{
  "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": {
    "epsilon": 0.5
  },
  "results": [
    {
      "suite": "accuracy",
      "impl": "dense",
      "n": 2048,
      "d": 8,
      "eps": 0.5,
      "lse_max_abs_err": 7.049782643875346e-06,
      "grad_rel_err": 1.8655037694046649e-06
    },
    {
      "suite": "accuracy",
      "impl": "xla",
      "n": 2048,
      "d": 8,
      "eps": 0.5,
      "lse_max_abs_err": 7.049782643875346e-06,
      "grad_rel_err": 1.813457059322335e-06
    },
    {
      "suite": "accuracy",
      "impl": "dense",
      "n": 2048,
      "d": 8,
      "eps": 0.01,
      "lse_max_abs_err": 0.0004763496506825504,
      "grad_rel_err": 4.946875479054067e-05
    },
    {
      "suite": "accuracy",
      "impl": "xla",
      "n": 2048,
      "d": 8,
      "eps": 0.01,
      "lse_max_abs_err": 0.0004763496506825504,
      "grad_rel_err": 5.9862297440027995e-05
    },
    {
      "suite": "accuracy",
      "impl": "dense",
      "n": 2048,
      "d": 32,
      "eps": 0.05,
      "lse_max_abs_err": 0.00033767065451684175,
      "grad_rel_err": 3.9025959318168096e-05
    },
    {
      "suite": "accuracy",
      "impl": "xla",
      "n": 2048,
      "d": 32,
      "eps": 0.05,
      "lse_max_abs_err": 0.00033767065451684175,
      "grad_rel_err": 3.7966150040008164e-05
    },
    {
      "suite": "lse",
      "impl": "xla",
      "mode": "fwd",
      "n": 1024,
      "m": 1024,
      "d": 2,
      "dtype": "float32",
      "xla_temp_mb": 4.0078125,
      "xla_arg_mb": 0.01953125,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.0012031959995510988,
      "inner": 74,
      "median_ms": 0.695846486475627,
      "q1_ms": 0.6792885337376533,
      "q3_ms": 0.6970537026948453
    },
    {
      "suite": "lse",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 1024,
      "m": 1024,
      "d": 2,
      "dtype": "float32",
      "xla_temp_mb": 8.0078125,
      "xla_arg_mb": 0.01953125,
      "xla_out_mb": 0.01955413818359375,
      "compile_s": 0.002828441000019666,
      "inner": 15,
      "median_ms": 2.4373371998080984,
      "q1_ms": 2.4121415333259693,
      "q3_ms": 2.6074133333167993
    },
    {
      "suite": "lse",
      "impl": "dense",
      "mode": "fwd",
      "n": 1024,
      "m": 1024,
      "d": 2,
      "dtype": "float32",
      "xla_temp_mb": 4.00390625,
      "xla_arg_mb": 0.01953125,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.0008575920001021586,
      "inner": 79,
      "median_ms": 0.6639463797541736,
      "q1_ms": 0.6013199240499754,
      "q3_ms": 0.6719753607596224
    },
    {
      "suite": "lse",
      "impl": "dense",
      "mode": "fwd_bwd",
      "n": 1024,
      "m": 1024,
      "d": 2,
      "dtype": "float32",
      "xla_temp_mb": 12.0,
      "xla_arg_mb": 0.01953125,
      "xla_out_mb": 0.01955413818359375,
      "compile_s": 0.003797534001932945,
      "inner": 16,
      "median_ms": 2.6612379374455486,
      "q1_ms": 2.5103593750372966,
      "q3_ms": 2.782267843713271
    },
    {
      "suite": "lse",
      "impl": "xla",
      "mode": "fwd",
      "n": 4096,
      "m": 4096,
      "d": 2,
      "dtype": "float32",
      "xla_temp_mb": 16.06279754638672,
      "xla_arg_mb": 0.078125,
      "xla_out_mb": 0.015625,
