{
  "device": {
    "jax": "0.11.2",
    "backend": "cpu",
    "devices": [
      "cpu:0"
    ],
    "python": "3.14.8",
    "device_kind": "cpu",
    "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": {
      "cpu_a.json": {
        "device": "cpu",
        "output": "/tmp/claude-1003/-home-manuel-src/80785a0e-56be-4da7-a2a8-e1d6effd6199/scratchpad/kmv/cpu_a.json",
        "sections": [
          "grid"
        ],
        "sizes": [
          1024,
          4096
        ],
        "dims": [
          2,
          8,
          32
        ],
        "outs": [
          1,
          16
        ],
        "modes": [
          "fwd",
          "fwd_bwd"
        ],
        "impls": [
          "dense",
          "xla",
          "pallas_interpret",
          "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
      },
      "cpu_b.json": {
        "device": "cpu",
        "output": "/tmp/claude-1003/-home-manuel-src/80785a0e-56be-4da7-a2a8-e1d6effd6199/scratchpad/kmv/cpu_b.json",
        "sections": [
          "grid"
        ],
        "sizes": [
          16384
        ],
        "dims": [
          8
        ],
        "outs": [
          16
        ],
        "modes": [
          "fwd",
          "fwd_bwd"
        ],
        "impls": [
          "dense",
          "xla",
          "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
      },
      "cpu_kde.json": {
        "device": "cpu",
        "output": "/tmp/claude-1003/-home-manuel-src/80785a0e-56be-4da7-a2a8-e1d6effd6199/scratchpad/kmv/cpu_kde.json",
        "sections": [
          "kde"
        ],
        "sizes": [
          1024,
          4096
        ],
        "dims": [
          2,
          8,
          32
        ],
        "outs": [
          1,
          16
        ],
        "modes": [
          "fwd",
          "fwd_bwd"
        ],
        "impls": [
          "dense",
          "xla",
          "pallas_interpret"
        ],
        "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
      },
      "cpu_maxn.json": {
        "device": "cpu",
        "output": "/tmp/claude-1003/-home-manuel-src/80785a0e-56be-4da7-a2a8-e1d6effd6199/scratchpad/kmv/cpu_maxn.json",
        "sections": [
          "max_n"
        ],
        "sizes": [
          1024,
          4096,
          16384
        ],
        "dims": [
          2,
          8,
          32
        ],
        "outs": [
          1,
          16
        ],
        "modes": [
          "fwd",
          "fwd_bwd"
        ],
        "impls": [
          "xla"
        ],
        "repeats": 5,
        "tiles": [
          "64,64,64",
          "64,64,16",
          "128,64,16",
          "64,128,16",
          "128,128,16"
        ],
        "ladder": [
          16384,
          32768,
          65536
        ],
        "timeout": 150,
        "child": null
      },
      "before_cpu.json": {
        "device": "cpu",
        "output": "/tmp/claude-1003/-home-manuel-src/80785a0e-56be-4da7-a2a8-e1d6effd6199/scratchpad/kmv/before_cpu.json",
        "sections": [
          "grid",
          "kde"
        ],
        "sizes": [
          1024
        ],
        "dims": [
          2,
          8,
          32
        ],
        "outs": [
          1,
          16
        ],
        "modes": [
          "fwd",
          "fwd_bwd"
        ],
        "impls": [
          "default"
        ],
        "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
      }
    }
  },
  "results": [
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 1,
      "xla_temp_mb": 8.007816314697266,
      "xla_arg_mb": 0.019535064697265625,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.04904211399843916,
      "inner": 36,
      "median_ms": 0.7201077779528634,
      "q1_ms": 0.7123926389896144,
      "q3_ms": 0.7209349167977861,
      "rel_err": 0.0
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 1,
      "xla_temp_mb": 8.007816314697266,
      "xla_arg_mb": 0.019535064697265625,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.05059160599921597,
      "inner": 78,
      "median_ms": 0.31681805119613926,
      "q1_ms": 0.31602636534863937,
      "q3_ms": 0.4486168012450575,
      "rel_err": 0.0
    },
    {
      "family": "kernel_mv",
      "impl": "pallas_interpret",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 1,
      "xla_temp_mb": 0.5711135864257812,
      "xla_arg_mb": 0.019535064697265625,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.1452298350050114,
      "inner": 20,
      "median_ms": 2.289550649948069,
      "q1_ms": 2.2885725749802077,
      "q3_ms": 2.289978000044357,
      "rel_err": 2.1091370188322762e-07
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 1,
      "compile_s": 0.005315726004482713,
      "inner": 13,
      "median_ms": 4.198798384795037,
      "q1_ms": 4.117168576787942,
      "q3_ms": 4.206464384669044,
      "rel_err": 2.734391698726909e-07
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 1,
      "xla_temp_mb": 20.00390625,
      "xla_arg_mb": 0.019535064697265625,
      "xla_out_mb": 0.019565582275390625,
      "compile_s": 0.0932777560010436,
      "inner": 8,
      "median_ms": 2.831145625350473,
      "q1_ms": 2.813703937590617,
      "q3_ms": 2.8586953749254462,
      "rel_err": {
        "dq": 2.3131273161799995e-07,
        "dk": 1.1491617609202143e-07,
        "dv": 0.0,
        "dlengthscale": 6.93839840117098e-08
      }
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 1,
      "xla_temp_mb": 16.007816314697266,
      "xla_arg_mb": 0.019535064697265625,
      "xla_out_mb": 0.019565582275390625,
      "compile_s": 0.08472320299915737,
      "inner": 23,
      "median_ms": 1.947319477988893,
      "q1_ms": 1.946056217212058,
      "q3_ms": 1.9586728911077522,
      "rel_err": {
        "dq": 0.0,
        "dk": 0.0,
        "dv": 0.0,
        "dlengthscale": 0.0
      }
    },
    {
      "family": "kernel_mv",
      "impl": "pallas_interpret",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 1,
      "xla_temp_mb": 1.549102783203125,
      "xla_arg_mb": 0.019535064697265625,
      "xla_out_mb": 0.019565582275390625,
      "compile_s": 0.23809257600078126,
      "inner": 4,
      "median_ms": 11.361878750903998,
      "q1_ms": 11.358824124727107,
      "q3_ms": 11.403612626054382,
      "rel_err": {
        "dq": 3.414265214088218e-07,
        "dk": 1.0389666674927372e-07,
        "dv": 4.7549680876554187e-07,
        "dlengthscale": 1.387679680234196e-07
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 1,
      "compile_s": 0.1573013000015635,
      "inner": 10,
      "median_ms": 4.011617899959674,
      "q1_ms": 4.004791999977897,
      "q3_ms": 4.021796150118462,
      "rel_err": {
        "dq": 7.006441948349668e-07,
