466,
      "q1_ms": 0.0465,
      "q3_ms": 0.0467
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0791,
      "q1_ms": 0.0791,
      "q3_ms": 0.0791
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "median_ms": 0.1901,
      "q1_ms": 0.1901,
      "q3_ms": 0.1901
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 1,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0425,
      "q1_ms": 0.0413,
      "q3_ms": 0.0426
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 1,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0382,
      "q1_ms": 0.038,
      "q3_ms": 0.0386
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 8,
      "out": 1,
      "median_ms": 0.1479,
      "q1_ms": 0.1476,
      "q3_ms": 0.1481
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 8,
      "out": 1,
      "median_ms": 0.1373,
      "q1_ms": 0.1373,
      "q3_ms": 0.1373
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 1,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0506,
      "q1_ms": 0.0506,
      "q3_ms": 0.0507
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 1,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0438,
      "q1_ms": 0.0419,
      "q3_ms": 0.044
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 8,
      "out": 16,
      "median_ms": 0.2212,
      "q1_ms": 0.2212,
      "q3_ms": 0.2213
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 8,
      "out": 16,
      "median_ms": 0.153,
      "q1_ms": 0.153,
      "q3_ms": 0.153
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 1,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 32,
      "out": 16,
      "median_ms": 0.053,
      "q1_ms": 0.0527,
      "q3_ms": 0.0533
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 1,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0701,
      "q1_ms": 0.0701,
      "q3_ms": 0.0702
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 32,
      "out": 16,
      "median_ms": 0.2284,
      "q1_ms": 0.2281,
      "q3_ms": 0.2286
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 32,
      "out": 16,
      "median_ms": 0.2782,
      "q1_ms": 0.2782,
      "q3_ms": 0.2782
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 1,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0512,
      "q1_ms": 0.0512,
      "q3_ms": 0.0512
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 1,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0436,
      "q1_ms": 0.0435,
      "q3_ms": 0.0436
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 8,
      "out": 1,
      "median_ms": 0.2183,
      "q1_ms": 0.2183,
      "q3_ms": 0.2184
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 8,
      "out": 1,
      "median_ms": 0.182,
      "q1_ms": 0.182,
      "q3_ms": 0.1821
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 1,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0923,
      "q1_ms": 0.0923,
      "q3_ms": 0.0924
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 1,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0409,
      "q1_ms": 0.0396,
      "q3_ms": 0.0442
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 8,
      "out": 16,
      "median_ms": 0.2873,
      "q1_ms": 0.2872,
      "q3_ms": 0.2873
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 8,
      "out": 16,
      "median_ms": 0.1989,
      "q1_ms": 0.1989,
      "q3_ms": 0.199
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 1,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0954,
      "q1_ms": 0.0951,
      "q3_ms": 0.0954
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 1,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0898,
      "q1_ms": 0.0898,
      "q3_ms": 0.0898
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 32,
      "out": 16,
      "median_ms": 0.2971,
      "q1_ms": 0.297,
      "q3_ms": 0.2971
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 32,
      "out": 16,
      "median_ms": 0.3718,
      "q1_ms": 0.3718,
      "q3_ms": 0.3719
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 1,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0474,
      "q1_ms": 0.0471,
      "q3_ms": 0.048
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 1,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0448,
      "q1_ms": 0.0446,
      "q3_ms": 0.0449
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0563,
      "q1_ms": 0.0563,
      "q3_ms": 0.0563
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0553,
      "q1_ms": 0.0552,
      "q3_ms": 0.0554
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 1,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 16,
      "median_ms": 0.048,
      "q1_ms": 0.0474,
      "q3_ms": 0.0485
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 1,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0388,
      "q1_ms": 0.0387,
      "q3_ms": 0.0396
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0766,
      "q1_ms": 0.0766,
      "q3_ms": 0.0767
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0654,
      "q1_ms": 0.0654,