      "compile_s": 0.005686386997695081,
      "inner": 8,
      "median_ms": 6.885971375595545,
      "q1_ms": 6.737174250702083,
      "q3_ms": 7.263693437380425
    },
    {
      "suite": "lse",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 4096,
      "m": 4096,
      "d": 2,
      "dtype": "float32",
      "xla_temp_mb": 16.149063110351562,
      "xla_arg_mb": 0.078125,
      "xla_out_mb": 0.07814788818359375,
      "compile_s": 0.030295677002868615,
      "inner": 1,
      "median_ms": 29.85711700603133,
      "q1_ms": 29.587001001345925,
      "q3_ms": 31.14183250363567
    },
    {
      "suite": "lse",
      "impl": "dense",
      "mode": "fwd",
      "n": 4096,
      "m": 4096,
      "d": 2,
      "dtype": "float32",
      "xla_temp_mb": 64.015625,
      "xla_arg_mb": 0.078125,
      "xla_out_mb": 0.015625,
      "compile_s": 0.02034215599996969,
      "inner": 2,
      "median_ms": 23.729013002594,
      "q1_ms": 19.753371250772034,
      "q3_ms": 23.92545975089888
    },
    {
      "suite": "lse",
      "impl": "dense",
      "mode": "fwd_bwd",
      "n": 4096,
      "m": 4096,
      "d": 2,
      "dtype": "float32",
      "xla_temp_mb": 192.0625,
      "xla_arg_mb": 0.078125,
      "xla_out_mb": 0.07814788818359375,
      "compile_s": 0.08817098099825671,
      "inner": 1,
      "median_ms": 94.77186999720288,
      "q1_ms": 92.00915849942248,
      "q3_ms": 95.35365049669053
    },
    {
      "suite": "lse",
      "impl": "xla",
      "mode": "fwd",
      "n": 16384,
      "m": 16384,
      "d": 2,
      "dtype": "float32",
      "xla_temp_mb": 16.06279754638672,
      "xla_arg_mb": 0.3125,
      "xla_out_mb": 0.0625,
      "compile_s": 0.09917682399827754,
      "inner": 1,
      "median_ms": 101.05651900084922,
      "q1_ms": 93.85095199831994,
      "q3_ms": 109.73045950231608
    },
    {
      "suite": "lse",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 16384,
      "m": 16384,
      "d": 2,
      "dtype": "float32",
      "xla_temp_mb": 16.242813110351562,
      "xla_arg_mb": 0.3125,
      "xla_out_mb": 0.31252288818359375,
      "compile_s": 0.4916773069999181,
      "inner": 1,
      "median_ms": 485.34456599736586,
      "q1_ms": 471.9210650000605,
      "q3_ms": 488.86966049758485
    },
    {
      "suite": "lse",
      "impl": "dense",
      "mode": "fwd",
      "n": 16384,
      "m": 16384,
      "d": 2,
      "dtype": "float32",
      "xla_temp_mb": 1024.0625,
      "xla_arg_mb": 0.3125,
      "xla_out_mb": 0.0625,
      "compile_s": 0.3270827800006373,
      "inner": 1,
      "median_ms": 327.88253300532233,
      "q1_ms": 320.5972500036296,
      "q3_ms": 330.45162850248744
    },
    {
      "suite": "lse",
      "impl": "dense",
      "mode": "fwd_bwd",
      "n": 16384,
      "m": 16384,
      "d": 2,
      "dtype": "float32",
      "xla_temp_mb": 3072.25,
      "xla_arg_mb": 0.3125,
      "xla_out_mb": 0.31252288818359375,
      "compile_s": 1.455930434000038,
      "inner": 1,
      "median_ms": 1386.5179920030641,
      "q1_ms": 1379.0914470009739,
      "q3_ms": 1387.385643502057
    },
    {
      "suite": "lse",
      "impl": "xla",
      "mode": "fwd",
      "n": 1024,
      "m": 1024,
      "d": 8,
      "dtype": "float32",