        "dk": 2.1929749022926025e-07,
        "dv": 1.0781886646877504e-07
      }
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 16,
      "xla_temp_mb": 8.007816314697266,
      "xla_arg_mb": 0.07812881469726562,
      "xla_out_mb": 0.0625,
      "compile_s": 0.04510439599835081,
      "inner": 61,
      "median_ms": 0.7270300982960836,
      "q1_ms": 0.7082459589702146,
      "q3_ms": 0.7356358851882492,
      "rel_err": 0.0
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 16,
      "xla_temp_mb": 8.007816314697266,
      "xla_arg_mb": 0.07812881469726562,
      "xla_out_mb": 0.0625,
      "compile_s": 0.04478518100222573,
      "inner": 73,
      "median_ms": 0.5080884246575034,
      "q1_ms": 0.4175516712153968,
      "q3_ms": 0.6097256232787415,
      "rel_err": 0.0
    },
    {
      "family": "kernel_mv",
      "impl": "pallas_interpret",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 16,
      "xla_temp_mb": 0.5559768676757812,
      "xla_arg_mb": 0.07812881469726562,
      "xla_out_mb": 0.0625,
      "compile_s": 0.11784646300657187,
      "inner": 21,
      "median_ms": 2.325888523912346,
      "q1_ms": 2.3028487381192724,
      "q3_ms": 2.3273097857392195,
      "rel_err": 5.640582347477708e-07
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 16,
      "compile_s": 0.0017421459997422062,
      "inner": 43,
      "median_ms": 1.0021495349234142,
      "q1_ms": 0.9991352209993258,
      "q3_ms": 1.0033968837253564,
      "rel_err": 5.731614162713022e-07
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 16,
      "xla_temp_mb": 16.12890625,
      "xla_arg_mb": 0.07812881469726562,
      "xla_out_mb": 0.07815933227539062,
      "compile_s": 0.08878612899570726,
      "inner": 17,
      "median_ms": 2.971322058703449,
      "q1_ms": 2.9413167940219864,
      "q3_ms": 2.9807764412373006,
      "rel_err": {
        "dq": 1.630812745097389e-07,
        "dk": 1.1900883424448616e-07,
        "dv": 0.0,
        "dlengthscale": 7.582849263193495e-08
      }
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 16,
      "xla_temp_mb": 12.062503814697266,
      "xla_arg_mb": 0.07812881469726562,
      "xla_out_mb": 0.07815933227539062,
      "compile_s": 0.08003995400213171,
      "inner": 18,
      "median_ms": 2.571905333348291,
      "q1_ms": 2.554799083251661,
      "q3_ms": 2.5958506388027063,
      "rel_err": {
        "dq": 0.0,
        "dk": 0.0,
        "dv": 0.0,
        "dlengthscale": 0.0
      }
    },
    {
      "family": "kernel_mv",
      "impl": "pallas_interpret",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 16,
      "xla_temp_mb": 1.533966064453125,
      "xla_arg_mb": 0.07812881469726562,
      "xla_out_mb": 0.07815933227539062,
      "compile_s": 0.24259558499761624,
      "inner": 3,
      "median_ms": 11.468812668075165,
      "q1_ms": 11.39218766790388,
      "q3_ms": 11.741972667853892,
      "rel_err": {
        "dq": 3.757237430003347e-07,
        "dk": 1.166141610263083e-07,
        "dv": 9.47534949915184e-08,
        "dlengthscale": 0.0
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 16,
      "compile_s": 0.007171917000960093,
      "inner": 9,
      "median_ms": 4.91923244408099,
      "q1_ms": 4.866867888469844,
      "q3_ms": 4.942773666294265,
      "rel_err": {
        "dq": 5.796119159028381e-07,
        "dk": 2.4611317068375404e-07,
        "dv": 1.2054033907206014e-07
      }
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 4.031253814697266,
      "xla_arg_mb": 0.06641006469726562,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.04424741399998311,
      "inner": 92,
      "median_ms": 0.4087386630050352,
      "q1_ms": 0.3900918152240869,
      "q3_ms": 0.4092332391060747,
      "rel_err": 0.0
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 4.031253814697266,
      "xla_arg_mb": 0.06641006469726562,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.04458520299522206,
      "inner": 90,
      "median_ms": 0.4348227889244703,
      "q1_ms": 0.4078745889273705,
      "q3_ms": 0.4359037833233338,
      "rel_err": 0.0
    },
    {
      "family": "kernel_mv",
      "impl": "pallas_interpret",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 0.5945510864257812,
      "xla_arg_mb": 0.06641006469726562,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.11599802700220607,
      "inner": 20,
      "median_ms": 2.3742008499539224,
      "q1_ms": 2.365228374947037,
      "q3_ms": 2.376126499984821,
      "rel_err": 4.822312056274688e-07
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "compile_s": 0.00216196000110358,
      "inner": 44,
      "median_ms": 0.8176711364120075,
      "q1_ms": 0.8169091137130056,
      "q3_ms": 0.858969034115117,
      "rel_err": 5.78870725399737e-07
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 16.0,
      "xla_arg_mb": 0.06641006469726562,
      "xla_out_mb": 0.06644058227539062,
      "compile_s": 0.08936892199562863,
      "inner": 18,
      "median_ms": 2.9681704998236254,
      "q1_ms": 2.9262158609021247,
      "q3_ms": 2.9874817775837275,
      "rel_err": {
        "dq": 2.3688929938700644e-07,
        "dk": 2.3076342036376968e-07,
        "dv": 0.0,
        "dlengthscale": 9.89052866161465e-08
      }
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 12.003910064697266,
      "xla_arg_mb": 0.06641006469726562,
      "xla_out_mb": 0.06644058227539062,
      "compile_s": 0.08515885700035142,
      "inner": 20,
      "median_ms": 1.9587589500588365,
      "q1_ms": 1.9532467500539497,
      "q3_ms": 2.0219652498781215,
      "rel_err": {
        "dq": 0.0,
        "dk": 0.0,
        "dv": 0.0,
        "dlengthscale": 0.0
      }
    },
    {
      "family": "kernel_mv",
      "impl": "pallas_interpret",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 1.644317626953125,
      "xla_arg_mb": 0.06641006469726562,
      "xla_out_mb": 0.06644058227539062,
      "compile_s": 0.24071529699722305,
      "inner": 4,
      "median_ms": 12.000652001006529,
      "q1_ms": 11.997050250101893,
      "q3_ms": 12.162604625700624,
      "rel_err": {
        "dq": 6.632070896711362e-07,
        "dk": 2.5319568724951726e-07,
        "dv": 4.853955772597559e-07,
        "dlengthscale": 4.945264330807326e-07
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "compile_s": 0.006418120996386278,
      "inner": 11,
      "median_ms": 4.007536636808337,
      "q1_ms": 4.002312182075746,
      "q3_ms": 4.059490409087051,
      "rel_err": {
        "dq": 8.967653152446449e-07,
        "dk": 5.239130254712595e-07,
        "dv": 2.866384925036506e-07
      }
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 0.031253814697265625,
      "xla_arg_mb": 0.12500381469726562,