      "q3_ms": 0.0655
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 1,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0446,
      "q1_ms": 0.044,
      "q3_ms": 0.0447
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 1,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0339,
      "q1_ms": 0.0335,
      "q3_ms": 0.0345
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0919,
      "q1_ms": 0.0919,
      "q3_ms": 0.0919
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 32,
      "out": 16,
      "median_ms": 0.1021,
      "q1_ms": 0.102,
      "q3_ms": 0.1021
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 1,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0519,
      "q1_ms": 0.0518,
      "q3_ms": 0.052
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 1,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0484,
      "q1_ms": 0.0453,
      "q3_ms": 0.049
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "median_ms": 0.1782,
      "q1_ms": 0.1779,
      "q3_ms": 0.1783
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0981,
      "q1_ms": 0.0981,
      "q3_ms": 0.0981
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 1,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0915,
      "q1_ms": 0.0914,
      "q3_ms": 0.0915
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 1,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0371,
      "q1_ms": 0.0368,
      "q3_ms": 0.0397
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "median_ms": 0.2873,
      "q1_ms": 0.2873,
      "q3_ms": 0.2874
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "median_ms": 0.106,
      "q1_ms": 0.106,
      "q3_ms": 0.106
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 1,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0948,
      "q1_ms": 0.0946,
      "q3_ms": 0.0949
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 1,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0477,
      "q1_ms": 0.0477,
      "q3_ms": 0.0477
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "median_ms": 0.2931,
      "q1_ms": 0.293,
      "q3_ms": 0.2932
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 1,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "median_ms": 0.1933,
      "q1_ms": 0.1933,
      "q3_ms": 0.1934
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 2,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0476,
      "q1_ms": 0.0458,
      "q3_ms": 0.0526
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 2,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0434,
      "q1_ms": 0.0433,
      "q3_ms": 0.0435
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0565,
      "q1_ms": 0.0558,
      "q3_ms": 0.0565
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0551,
      "q1_ms": 0.0551,
      "q3_ms": 0.0551
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 2,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 16,
      "median_ms": 0.044,
      "q1_ms": 0.0432,
      "q3_ms": 0.0452
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 2,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0431,
      "q1_ms": 0.043,
      "q3_ms": 0.0528
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0682,
      "q1_ms": 0.066,
      "q3_ms": 0.0682
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0644,
      "q1_ms": 0.0635,
      "q3_ms": 0.0645
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 2,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0473,
      "q1_ms": 0.046,
      "q3_ms": 0.0478
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 2,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0353,
      "q1_ms": 0.0347,
      "q3_ms": 0.0364
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0577,
      "q1_ms": 0.0558,
      "q3_ms": 0.0584
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 1,
      "n": 512,
      "m": 512,
      "dim": 32,
      "out": 16,
      "median_ms": 0.1146,
      "q1_ms": 0.1146,
      "q3_ms": 0.1148
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 2,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0491,
      "q1_ms": 0.0488,
      "q3_ms": 0.0491
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 2,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0461,
      "q1_ms": 0.0461,
      "q3_ms": 0.0467
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0561,
      "q1_ms": 0.056,
      "q3_ms": 0.0561
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0964,
      "q1_ms": 0.0964,
      "q3_ms": 0.0965
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 2,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0439,
      "q1_ms": 0.0429,
      "q3_ms": 0.0449
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 2,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0418,
      "q1_ms": 0.0415,
      "q3_ms": 0.042
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0755,
      "q1_ms": 0.0755,
      "q3_ms": 0.0755
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "median_ms": 0.1034,