      "xla_temp_mb": 4.0078125,
      "xla_arg_mb": 0.06640625,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.000811865997093264,
      "inner": 90,
      "median_ms": 0.6979301999510628,
      "q1_ms": 0.6848841277840418,
      "q3_ms": 0.7066878777550301
    },
    {
      "suite": "lse",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 1024,
      "m": 1024,
      "d": 8,
      "dtype": "float32",
      "xla_temp_mb": 8.0,
      "xla_arg_mb": 0.06640625,
      "xla_out_mb": 0.06642913818359375,
      "compile_s": 0.005386730997997802,
      "inner": 11,
      "median_ms": 3.574220273135738,
      "q1_ms": 3.33913595452693,
      "q3_ms": 3.932123727281578
    },
    {
      "suite": "lse",
      "impl": "dense",
      "mode": "fwd",
      "n": 1024,
      "m": 1024,
      "d": 8,
      "dtype": "float32",
      "xla_temp_mb": 4.00390625,
      "xla_arg_mb": 0.06640625,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.0009889509965432808,
      "inner": 70,
      "median_ms": 0.6854073000015758,
      "q1_ms": 0.6323419785725751,
      "q3_ms": 0.7220235714547536
    },
    {
      "suite": "lse",
      "impl": "dense",
      "mode": "fwd_bwd",
      "n": 1024,
      "m": 1024,
      "d": 8,
      "dtype": "float32",
      "xla_temp_mb": 12.0,
      "xla_arg_mb": 0.06640625,
      "xla_out_mb": 0.06642913818359375,
      "compile_s": 0.003446510003414005,
      "inner": 13,
      "median_ms": 3.372639154034774,
      "q1_ms": 3.266187769230992,
      "q3_ms": 3.5945929230365437
    },
    {
      "suite": "lse",
      "impl": "xla",
      "mode": "fwd",
      "n": 4096,
      "m": 4096,
      "d": 8,
      "dtype": "float32",
      "xla_temp_mb": 16.15654754638672,
      "xla_arg_mb": 0.265625,
      "xla_out_mb": 0.015625,
      "compile_s": 0.005641843999910634,
      "inner": 8,
      "median_ms": 6.845144124781655,
      "q1_ms": 6.844330062449444,
      "q3_ms": 6.959172374990885
    },
    {
      "suite": "lse",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 4096,
      "m": 4096,
      "d": 8,
      "dtype": "float32",
      "xla_temp_mb": 16.430313110351562,
      "xla_arg_mb": 0.265625,
      "xla_out_mb": 0.26564788818359375,
      "compile_s": 0.038663017003273126,
      "inner": 1,
      "median_ms": 32.87997600273229,
      "q1_ms": 31.606642001861474,
      "q3_ms": 33.26147750340169
    },
    {
      "suite": "lse",
      "impl": "dense",
      "mode": "fwd",
      "n": 4096,
      "m": 4096,
      "d": 8,
      "dtype": "float32",
      "xla_temp_mb": 64.015625,
      "xla_arg_mb": 0.265625,
      "xla_out_mb": 0.015625,
      "compile_s": 0.019628444999398198,
      "inner": 2,
      "median_ms": 17.621753500861814,
      "q1_ms": 16.995994748867815,
      "q3_ms": 18.76378174893034
    },
    {
      "suite": "lse",
      "impl": "dense",
      "mode": "fwd_bwd",
      "n": 4096,
      "m": 4096,
      "d": 8,
      "dtype": "float32",
      "xla_temp_mb": 192.25,
      "xla_arg_mb": 0.265625,
      "xla_out_mb": 0.26564788818359375,
      "compile_s": 0.08397125599731226,
      "inner": 1,
      "median_ms": 84.81997100170702,
      "q1_ms": 84.4031630003883,
      "q3_ms": 85.72484850083129
    },
    {
      "suite": "lse",