      "xla_out_mb": 0.0625,
      "compile_s": 0.037841555997147225,
      "inner": 75,
      "median_ms": 0.39534561336040497,
      "q1_ms": 0.3495323533328095,
      "q3_ms": 0.39948498000740074,
      "rel_err": 0.0
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 0.031253814697265625,
      "xla_arg_mb": 0.12500381469726562,
      "xla_out_mb": 0.0625,
      "compile_s": 0.040482201002305374,
      "inner": 79,
      "median_ms": 0.3686213797353255,
      "q1_ms": 0.35539566453758487,
      "q3_ms": 0.3846324936971599,
      "rel_err": 0.0
    },
    {
      "family": "kernel_mv",
      "impl": "pallas_interpret",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 0.5652542114257812,
      "xla_arg_mb": 0.12500381469726562,
      "xla_out_mb": 0.0625,
      "compile_s": 0.1144243389935582,
      "inner": 19,
      "median_ms": 2.3785854209563695,
      "q1_ms": 2.372612000104518,
      "q3_ms": 2.396582868402076,
      "rel_err": 6.013135215604066e-07
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "compile_s": 0.0016683270005160011,
      "inner": 43,
      "median_ms": 1.001665814032475,
      "q1_ms": 1.0010441744038132,
      "q3_ms": 1.00625110475406,
      "rel_err": 6.182778406808962e-07
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 16.125,
      "xla_arg_mb": 0.12500381469726562,
      "xla_out_mb": 0.12503433227539062,
      "compile_s": 0.0893151689961087,
      "inner": 14,
      "median_ms": 3.273389500204108,
      "q1_ms": 3.056515285736948,
      "q3_ms": 3.3495160358662748,
      "rel_err": {
        "dq": 1.4013121648509653e-07,
        "dk": 1.0185494718945941e-07,
        "dv": 0.0,
        "dlengthscale": 0.0
      }
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 12.0625,
      "xla_arg_mb": 0.12500381469726562,
      "xla_out_mb": 0.12503433227539062,
      "compile_s": 0.08067770000343444,
      "inner": 18,
      "median_ms": 2.833696666736311,
      "q1_ms": 2.8016596944427063,
      "q3_ms": 2.834018388966797,
      "rel_err": {
        "dq": 0.0,
        "dk": 0.0,
        "dv": 0.0,
        "dlengthscale": 0.0
      }
    },
    {
      "family": "kernel_mv",
      "impl": "pallas_interpret",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 1.615020751953125,
      "xla_arg_mb": 0.12500381469726562,
      "xla_out_mb": 0.12503433227539062,
      "compile_s": 0.24358320300234482,
      "inner": 4,
      "median_ms": 11.880038749950472,
      "q1_ms": 11.858836250212335,
      "q3_ms": 11.933601374948921,
      "rel_err": {
        "dq": 3.908861762163557e-07,
        "dk": 1.3034164480634804e-07,
        "dv": 1.049487561663728e-07,
        "dlengthscale": 0.0
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 16,
      "compile_s": 0.010907003997999709,
      "inner": 7,
      "median_ms": 5.718858571656581,
      "q1_ms": 5.312506929060744,
      "q3_ms": 5.906155000307731,
      "rel_err": {
        "dq": 4.5467546937381197e-07,
        "dk": 1.5821312665996588e-07,
        "dv": 1.353587438344082e-07
      }
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 4.125003814697266,
      "xla_arg_mb": 0.2539100646972656,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.045707195000431966,
      "inner": 68,
      "median_ms": 0.48077788233058527,
      "q1_ms": 0.47734808825172126,
      "q3_ms": 0.4961379705384709,
      "rel_err": 0.0
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 4.125003814697266,
      "xla_arg_mb": 0.2539100646972656,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.04705333200399764,
      "inner": 63,
      "median_ms": 0.4677624920510598,
      "q1_ms": 0.4460696348807745,
      "q3_ms": 0.4739628730310378,
      "rel_err": 0.0
    },
    {
      "family": "kernel_mv",
      "impl": "pallas_interpret",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 0.7739944458007812,
      "xla_arg_mb": 0.2539100646972656,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.1151695890002884,
      "inner": 14,
      "median_ms": 2.9592737856936373,
      "q1_ms": 2.9561046069699968,
      "q3_ms": 2.9785914284730097,
      "rel_err": 3.238302501810531e-07
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 1,
      "compile_s": 0.0017677239957265556,
      "inner": 41,
      "median_ms": 1.0184837803713092,
      "q1_ms": 1.0040526828627818,
      "q3_ms": 1.0288455609619764,
      "rel_err": 4.13640964536018e-07
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 16.0,
      "xla_arg_mb": 0.2539100646972656,
      "xla_out_mb": 0.2539405822753906,
      "compile_s": 0.09044424599414924,
      "inner": 14,
      "median_ms": 3.13007414245346,
      "q1_ms": 3.0757356068664063,
      "q3_ms": 3.2120363212015945,
      "rel_err": {
        "dq": 2.785974434588553e-07,
        "dk": 9.737766407275531e-08,
        "dv": 0.0,
        "dlengthscale": 0.0
      }
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 12.003910064697266,
      "xla_arg_mb": 0.2539100646972656,
      "xla_out_mb": 0.2539405822753906,
      "compile_s": 0.08504230700054904,
      "inner": 20,
      "median_ms": 2.0581559499987634,
      "q1_ms": 2.0132575000388897,
      "q3_ms": 2.2189726749274996,
      "rel_err": {
        "dq": 0.0,
        "dk": 0.0,
        "dv": 0.0,
        "dlengthscale": 0.0
      }
    },
    {
      "family": "kernel_mv",
      "impl": "pallas_interpret",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 2.032989501953125,
      "xla_arg_mb": 0.2539100646972656,
      "xla_out_mb": 0.2539405822753906,
      "compile_s": 0.23923229199863272,
      "inner": 3,
      "median_ms": 15.269018333735099,
      "q1_ms": 15.23001116705321,
      "q3_ms": 15.739625165830756,
      "rel_err": {
        "dq": 5.95765672062591e-07,
        "dk": 1.201738168538291e-07,
        "dv": 4.754733897282389e-07,
        "dlengthscale": 0.0
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 1,
      "compile_s": 0.0059716070027207024,
      "inner": 11,
      "median_ms": 4.141392545451791,
      "q1_ms": 4.131969272500349,
      "q3_ms": 4.1540701816054275,
      "rel_err": {
        "dq": 6.63021690084746e-07,
        "dk": 9.923463188353231e-08,
        "dv": 1.2397476685464777e-07
      }
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 0.12500381469726562,
      "xla_arg_mb": 0.3125038146972656,
      "xla_out_mb": 0.0625,
      "compile_s": 0.04295070999796735,
      "inner": 68,
      "median_ms": 0.4569797794615357,
      "q1_ms": 0.451711727959371,
      "q3_ms": 0.4707734485825366,
      "rel_err": 0.0
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 0.12500381469726562,
      "xla_arg_mb": 0.3125038146972656,
      "xla_out_mb": 0.0625,
      "compile_s": 0.04325775199686177,
      "inner": 71,