      "q1_ms": 0.1033,
      "q3_ms": 0.1034
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 2,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0436,
      "q1_ms": 0.0433,
      "q3_ms": 0.0436
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 2,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0466,
      "q1_ms": 0.0466,
      "q3_ms": 0.0467
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0792,
      "q1_ms": 0.0791,
      "q3_ms": 0.0792
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "median_ms": 0.1904,
      "q1_ms": 0.1904,
      "q3_ms": 0.1904
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 2,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0389,
      "q1_ms": 0.0384,
      "q3_ms": 0.0391
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 2,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0366,
      "q1_ms": 0.0365,
      "q3_ms": 0.0367
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 8,
      "out": 1,
      "median_ms": 0.148,
      "q1_ms": 0.148,
      "q3_ms": 0.1483
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 8,
      "out": 1,
      "median_ms": 0.1364,
      "q1_ms": 0.1364,
      "q3_ms": 0.1364
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 2,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 8,
      "out": 16,
      "median_ms": 0.051,
      "q1_ms": 0.0508,
      "q3_ms": 0.0512
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 2,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0495,
      "q1_ms": 0.0417,
      "q3_ms": 0.0502
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 8,
      "out": 16,
      "median_ms": 0.2215,
      "q1_ms": 0.2213,
      "q3_ms": 0.2216
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 8,
      "out": 16,
      "median_ms": 0.1531,
      "q1_ms": 0.1531,
      "q3_ms": 0.1531
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 2,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0532,
      "q1_ms": 0.0531,
      "q3_ms": 0.0534
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 2,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0707,
      "q1_ms": 0.0707,
      "q3_ms": 0.0707
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 32,
      "out": 16,
      "median_ms": 0.2286,
      "q1_ms": 0.2285,
      "q3_ms": 0.2287
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 1,
      "n": 1448,
      "m": 1448,
      "dim": 32,
      "out": 16,
      "median_ms": 0.2787,
      "q1_ms": 0.2787,
      "q3_ms": 0.2787
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 2,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0514,
      "q1_ms": 0.0514,
      "q3_ms": 0.0515
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 2,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0439,
      "q1_ms": 0.0437,
      "q3_ms": 0.0463
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 8,
      "out": 1,
      "median_ms": 0.2184,
      "q1_ms": 0.2184,
      "q3_ms": 0.2184
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 8,
      "out": 1,
      "median_ms": 0.1818,
      "q1_ms": 0.1818,
      "q3_ms": 0.1818
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 2,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0925,
      "q1_ms": 0.0925,
      "q3_ms": 0.0925
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 2,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0398,
      "q1_ms": 0.0398,
      "q3_ms": 0.0405
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 8,
      "out": 16,
      "median_ms": 0.2873,
      "q1_ms": 0.2872,
      "q3_ms": 0.2873
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 8,
      "out": 16,
      "median_ms": 0.1989,
      "q1_ms": 0.1989,
      "q3_ms": 0.1989
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 2,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0954,
      "q1_ms": 0.0952,
      "q3_ms": 0.0954
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 2,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0898,
      "q1_ms": 0.0898,
      "q3_ms": 0.0898
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 32,
      "out": 16,
      "median_ms": 0.297,
      "q1_ms": 0.297,
      "q3_ms": 0.2971
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 1,
      "n": 2048,
      "m": 2048,
      "dim": 32,
      "out": 16,
      "median_ms": 0.372,
      "q1_ms": 0.372,
      "q3_ms": 0.3721
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 2,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0494,
      "q1_ms": 0.0491,
      "q3_ms": 0.0508
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 2,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0463,
      "q1_ms": 0.0455,
      "q3_ms": 0.0491
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 1,
      "median_ms": 0.057,
      "q1_ms": 0.057,
      "q3_ms": 0.0584
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0554,
      "q1_ms": 0.0554,
      "q3_ms": 0.0554
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 2,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0558,
      "q1_ms": 0.0544,
      "q3_ms": 0.0566
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 2,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 16,
      "median_ms": 0.046,
      "q1_ms": 0.0458,
      "q3_ms": 0.0462
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0766,
      "q1_ms": 0.0766,
      "q3_ms": 0.0767
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0654,
      "q1_ms": 0.0654,
      "q3_ms": 0.0655
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 2,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0492,
      "q1_ms": 0.0487,
      "q3_ms": 0.0492
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 2,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0386,
      "q1_ms": 0.0381,
      "q3_ms": 0.0392
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0919,
      "q1_ms": 0.0919,
      "q3_ms": 0.092
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 4,
      "n": 512,
      "m": 512,
      "dim": 32,
      "out": 16,
      "median_ms": 0.1157,
      "q1_ms": 0.1157,
      "q3_ms": 0.1157
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 2,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0514,
      "q1_ms": 0.0513,
      "q3_ms": 0.0517
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 2,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "median_ms": 0.044,
      "q1_ms": 0.0439,
      "q3_ms": 0.0442
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "median_ms": 0.178,
      "q1_ms": 0.1778,
      "q3_ms": 0.1781
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "median_ms": 0.0979,
      "q1_ms": 0.0978,
      "q3_ms": 0.0979
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 2,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0913,
      "q1_ms": 0.0911,
      "q3_ms": 0.0914
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 2,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "median_ms": 0.0374,
      "q1_ms": 0.0372,
      "q3_ms": 0.0382
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "median_ms": 0.2872,
      "q1_ms": 0.2871,
      "q3_ms": 0.2873
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "median_ms": 0.106,
      "q1_ms": 0.106,
      "q3_ms": 0.106
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd",
      "repeat": 2,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0949,
      "q1_ms": 0.0947,
      "q3_ms": 0.0949
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd",
      "repeat": 2,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "median_ms": 0.0479,
      "q1_ms": 0.0477,
      "q3_ms": 0.048
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "xla",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "median_ms": 0.2935,
      "q1_ms": 0.293,
      "q3_ms": 0.2935
    },
    {
      "family": "kernel_mv_crossover",
      "impl": "pallas",
      "mode": "fwd_bwd",
      "repeat": 2,
      "batch": 4,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "median_ms": 0.1934,
      "q1_ms": 0.1934,
      "q3_ms": 0.1935
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 0.5,
      "xla_arg_mb": 0.06641006469726562,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.31735403000493534,
      "inner": 16,
      "median_ms": 0.24304525004481548,
      "q1_ms": 0.1541049373372516,
      "q3_ms": 0.25258837467845296,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 0.7500152587890625,
      "xla_arg_mb": 0.06641006469726562,
      "xla_out_mb": 0.06644058227539062,
      "compile_s": 0.5400740559998667,
      "inner": 156,
      "median_ms": 0.20502108334487554,
      "q1_ms": 0.20147153848553828,
      "q3_ms": 0.23652670512465426,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 0.5,
      "xla_arg_mb": 0.12500381469726562,
      "xla_out_mb": 0.0625,
      "compile_s": 0.39419617100065807,
      "inner": 200,
      "median_ms": 0.04960955000569811,
      "q1_ms": 0.04958660501870327,
      "q3_ms": 0.06271446502069011,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 0.7500152587890625,
      "xla_arg_mb": 0.12500381469726562,
      "xla_out_mb": 0.12503433227539062,
      "compile_s": 0.4534760910028126,
      "inner": 176,
      "median_ms": 0.18188632384408265,
      "q1_ms": 0.18045685795765495,
      "q3_ms": 0.18404592612403742,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 0.5,
      "xla_arg_mb": 0.2539100646972656,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.36276596899551805,
      "inner": 159,
      "median_ms": 0.08471854714385334,
      "q1_ms": 0.0840997044344831,
      "q3_ms": 0.08506310062004113,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 0.7500152587890625,
      "xla_arg_mb": 0.2539100646972656,
      "xla_out_mb": 0.2539405822753906,
      "compile_s": 0.5722299619956175,
      "inner": 114,
      "median_ms": 0.33247048243841465,
      "q1_ms": 0.3300238421249097,
      "q3_ms": 0.3509256228490063,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 0.5,
      "xla_arg_mb": 0.3125038146972656,
      "xla_out_mb": 0.0625,