      "median_ms": 0.4391902675450465,
      "q1_ms": 0.4198998309656302,
      "q3_ms": 0.4491715422568312,
      "rel_err": 0.0
    },
    {
      "family": "kernel_mv",
      "impl": "pallas_interpret",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 0.7586135864257812,
      "xla_arg_mb": 0.3125038146972656,
      "xla_out_mb": 0.0625,
      "compile_s": 0.11591582999972161,
      "inner": 14,
      "median_ms": 2.973643857071043,
      "q1_ms": 2.9695976427319692,
      "q3_ms": 3.0191796785012617,
      "rel_err": 6.400829612130672e-07
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "compile_s": 0.001964907998626586,
      "inner": 35,
      "median_ms": 1.0033563712827993,
      "q1_ms": 1.0008599284836754,
      "q3_ms": 1.0869620998814105,
      "rel_err": 6.633030393513967e-07
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 16.0625,
      "xla_arg_mb": 0.3125038146972656,
      "xla_out_mb": 0.3125343322753906,
      "compile_s": 0.09080698199977633,
      "inner": 13,
      "median_ms": 3.480372307911659,
      "q1_ms": 3.4526102309218216,
      "q3_ms": 3.5120025769664123,
      "rel_err": {
        "dq": 1.2481858227021594e-07,
        "dk": 9.783383555447191e-08,
        "dv": 0.0,
        "dlengthscale": 8.601062179572805e-08
      }
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 12.0,
      "xla_arg_mb": 0.3125038146972656,
      "xla_out_mb": 0.3125343322753906,
      "compile_s": 0.07708685700345086,
      "inner": 15,
      "median_ms": 3.0436337333715833,
      "q1_ms": 3.0405549000230776,
      "q3_ms": 3.07794963324947,
      "rel_err": {
        "dq": 0.0,
        "dk": 0.0,
        "dv": 0.0,
        "dlengthscale": 0.0
      }
    },
    {
      "family": "kernel_mv",
      "impl": "pallas_interpret",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 2.017364501953125,
      "xla_arg_mb": 0.3125038146972656,
      "xla_out_mb": 0.3125343322753906,
      "compile_s": 0.24262311599886743,
      "inner": 2,
      "median_ms": 15.577308000501944,
      "q1_ms": 15.451151999513968,
      "q3_ms": 15.842594750210992,
      "rel_err": {
        "dq": 3.6050699820151126e-07,
        "dk": 1.253606062287492e-07,
        "dv": 1.1541110028188569e-07,
        "dlengthscale": 8.601062179572805e-08
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 32,
      "out": 16,
      "compile_s": 0.0078098840021993965,
      "inner": 8,
      "median_ms": 5.1629668751047575,
      "q1_ms": 5.0964686251973035,
      "q3_ms": 5.270849750104389,
      "rel_err": {
        "dq": 3.6735381228208164e-07,
        "dk": 1.0870277000772894e-07,
        "dv": 1.2791730587800022e-07
      }
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "out": 1,
      "xla_temp_mb": 64.03125381469727,
      "xla_arg_mb": 0.07812881469726562,
      "xla_out_mb": 0.015625,
      "compile_s": 0.043791390999103896,
      "inner": 3,
      "median_ms": 15.543571001520226,
      "q1_ms": 15.219948834176952,
      "q3_ms": 16.64889200158844,
      "rel_err": 9.490572982114094e-08
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "out": 1,
      "xla_temp_mb": 32.06664276123047,
      "xla_arg_mb": 0.07812881469726562,
      "xla_out_mb": 0.015625,
      "compile_s": 0.07463099099550163,
      "inner": 3,
      "median_ms": 13.739160667076552,
      "q1_ms": 13.620002000000872,
      "q3_ms": 13.828103999306526,
      "rel_err": 0.0
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "out": 1,
      "compile_s": 0.07801059500343399,
      "inner": 1,
      "median_ms": 81.22609699785244,
      "q1_ms": 80.63075549944188,
      "q3_ms": 82.15168849710608,
      "rel_err": 6.251414222739596e-07
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "out": 1,
      "xla_temp_mb": 256.0,
      "xla_arg_mb": 0.07812881469726562,
      "xla_out_mb": 0.07815933227539062,
      "compile_s": 0.08788992399786366,
      "inner": 1,
      "median_ms": 97.97568299836712,
      "q1_ms": 97.48717399997986,
      "q3_ms": 98.18665749844513,
      "rel_err": {
        "dq": 3.0576314755308254e-07,
        "dk": 9.589624956520177e-08,
        "dv": 9.89898790721658e-07,
        "dlengthscale": 7.846490629603193e-08
      }
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "out": 1,
      "xla_temp_mb": 64.07860565185547,
      "xla_arg_mb": 0.07812881469726562,
      "xla_out_mb": 0.07815933227539062,
      "compile_s": 0.12526687100034906,
      "inner": 1,
      "median_ms": 60.9076409964473,
      "q1_ms": 60.42617999628419,
      "q3_ms": 62.48209399564075,
      "rel_err": {
        "dq": 0.0,
        "dk": 0.0,
        "dv": 0.0,
        "dlengthscale": 0.0
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "out": 1,
      "compile_s": 0.26028302499616984,
      "inner": 1,
      "median_ms": 253.37749100435758,
      "q1_ms": 253.29620699994848,
      "q3_ms": 257.41611100238515,
      "rel_err": {
        "dq": 1.3645297193154327e-06,
        "dk": 4.984619690544233e-07,
        "dv": 9.898608925900294e-07
      }
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "out": 16,
      "xla_temp_mb": 0.031253814697265625,
      "xla_arg_mb": 0.3125038146972656,
      "xla_out_mb": 0.25,
      "compile_s": 0.03730383900256129,
      "inner": 3,
      "median_ms": 20.85062366677448,
      "q1_ms": 18.7800050007354,
      "q3_ms": 21.28337900042728,
      "rel_err": 1.1325038094971399e-07
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "out": 16,
      "xla_temp_mb": 32.05101776123047,
      "xla_arg_mb": 0.3125038146972656,
      "xla_out_mb": 0.25,
      "compile_s": 0.0675650749981287,
      "inner": 3,
      "median_ms": 14.00265899913696,
      "q1_ms": 13.509582999783259,
      "q3_ms": 14.24756899965966,
      "rel_err": 0.0
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "out": 16,
      "compile_s": 0.0844058739967295,
      "inner": 1,
      "median_ms": 85.02836000116076,
      "q1_ms": 84.85707650106633,
      "q3_ms": 85.18678999826079,
      "rel_err": 8.517833722396528e-07
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "out": 16,
      "xla_temp_mb": 256.5,
      "xla_arg_mb": 0.3125038146972656,
      "xla_out_mb": 0.3125343322753906,
      "compile_s": 0.09121467300428776,
      "inner": 1,
      "median_ms": 98.53470999951242,
      "q1_ms": 97.1134270002949,
      "q3_ms": 100.25400250015082,
      "rel_err": {
        "dq": 2.271188070080682e-07,
        "dk": 9.940218638767264e-08,
        "dv": 1.0561316272740374e-07,
        "dlengthscale": 0.0
      }
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "out": 16,
      "xla_temp_mb": 80.37157440185547,
      "xla_arg_mb": 0.3125038146972656,
      "xla_out_mb": 0.3125343322753906,
      "compile_s": 0.12352643800113583,