      "compile_s": 0.3541445820010267,
      "inner": 200,
      "median_ms": 0.07805240999005036,
      "q1_ms": 0.07773671502945945,
      "q3_ms": 0.07863466500566574,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 0.7500152587890625,
      "xla_arg_mb": 0.3125038146972656,
      "xla_out_mb": 0.3125343322753906,
      "compile_s": 0.5427536509960191,
      "inner": 133,
      "median_ms": 0.2979979398148309,
      "q1_ms": 0.29577048870916933,
      "q3_ms": 0.2995068195347737,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 2.0,
      "xla_arg_mb": 0.2656288146972656,
      "xla_out_mb": 0.015625,
      "compile_s": 0.30276296800002456,
      "inner": 165,
      "median_ms": 0.170001393946057,
      "q1_ms": 0.1649060484575757,
      "q3_ms": 0.17719131516941794,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 3.0000152587890625,
      "xla_arg_mb": 0.2656288146972656,
      "xla_out_mb": 0.2656593322753906,
      "compile_s": 0.5683020139986183,
      "inner": 62,
      "median_ms": 0.6820993064226764,
      "q1_ms": 0.6795628870613948,
      "q3_ms": 0.7014876290697665,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 2.0,
      "xla_arg_mb": 0.5000038146972656,
      "xla_out_mb": 0.25,
      "compile_s": 0.3186722060054308,
      "inner": 199,
      "median_ms": 0.1751105125581649,
      "q1_ms": 0.16339873367377636,
      "q3_ms": 0.1813295728690225,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 3.0000152587890625,
      "xla_arg_mb": 0.5000038146972656,
      "xla_out_mb": 0.5000343322753906,
      "compile_s": 0.42744728999969084,
      "inner": 64,
      "median_ms": 0.7304446875195936,
      "q1_ms": 0.7150156250190776,
      "q3_ms": 0.7629653437106754,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 2.0,
      "xla_arg_mb": 1.0156288146972656,
      "xla_out_mb": 0.015625,
      "compile_s": 0.35862007700052345,
      "inner": 135,
      "median_ms": 0.2851480741002079,
      "q1_ms": 0.2844877852045034,
      "q3_ms": 0.28537628893024736,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 3.0000152587890625,
      "xla_arg_mb": 1.0156288146972656,
      "xla_out_mb": 1.0156593322753906,
      "compile_s": 0.5526469459946384,
      "inner": 40,
      "median_ms": 1.1356894501659553,
      "q1_ms": 1.1289702499198029,
      "q3_ms": 1.1394781249691732,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 2.0,
      "xla_arg_mb": 1.2500038146972656,
      "xla_out_mb": 0.25,
      "compile_s": 0.36183155899925623,
      "inner": 135,
      "median_ms": 0.29002115555124836,
      "q1_ms": 0.288437044407517,
      "q3_ms": 0.29295642226417035,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 3.0000152587890625,
      "xla_arg_mb": 1.2500038146972656,
      "xla_out_mb": 1.2500343322753906,
      "compile_s": 0.566670152998995,
      "inner": 40,
      "median_ms": 1.2508502999480697,
      "q1_ms": 1.1497936749947257,
      "q3_ms": 1.258115575001284,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 8.0,
      "xla_arg_mb": 1.0625038146972656,
      "xla_out_mb": 0.0625,
      "compile_s": 0.37660002700431505,
      "inner": 24,
      "median_ms": 1.9426505414230633,
      "q1_ms": 1.9390979999419262,
      "q3_ms": 2.03125545825363,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 12.000015258789062,
      "xla_arg_mb": 1.0625038146972656,
      "xla_out_mb": 1.0625343322753906,
      "compile_s": 0.545857839002565,
      "inner": 5,
      "median_ms": 9.55823279946344,
      "q1_ms": 9.212840799591504,
      "q3_ms": 10.190963199420366,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 8.0,
      "xla_arg_mb": 2.0000038146972656,
      "xla_out_mb": 1.0,
      "compile_s": 0.31552606799959904,
      "inner": 24,
      "median_ms": 2.113690749865782,
      "q1_ms": 2.055633250165556,
      "q3_ms": 2.2099972917809887,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 12.000015258789062,
      "xla_arg_mb": 2.0000038146972656,
      "xla_out_mb": 2.0000343322753906,
      "compile_s": 0.47655399500217754,
      "inner": 5,
      "median_ms": 9.173904999624938,
      "q1_ms": 9.028589799709152,
      "q3_ms": 9.176582199870609,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 8.0,
      "xla_arg_mb": 4.062503814697266,
      "xla_out_mb": 0.0625,
      "compile_s": 0.49692261799646076,
      "inner": 13,
      "median_ms": 3.937934384269353,
      "q1_ms": 3.7211900769482154,
      "q3_ms": 4.463653845820003,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 12.000015258789062,
      "xla_arg_mb": 4.062503814697266,
      "xla_out_mb": 4.062534332275391,
      "compile_s": 0.6428626830020221,
      "inner": 3,
      "median_ms": 15.755173000798095,
      "q1_ms": 15.351678666775115,
      "q3_ms": 16.476479000023875,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 8.0,
      "xla_arg_mb": 5.000003814697266,
      "xla_out_mb": 1.0,
      "compile_s": 0.3926979720054078,
      "inner": 13,
      "median_ms": 4.023856384894596,
      "q1_ms": 3.829764154103871,