      "inner": 1,
      "median_ms": 73.51138300145976,
      "q1_ms": 72.81534600042505,
      "q3_ms": 73.609569502878,
      "rel_err": {
        "dq": 0.0,
        "dk": 0.0,
        "dv": 0.0,
        "dlengthscale": 0.0
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "out": 16,
      "compile_s": 0.27039069500460755,
      "inner": 1,
      "median_ms": 281.9904880016111,
      "q1_ms": 281.41064699957496,
      "q3_ms": 283.5987060025218,
      "rel_err": {
        "dq": 1.2669515879154594e-06,
        "dk": 5.069882773777279e-07,
        "dv": 1.1027252771403406e-07
      }
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 64.12500381469727,
      "xla_arg_mb": 0.2656288146972656,
      "xla_out_mb": 0.015625,
      "compile_s": 0.045395966997602955,
      "inner": 3,
      "median_ms": 20.116829000471625,
      "q1_ms": 17.27238466628478,
      "q3_ms": 20.24451866721696,
      "rel_err": 0.0
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 16.31273651123047,
      "xla_arg_mb": 0.2656288146972656,
      "xla_out_mb": 0.015625,
      "compile_s": 0.07000950899964664,
      "inner": 9,
      "median_ms": 4.956262555626584,
      "q1_ms": 4.936137777905161,
      "q3_ms": 4.9880838887474965,
      "rel_err": 0.0
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 1,
      "compile_s": 0.07448684799601324,
      "inner": 1,
      "median_ms": 82.42050700209802,
      "q1_ms": 81.95751150196884,
      "q3_ms": 83.0501380005444,
      "rel_err": 9.102093330721248e-07
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 256.0,
      "xla_arg_mb": 0.2656288146972656,
      "xla_out_mb": 0.2656593322753906,
      "compile_s": 0.09245246399950702,
      "inner": 1,
      "median_ms": 97.12881099403603,
      "q1_ms": 95.54932349783485,
      "q3_ms": 97.60531749634538,
      "rel_err": {
        "dq": 4.092088441765005e-07,
        "dk": 9.031034113244833e-08,
        "dv": 9.586077956726323e-07,
        "dlengthscale": 0.0
      }
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 64.23876190185547,
      "xla_arg_mb": 0.2656288146972656,
      "xla_out_mb": 0.2656593322753906,
      "compile_s": 0.12873331800074084,
      "inner": 1,
      "median_ms": 83.49663300032262,
      "q1_ms": 80.968665999535,
      "q3_ms": 89.2962889993214,
      "rel_err": {
        "dq": 0.0,
        "dk": 0.0,
        "dv": 0.0,
        "dlengthscale": 0.0
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 1,
      "compile_s": 0.25821897899731994,
      "inner": 1,
      "median_ms": 253.2267860005959,
      "q1_ms": 249.30023750130204,
      "q3_ms": 254.0402395025012,
      "rel_err": {
        "dq": 1.5726282481312776e-06,
        "dk": 2.609336962450539e-07,
        "dv": 9.582590780010205e-07
      }
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 0.12500381469726562,
      "xla_arg_mb": 0.5000038146972656,
      "xla_out_mb": 0.25,
      "compile_s": 0.03888018100406043,
      "inner": 3,
      "median_ms": 19.74737999989884,
      "q1_ms": 19.51357099945502,
      "q3_ms": 21.18188866734272,
      "rel_err": 0.0
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 0.35961151123046875,
      "xla_arg_mb": 0.5000038146972656,
      "xla_out_mb": 0.25,
      "compile_s": 0.058556633004627656,
      "inner": 9,
      "median_ms": 4.885163555930679,
      "q1_ms": 4.844678611132419,
      "q3_ms": 4.931927055925674,
      "rel_err": 0.0
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 16,
      "compile_s": 0.0749167519970797,
      "inner": 1,
      "median_ms": 83.54893099749461,
      "q1_ms": 83.29619299911428,
      "q3_ms": 85.41938499911339,
      "rel_err": 6.812514170580017e-07
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 256.5,
      "xla_arg_mb": 0.5000038146972656,
      "xla_out_mb": 0.5000343322753906,
      "compile_s": 0.08824499500042293,
      "inner": 1,
      "median_ms": 100.56385899952147,
      "q1_ms": 99.24636699724942,
      "q3_ms": 101.00418699948932,
      "rel_err": {
        "dq": 1.6841884943454598e-07,
        "dk": 8.665011155818498e-08,
        "dv": 1.0649095779456192e-07,
        "dlengthscale": 0.0
      }
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 80.51610565185547,
      "xla_arg_mb": 0.5000038146972656,
      "xla_out_mb": 0.5000343322753906,
      "compile_s": 0.12612752900167834,
      "inner": 1,
      "median_ms": 89.53360300074564,
      "q1_ms": 89.21402900159592,
      "q3_ms": 98.48436050015152,
      "rel_err": {
        "dq": 0.0,
        "dk": 0.0,
        "dv": 0.0,
        "dlengthscale": 0.0
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "out": 16,
      "compile_s": 0.2840771670016693,
      "inner": 1,
      "median_ms": 288.80157400271855,
      "q1_ms": 281.95480150316143,
      "q3_ms": 290.37852699912037,
      "rel_err": {
        "dq": 5.19692517121903e-07,
        "dk": 1.7870438119855616e-07,
        "dv": 1.2145825875819032e-07
      }
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 64.50000381469727,
      "xla_arg_mb": 1.0156288146972656,
      "xla_out_mb": 0.015625,
      "compile_s": 0.04869400500319898,
      "inner": 1,
      "median_ms": 20.314930996391922,
      "q1_ms": 19.56253749813186,
      "q3_ms": 24.946101497334894,
      "rel_err": 0.0
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 17.15648651123047,
      "xla_arg_mb": 1.0156288146972656,
      "xla_out_mb": 0.015625,
      "compile_s": 0.07258267699944554,
      "inner": 8,
      "median_ms": 6.077057375478034,
      "q1_ms": 6.02417481240991,
      "q3_ms": 6.098233687680477,
      "rel_err": 0.0
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 32,
      "out": 1,
      "compile_s": 0.07473400599701563,
      "inner": 1,
      "median_ms": 79.73737100110156,
      "q1_ms": 77.24077850070898,
      "q3_ms": 80.54460849962197,
      "rel_err": 1.1140050160274918e-06
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 256.0,
      "xla_arg_mb": 1.0156288146972656,
      "xla_out_mb": 1.0156593322753906,
      "compile_s": 0.09428167399892118,
      "inner": 1,
      "median_ms": 98.99629399296828,
      "q1_ms": 98.95444449648494,
      "q3_ms": 99.65917799854651,
      "rel_err": {
        "dq": 4.1031839060306593e-07,
        "dk": 8.381253463454264e-08,
        "dv": 9.57425694056896e-07,
        "dlengthscale": 0.0
      }
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 32,
      "out": 1,
      "xla_temp_mb": 64.80126190185547,
      "xla_arg_mb": 1.0156288146972656,
      "xla_out_mb": 1.0156593322753906,
      "compile_s": 0.12640266000380507,
      "inner": 1,
      "median_ms": 82.80290099355625,
      "q1_ms": 82.14116399540217,
      "q3_ms": 83.81199949872098,
      "rel_err": {
        "dq": 0.0,
        "dk": 0.0,
        "dv": 0.0,
        "dlengthscale": 0.0
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 32,
      "out": 1,
      "compile_s": 0.2613965909986291,
      "inner": 1,
      "median_ms": 256.09630400140304,
      "q1_ms": 254.8075015001814,
      "q3_ms": 257.35720900047454,
      "rel_err": {
        "dq": 1.3205530502079708e-06,
        "dk": 1.5235556984348805e-07,
        "dv": 9.621552630198916e-07
      }
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 0.5000038146972656,
      "xla_arg_mb": 1.2500038146972656,
      "xla_out_mb": 0.25,
      "compile_s": 0.0470509380029398,
      "inner": 2,
      "median_ms": 20.367945500765927,
      "q1_ms": 19.823395250568865,
      "q3_ms": 20.45865475156461,
      "rel_err": 0.0
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 1.2033615112304688,
      "xla_arg_mb": 1.2500038146972656,
      "xla_out_mb": 0.25,
      "compile_s": 0.07050482599879615,
      "inner": 4,
      "median_ms": 9.667706250183983,
      "q1_ms": 9.550166374538094,
      "q3_ms": 9.773042875167448,
      "rel_err": 0.0
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 32,
      "out": 16,
      "compile_s": 0.08362415400188183,
      "inner": 1,
      "median_ms": 89.28303800348658,
      "q1_ms": 88.86647550025373,
      "q3_ms": 90.07991600083187,
      "rel_err": 1.0550311193858322e-06
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 256.25,
      "xla_arg_mb": 1.2500038146972656,
      "xla_out_mb": 1.2500343322753906,
      "compile_s": 0.09742639400064945,
      "inner": 1,
      "median_ms": 102.21837899734965,
      "q1_ms": 99.67487749963766,
      "q3_ms": 103.73991249798564,
      "rel_err": {
        "dq": 1.7234145853243992e-07,
        "dk": 8.163750119377922e-08,
        "dv": 1.0714112437121476e-07,
        "dlengthscale": 0.0
      }
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 32,
      "out": 16,
      "xla_temp_mb": 81.07860565185547,
      "xla_arg_mb": 1.2500038146972656,
      "xla_out_mb": 1.2500343322753906,
      "compile_s": 0.1252849249940482,
      "inner": 1,
      "median_ms": 95.31552099360852,
      "q1_ms": 95.24849399531377,
      "q3_ms": 95.94989599645487,
      "rel_err": {
        "dq": 0.0,
        "dk": 0.0,
        "dv": 0.0,
        "dlengthscale": 0.0
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 4096,
      "m": 4096,
      "dim": 32,
      "out": 16,
      "compile_s": 0.28616717999830144,
      "inner": 1,
      "median_ms": 278.993839994655,
      "q1_ms": 274.4150679973245,
      "q3_ms": 279.95967699825997,
      "rel_err": {
        "dq": 5.160128435064089e-07,
        "dk": 1.144358924238084e-07,
        "dv": 1.3382006588010812e-07
      }
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 0.5000038146972656,
      "xla_arg_mb": 2.0000038146972656,
      "xla_out_mb": 1.0,
      "compile_s": 0.04184916699887253,
      "inner": 1,
      "median_ms": 154.55297200242057,
      "q1_ms": 154.05402800024603,
      "q3_ms": 155.109885501588,
      "rel_err": 3.134581260213732e-07
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 32.56371307373047,
      "xla_arg_mb": 2.0000038146972656,
      "xla_out_mb": 1.0,
      "compile_s": 0.07125273100245977,
      "inner": 1,
      "median_ms": 156.3173500035191,
      "q1_ms": 156.03840250332723,
      "q3_ms": 156.78877150276094,
      "rel_err": 0.0
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 8,
      "out": 16,
      "compile_s": 1.4781135759985773,
      "inner": 1,
      "median_ms": 1464.4183080017683,
      "q1_ms": 1456.638555999234,
      "q3_ms": 1464.9884115024179,
      "rel_err": 1.3837989205449114e-06
    },
    {
      "family": "kernel_mv",
      "impl": "dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 4098.0,
      "xla_arg_mb": 2.0000038146972656,
      "xla_out_mb": 2.0000343322753906,
      "compile_s": 0.08706929100299021,
      "inner": 1,
      "median_ms": 1569.5314040058292,
      "q1_ms": 1567.6508660035324,
      "q3_ms": 1571.368845005054,
      "rel_err": {
        "dq": 3.825937919319912e-07,
        "dk": 1.399201602909189e-07,
        "dv": 1.9304321303951077e-07,
        "dlengthscale": 2.8588115863059494e-07
      }
    },
    {
      "family": "kernel_mv",
      "impl": "xla",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 8,
      "out": 16,
      "xla_temp_mb": 81.58739471435547,
      "xla_arg_mb": 2.0000038146972656,
      "xla_out_mb": 2.0000343322753906,
      "compile_s": 0.12159796999912942,
      "inner": 1,
      "median_ms": 995.9429509981419,
      "q1_ms": 995.4444135000813,
      "q3_ms": 1000.5251395014056,
      "rel_err": {
        "dq": 0.0,
        "dk": 0.0,
        "dv": 0.0,
        "dlengthscale": 0.0
      }
    },
    {
      "family": "kernel_mv",
      "impl": "torch_dense",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 16384,
      "m": 16384,
      "dim": 8,
      "out": 16,
      "compile_s": 4.799544894995051,
      "inner": 1,
      "median_ms": 4630.335106005077,
      "q1_ms": 4593.955656502658,
      "q3_ms": 4633.783935500105,
      "rel_err": {
        "dq": 1.5248530276206498e-06,
        "dk": 4.1859635645296555e-07,
        "dv": 1.9233501724071175e-07
      }
    },
    {
      "family": "kde_density",
      "impl": "dense",
      "mode": "fwd",
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "xla_temp_mb": 8.007877349853516,
      "xla_arg_mb": 0.015628814697265625,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.00277260500297416,
      "inner": 22,
      "median_ms": 0.7263845453218727,
      "q1_ms": 0.7039297953187171,
      "q3_ms": 0.752359613629365
    },
    {
      "family": "kde_density",
      "impl": "xla",
      "mode": "fwd",
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "xla_temp_mb": 8.007877349853516,
      "xla_arg_mb": 0.015628814697265625,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.0014918119995854795,
      "inner": 96,
      "median_ms": 0.6304054687310175,
      "q1_ms": 0.5122534687416191,
      "q3_ms": 0.6708988437367225
    },
    {
      "family": "kde_density",
      "impl": "pallas_interpret",
      "mode": "fwd",
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "xla_temp_mb": 0.02991485595703125,
      "xla_arg_mb": 0.015628814697265625,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.0019725210004253313,
      "inner": 44,
      "median_ms": 0.8632878636324841,
      "q1_ms": 0.8609592499520988,
      "q3_ms": 0.8674597386495655
    },
    {
      "family": "kde_density",
      "impl": "dense",
      "mode": "fwd_bwd",
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "xla_temp_mb": 16.015625,
      "xla_arg_mb": 0.015628814697265625,