      "q3_ms": 4.060049538496129,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 12.000015258789062,
      "xla_arg_mb": 5.000003814697266,
      "xla_out_mb": 5.000034332275391,
      "compile_s": 0.8288730400017812,
      "inner": 3,
      "median_ms": 15.486090332463695,
      "q1_ms": 15.291661999071948,
      "q3_ms": 15.651720333456371,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 32.0,
      "xla_arg_mb": 4.250003814697266,
      "xla_out_mb": 0.25,
      "compile_s": 0.3670511840027757,
      "inner": 1,
      "median_ms": 25.72983600111911,
      "q1_ms": 25.712101996759884,
      "q3_ms": 26.09175099496497,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 48.00001525878906,
      "xla_arg_mb": 4.250003814697266,
      "xla_out_mb": 4.250034332275391,
      "compile_s": 0.5102368319930974,
      "inner": 1,
      "median_ms": 122.19814299896825,
      "q1_ms": 120.1123720020405,
      "q3_ms": 125.74128300184384,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 32.0,
      "xla_arg_mb": 8.000003814697266,
      "xla_out_mb": 4.0,
      "compile_s": 0.3623347620014101,
      "inner": 1,
      "median_ms": 26.581138001347426,
      "q1_ms": 26.4368749994901,
      "q3_ms": 26.619149997713976,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 48.00001525878906,
      "xla_arg_mb": 8.000003814697266,
      "xla_out_mb": 8.00003433227539,
      "compile_s": 0.6622893620005925,
      "inner": 1,
      "median_ms": 164.36756099574268,
      "q1_ms": 133.5887739987811,
      "q3_ms": 191.83223199797794,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 32.0,
      "xla_arg_mb": 16.250003814697266,
      "xla_out_mb": 0.25,
      "compile_s": 0.40587812299781945,
      "inner": 1,
      "median_ms": 75.35761900362559,
      "q1_ms": 72.84445899858838,
      "q3_ms": 75.44861099449918,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 48.00001525878906,
      "xla_arg_mb": 16.250003814697266,
      "xla_out_mb": 16.25003433227539,
      "compile_s": 0.5992952960004914,
      "inner": 1,
      "median_ms": 214.6220309950877,
      "q1_ms": 205.5922819999978,
      "q3_ms": 216.35272900311975,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 32.0,
      "xla_arg_mb": 20.000003814697266,
      "xla_out_mb": 4.0,
      "compile_s": 0.37894001899985597,
      "inner": 1,
      "median_ms": 90.95401399827097,
      "q1_ms": 89.75259900034871,
      "q3_ms": 91.70312299829675,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 65536,
      "m": 65536,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 48.00001525878906,
      "xla_arg_mb": 20.000003814697266,
      "xla_out_mb": 20.00003433227539,
      "compile_s": 0.5438813769942499,
      "inner": 1,
      "median_ms": 245.89469299826305,
      "q1_ms": 245.7060669985367,
      "q3_ms": 285.7638460045564,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kde_density",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd",
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "xla_temp_mb": 0.00390625,
      "xla_arg_mb": 0.015628814697265625,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.009439795998332556,
      "inner": 19,
      "median_ms": 0.14636400010203943,
      "q1_ms": 0.1462886315206752,
      "q3_ms": 0.15202894734868191,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kde_density",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd_bwd",
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "xla_temp_mb": 0.0083160400390625,
      "xla_arg_mb": 0.015628814697265625,
      "xla_out_mb": 0.007831573486328125,
      "compile_s": 0.007435981999151409,
      "inner": 21,
      "median_ms": 0.15694676195770235,
      "q1_ms": 0.15617338102033718,
      "q3_ms": 0.15701785741839558,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kde_density",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd",
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "xla_temp_mb": 0.00390625,
      "xla_arg_mb": 0.06250381469726562,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.007750527998723555,
      "inner": 20,
      "median_ms": 0.14547450009558816,
      "q1_ms": 0.1419959000486415,
      "q3_ms": 0.1490722999733407,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kde_density",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd_bwd",
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "xla_temp_mb": 0.0083160400390625,
      "xla_arg_mb": 0.06250381469726562,
      "xla_out_mb": 0.031269073486328125,
      "compile_s": 0.014610459002142306,
      "inner": 21,
      "median_ms": 0.15550019071226764,
      "q1_ms": 0.1553971430569488,
      "q3_ms": 0.15562042868363538,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kde_density",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd",
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "xla_temp_mb": 0.015625,
      "xla_arg_mb": 0.06250381469726562,
      "xla_out_mb": 0.015625,
      "compile_s": 0.006103458996221889,
      "inner": 9,
      "median_ms": 0.3649377775016344,