      "xla_out_mb": 0.007831573486328125,
      "compile_s": 0.006343449000269175,
      "inner": 11,
      "median_ms": 1.9678608177028682,
      "q1_ms": 1.9489987726873634,
      "q3_ms": 2.019627409026725
    },
    {
      "family": "kde_density",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "xla_temp_mb": 16.007816314697266,
      "xla_arg_mb": 0.015628814697265625,
      "xla_out_mb": 0.007831573486328125,
      "compile_s": 0.004498260001128074,
      "inner": 28,
      "median_ms": 1.764899071401617,
      "q1_ms": 1.7540826963470733,
      "q3_ms": 1.7747176428071856
    },
    {
      "family": "kde_density",
      "impl": "pallas_interpret",
      "mode": "fwd_bwd",
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "xla_temp_mb": 0.050506591796875,
      "xla_arg_mb": 0.015628814697265625,
      "xla_out_mb": 0.007831573486328125,
      "compile_s": 0.005357194997486658,
      "inner": 19,
      "median_ms": 2.5389730523640632,
      "q1_ms": 2.5168749472006886,
      "q3_ms": 2.5413208945553234
    },
    {
      "family": "kde_density",
      "impl": "dense",
      "mode": "fwd",
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "xla_temp_mb": 4.031314849853516,
      "xla_arg_mb": 0.06250381469726562,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.002944721003586892,
      "inner": 88,
      "median_ms": 0.3857513182248857,
      "q1_ms": 0.38417937503254507,
      "q3_ms": 0.4067292443323822
    },
    {
      "family": "kde_density",
      "impl": "xla",
      "mode": "fwd",
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "xla_temp_mb": 4.031314849853516,
      "xla_arg_mb": 0.06250381469726562,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.0024544830012018792,
      "inner": 103,
      "median_ms": 0.41513638830861277,
      "q1_ms": 0.3711150048085312,
      "q3_ms": 0.42225456794749866
    },
    {
      "family": "kde_density",
      "impl": "pallas_interpret",
      "mode": "fwd",
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "xla_temp_mb": 0.06946563720703125,
      "xla_arg_mb": 0.06250381469726562,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.002079414000036195,
      "inner": 36,
      "median_ms": 0.9380506388778383,
      "q1_ms": 0.9353329999915634,
      "q3_ms": 0.9682202221989024
    },
    {
      "family": "kde_density",
      "impl": "dense",
      "mode": "fwd_bwd",
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "xla_temp_mb": 16.0078125,
      "xla_arg_mb": 0.06250381469726562,
      "xla_out_mb": 0.031269073486328125,
      "compile_s": 0.005962549003015738,
      "inner": 26,
      "median_ms": 2.1594073077721987,
      "q1_ms": 2.155539230723382,
      "q3_ms": 2.1931546154799153
    },
    {
      "family": "kde_density",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "xla_temp_mb": 8.003910064697266,
      "xla_arg_mb": 0.06250381469726562,
      "xla_out_mb": 0.031269073486328125,
      "compile_s": 0.005067156002041884,
      "inner": 34,
      "median_ms": 1.2635959706933457,
      "q1_ms": 1.2542031912316283,
      "q3_ms": 1.2733769706434996
    },
    {
      "family": "kde_density",
      "impl": "pallas_interpret",
      "mode": "fwd_bwd",
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "xla_temp_mb": 0.105194091796875,
      "xla_arg_mb": 0.06250381469726562,
      "xla_out_mb": 0.031269073486328125,
      "compile_s": 0.004865153998252936,
      "inner": 17,
      "median_ms": 2.806634117886835,
      "q1_ms": 2.8035809707088286,
      "q3_ms": 2.825516911851831
    },
    {
      "family": "kde_density",
      "impl": "dense",
      "mode": "fwd",
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "xla_temp_mb": 64.03131484985352,
      "xla_arg_mb": 0.06250381469726562,
      "xla_out_mb": 0.015625,
      "compile_s": 0.011247267000726424,
      "inner": 3,
      "median_ms": 18.343567666306626,
      "q1_ms": 16.99228466653343,
      "q3_ms": 19.92269316663927
    },
    {
      "family": "kde_density",
      "impl": "xla",
      "mode": "fwd",
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "xla_temp_mb": 32.06669616699219,
      "xla_arg_mb": 0.06250381469726562,
      "xla_out_mb": 0.015625,
      "compile_s": 0.01190512200264493,
      "inner": 3,
      "median_ms": 13.550864333713738,
      "q1_ms": 13.07908183298423,
      "q3_ms": 13.685379000283623
    },
    {
      "family": "kde_density",
      "impl": "dense",
      "mode": "fwd_bwd",
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "xla_temp_mb": 256.01562881469727,
      "xla_arg_mb": 0.06250381469726562,
      "xla_out_mb": 0.031269073486328125,
      "compile_s": 0.06769597299717134,
      "inner": 1,
      "median_ms": 86.62921899667708,
      "q1_ms": 86.20901399990544,
      "q3_ms": 87.42443799928878
    },
    {
      "family": "kde_density",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 4096,
      "m": 4096,
      "dim": 2,
      "xla_temp_mb": 64.06300354003906,
      "xla_arg_mb": 0.06250381469726562,
      "xla_out_mb": 0.031269073486328125,
      "compile_s": 0.04476242799864849,
      "inner": 1,
      "median_ms": 51.22614600259112,
      "q1_ms": 49.61175700373133,
      "q3_ms": 51.39814650101471
    },
    {
      "family": "kde_density",
      "impl": "dense",
      "mode": "fwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "xla_temp_mb": 64.12506484985352,
      "xla_arg_mb": 0.2500038146972656,
      "xla_out_mb": 0.015625,
      "compile_s": 0.011754665996704716,
      "inner": 1,
      "median_ms": 20.04929699614877,
      "q1_ms": 18.668059499759693,
      "q3_ms": 20.83503399626352
    },
    {
      "family": "kde_density",
      "impl": "xla",
      "mode": "fwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "xla_temp_mb": 16.312789916992188,
      "xla_arg_mb": 0.2500038146972656,
      "xla_out_mb": 0.015625,
      "compile_s": 0.009366579004563391,
      "inner": 8,
      "median_ms": 4.910149875286152,
      "q1_ms": 4.8812846875989635,
      "q3_ms": 4.946282625041931
    },
    {
      "family": "kde_density",
      "impl": "dense",
      "mode": "fwd_bwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "xla_temp_mb": 256.01562881469727,
      "xla_arg_mb": 0.2500038146972656,
      "xla_out_mb": 0.12501907348632812,
      "compile_s": 0.06563479299802566,
      "inner": 1,
      "median_ms": 87.51638700050535,
      "q1_ms": 87.07084400157328,
      "q3_ms": 87.81449099842575
    },
    {
      "family": "kde_density",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 4096,
      "m": 4096,
      "dim": 8,
      "xla_temp_mb": 32.18806457519531,
      "xla_arg_mb": 0.2500038146972656,
      "xla_out_mb": 0.12501907348632812,
      "compile_s": 0.03500032100419048,
      "inner": 1,