      "q1_ms": 0.3349108883412555,
      "q3_ms": 0.39146022269657504,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kde_density",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd_bwd",
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "xla_temp_mb": 0.0319976806640625,
      "xla_arg_mb": 0.06250381469726562,
      "xla_out_mb": 0.031269073486328125,
      "compile_s": 0.01786564299982274,
      "inner": 24,
      "median_ms": 0.3373100416865782,
      "q1_ms": 0.17161000020375164,
      "q3_ms": 0.34327045826406294,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kde_density",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "xla_temp_mb": 0.015625,
      "xla_arg_mb": 0.2500038146972656,
      "xla_out_mb": 0.015625,
      "compile_s": 0.0025174670008709654,
      "inner": 42,
      "median_ms": 0.141263238003289,
      "q1_ms": 0.07878014280105985,
      "q3_ms": 0.16379902375601452,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kde_density",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd_bwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "xla_temp_mb": 0.0319976806640625,
      "xla_arg_mb": 0.2500038146972656,
      "xla_out_mb": 0.12501907348632812,
      "compile_s": 0.001092028003768064,
      "inner": 195,
      "median_ms": 0.16446498973378673,
      "q1_ms": 0.162258820506768,
      "q3_ms": 0.17434406151714474,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kde_density",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd",
      "n": 16384,
      "m": 16384,
      "dim": 2,
      "xla_temp_mb": 0.0625,
      "xla_arg_mb": 0.2500038146972656,
      "xla_out_mb": 0.0625,
      "compile_s": 0.0014716970035806298,
      "inner": 67,
      "median_ms": 0.7741763880919429,
      "q1_ms": 0.7234452537316551,
      "q3_ms": 0.9978860149719515,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kde_density",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd_bwd",
      "n": 16384,
      "m": 16384,
      "dim": 2,
      "xla_temp_mb": 0.1264801025390625,
      "xla_arg_mb": 0.2500038146972656,
      "xla_out_mb": 0.12501907348632812,
      "compile_s": 0.0022469549949164502,
      "inner": 32,
      "median_ms": 1.5312106875171594,
      "q1_ms": 1.5248980937485612,
      "q3_ms": 1.5409913437451905,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kde_density",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd",
      "n": 16384,
      "m": 16384,
      "dim": 8,
      "xla_temp_mb": 0.0625,
      "xla_arg_mb": 1.0000038146972656,
      "xla_out_mb": 0.0625,
      "compile_s": 0.00482448600087082,
      "inner": 26,
      "median_ms": 1.0370458462253178,
      "q1_ms": 1.0312849232062804,
      "q3_ms": 1.0948953847848597,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kde_density",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd_bwd",
      "n": 16384,
      "m": 16384,
      "dim": 8,
      "xla_temp_mb": 0.1264801025390625,
      "xla_arg_mb": 1.0000038146972656,
      "xla_out_mb": 0.5000190734863281,
      "compile_s": 0.008254471998952795,
      "inner": 22,
      "median_ms": 1.5377653180621564,
      "q1_ms": 1.5179791817817285,
      "q3_ms": 1.5409541364616335,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kde_density",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd",
      "n": 65536,
      "m": 65536,
      "dim": 2,
      "xla_temp_mb": 0.25,
      "xla_arg_mb": 1.0000038146972656,
      "xla_out_mb": 0.25,
      "compile_s": 0.008395759999984875,
      "inner": 6,
      "median_ms": 8.410921667139823,
      "q1_ms": 8.345657833463823,
      "q3_ms": 8.57945233292412,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kde_density",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd_bwd",
      "n": 65536,
      "m": 65536,
      "dim": 2,
      "xla_temp_mb": 0.5044097900390625,
      "xla_arg_mb": 1.0000038146972656,
      "xla_out_mb": 0.5000190734863281,
      "compile_s": 0.01976019099674886,
      "inner": 2,
      "median_ms": 19.781201001023874,
      "q1_ms": 19.661358499433845,
      "q3_ms": 19.790072499745293,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kde_density",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd",
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "xla_temp_mb": 0.25,
      "xla_arg_mb": 4.000003814697266,
      "xla_out_mb": 0.25,
      "compile_s": 0.017723092998494394,
      "inner": 3,
      "median_ms": 16.92642766526357,
      "q1_ms": 16.892777668545023,
      "q3_ms": 17.156920999696013,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kde_density",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd_bwd",
      "n": 65536,
      "m": 65536,
      "dim": 8,
      "xla_temp_mb": 0.5044097900390625,
      "xla_arg_mb": 4.000003814697266,
      "xla_out_mb": 2.000019073486328,
      "compile_s": 0.020000476004497614,
      "inner": 2,
      "median_ms": 20.297197999752825,
      "q1_ms": 20.27321750210831,
      "q3_ms": 20.305252997786738,
      "checkout": "gpu-benchmarks (before)"
    }
  ]
}