      "median_ms": 28.15192600246519,
      "q1_ms": 27.657826500217197,
      "q3_ms": 28.269373000512132
    },
    {
      "family": "kernel_mv_max_n",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 16384,
      "dim": 8,
      "out": 16,
      "run_s": 1.1267344579973724
    },
    {
      "family": "kernel_mv_max_n",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 32768,
      "dim": 8,
      "out": 16,
      "run_s": 4.397505105000164
    },
    {
      "family": "kernel_mv_max_n",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 65536,
      "dim": 8,
      "out": 16,
      "run_s": 18.564103108998097
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 1,
      "xla_temp_mb": 0.5711135864257812,
      "xla_arg_mb": 0.019535064697265625,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.17784618699806742,
      "inner": 17,
      "median_ms": 2.7694835294181863,
      "q1_ms": 2.648427529493347,
      "q3_ms": 2.838705500086208,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 1,
      "xla_temp_mb": 1.549102783203125,
      "xla_arg_mb": 0.019535064697265625,
      "xla_out_mb": 0.019565582275390625,
      "compile_s": 0.29758655600016937,
      "inner": 3,
      "median_ms": 13.595202666086456,
      "q1_ms": 13.091453666371915,
      "q3_ms": 13.671655833604746,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 16,
      "xla_temp_mb": 0.5559768676757812,
      "xla_arg_mb": 0.07812881469726562,
      "xla_out_mb": 0.0625,
      "compile_s": 0.15312224599620095,
      "inner": 20,
      "median_ms": 2.943954100192059,
      "q1_ms": 2.8390048752044095,
      "q3_ms": 3.009791800104722,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd_bwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "out": 16,
      "xla_temp_mb": 1.533966064453125,
      "xla_arg_mb": 0.07812881469726562,
      "xla_out_mb": 0.07815933227539062,
      "compile_s": 0.4486334809989785,
      "inner": 2,
      "median_ms": 22.553506001713686,
      "q1_ms": 17.86395000090124,
      "q3_ms": 23.25610125080857,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kernel_mv",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd",
      "batch": 1,
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "out": 1,
      "xla_temp_mb": 0.5945510864257812,
      "xla_arg_mb": 0.06641006469726562,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.17031176899763523,
      "inner": 13,
      "median_ms": 3.6415485385357615,
      "q1_ms": 3.3730816539900843,
      "q3_ms": 3.722313769125881,
      "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": 1.644317626953125,
      "xla_arg_mb": 0.06641006469726562,
      "xla_out_mb": 0.06644058227539062,
      "compile_s": 0.28241963300388306,
      "inner": 3,
      "median_ms": 13.888050331540095,
      "q1_ms": 13.299846499042662,
      "q3_ms": 13.933522999044119,
      "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": 0.5652542114257812,
      "xla_arg_mb": 0.12500381469726562,
      "xla_out_mb": 0.0625,
      "compile_s": 0.1364238230016781,
      "inner": 18,
      "median_ms": 2.834877999703167,
      "q1_ms": 2.643542333316873,
      "q3_ms": 2.837035138630502,
      "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": 1.615020751953125,
      "xla_arg_mb": 0.12500381469726562,
      "xla_out_mb": 0.12503433227539062,
      "compile_s": 0.29863271700014593,
      "inner": 3,
      "median_ms": 13.546854332768513,
      "q1_ms": 13.071981666143984,
      "q3_ms": 13.961875333431331,
      "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": 0.7739944458007812,
      "xla_arg_mb": 0.2539100646972656,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.1407131149971974,
      "inner": 13,
      "median_ms": 3.614353153702373,
      "q1_ms": 3.5985289614473004,
      "q3_ms": 3.7227056536022034,
      "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": 2.032989501953125,
      "xla_arg_mb": 0.2539100646972656,
      "xla_out_mb": 0.2539405822753906,
      "compile_s": 0.3554096980005852,
      "inner": 2,
      "median_ms": 20.407260999490973,
      "q1_ms": 19.344222999279737,
      "q3_ms": 22.96384524925088,
      "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": 0.7586135864257812,
      "xla_arg_mb": 0.3125038146972656,
      "xla_out_mb": 0.0625,
      "compile_s": 0.1455921849992592,
      "inner": 10,
      "median_ms": 3.5614595995866694,
      "q1_ms": 3.517944649865967,
      "q3_ms": 3.5881821499060607,
      "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": 2.017364501953125,
      "xla_arg_mb": 0.3125038146972656,
      "xla_out_mb": 0.3125343322753906,
      "compile_s": 0.364593418998993,
      "inner": 2,
      "median_ms": 17.277729497436667,
      "q1_ms": 17.20286549971206,
      "q3_ms": 18.126339247828582,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kde_density",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd",
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "xla_temp_mb": 0.02991485595703125,
      "xla_arg_mb": 0.015628814697265625,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.0022547100015799515,
      "inner": 40,
      "median_ms": 1.1447869499534136,
      "q1_ms": 1.077235274897248,
      "q3_ms": 1.3331212374396273,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kde_density",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd_bwd",
      "n": 1024,
      "m": 1024,
      "dim": 2,
      "xla_temp_mb": 0.050506591796875,
      "xla_arg_mb": 0.015628814697265625,
      "xla_out_mb": 0.007831573486328125,
      "compile_s": 0.005285339000693057,
      "inner": 16,
      "median_ms": 3.0490134372485045,
      "q1_ms": 2.9893379373788775,
      "q3_ms": 3.160230624871474,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kde_density",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd",
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "xla_temp_mb": 0.06946563720703125,
      "xla_arg_mb": 0.06250381469726562,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.0022636069988948293,
      "inner": 39,
      "median_ms": 1.0648547693749721,
      "q1_ms": 1.0645890128776396,
      "q3_ms": 1.0847003077521011,
      "checkout": "gpu-benchmarks (before)"
    },
    {
      "family": "kde_density",
      "impl": "default@gpu-benchmarks",
      "mode": "fwd_bwd",
      "n": 1024,
      "m": 1024,
      "dim": 8,
      "xla_temp_mb": 0.105194091796875,
      "xla_arg_mb": 0.06250381469726562,
      "xla_out_mb": 0.031269073486328125,
      "compile_s": 0.005188595998333767,
      "inner": 16,
      "median_ms": 3.241697687371925,
      "q1_ms": 3.128300124899397,
      "q3_ms": 3.675091249988327,
      "checkout": "gpu-benchmarks (before)"
    }
  ]
}