      "impl": "xla",
      "mode": "fwd",
      "n": 16384,
      "m": 16384,
      "d": 8,
      "dtype": "float32",
      "xla_temp_mb": 16.15654754638672,
      "xla_arg_mb": 1.0625,
      "xla_out_mb": 0.0625,
      "compile_s": 0.09126040599949192,
      "inner": 1,
      "median_ms": 92.97091199550778,
      "q1_ms": 89.63534149734187,
      "q3_ms": 94.58905849896837
    },
    {
      "suite": "lse",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 16384,
      "m": 16384,
      "d": 8,
      "dtype": "float32",
      "xla_temp_mb": 16.524063110351562,
      "xla_arg_mb": 1.0625,
      "xla_out_mb": 1.0625228881835938,
      "compile_s": 0.4663486879944685,
      "inner": 1,
      "median_ms": 464.76079399872106,
      "q1_ms": 460.82487749663414,
      "q3_ms": 504.56859649784747
    },
    {
      "suite": "lse",
      "impl": "dense",
      "mode": "fwd",
      "n": 16384,
      "m": 16384,
      "d": 8,
      "dtype": "float32",
      "xla_temp_mb": 1024.0625,
      "xla_arg_mb": 1.0625,
      "xla_out_mb": 0.0625,
      "compile_s": 0.31062384100368945,
      "inner": 1,
      "median_ms": 301.75665600108914,
      "q1_ms": 300.5753504985478,
      "q3_ms": 301.92717850150075
    },
    {
      "suite": "lse",
      "impl": "dense",
      "mode": "fwd_bwd",
      "n": 16384,
      "m": 16384,
      "d": 8,
      "dtype": "float32",
      "xla_temp_mb": 3073.0,
      "xla_arg_mb": 1.0625,
      "xla_out_mb": 1.0625228881835938,
      "compile_s": 1.2427325430035125,
      "inner": 1,
      "median_ms": 1246.2570019997656,
      "q1_ms": 1237.0046875003027,
      "q3_ms": 1273.768345501594
    },
    {
      "suite": "lse",
      "impl": "xla",
      "mode": "fwd",
      "n": 1024,
      "m": 1024,
      "d": 32,
      "dtype": "float32",
      "xla_temp_mb": 4.0078125,
      "xla_arg_mb": 0.25390625,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.0008130679998430423,
      "inner": 66,
      "median_ms": 0.684599287827344,
      "q1_ms": 0.664854053011919,
      "q3_ms": 0.688850053021187
    },
    {
      "suite": "lse",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 1024,
      "m": 1024,
      "d": 32,
      "dtype": "float32",
      "xla_temp_mb": 8.0,
      "xla_arg_mb": 0.25390625,
      "xla_out_mb": 0.25392913818359375,
      "compile_s": 0.002397425996605307,
      "inner": 18,
      "median_ms": 2.7794607222959815,
      "q1_ms": 2.7403703613041825,
      "q3_ms": 2.79689063892066
    },
    {
      "suite": "lse",
      "impl": "dense",
      "mode": "fwd",
      "n": 1024,
      "m": 1024,
      "d": 32,
      "dtype": "float32",
      "xla_temp_mb": 4.00390625,
      "xla_arg_mb": 0.25390625,
      "xla_out_mb": 0.00390625,
      "compile_s": 0.0008235780042014085,
      "inner": 66,
      "median_ms": 0.6435004696399891,
      "q1_ms": 0.6349215908619843,
      "q3_ms": 0.6554584772609153
    },
    {
      "suite": "lse",
      "impl": "dense",
      "mode": "fwd_bwd",
      "n": 1024,
      "m": 1024,
      "d": 32,
      "dtype": "float32",
      "xla_temp_mb": 12.0,
      "xla_arg_mb": 0.25390625,
      "xla_out_mb": 0.25392913818359375,
      "compile_s": 0.002922339001088403,
      "inner": 15,
      "median_ms": 3.066556333214976,
      "q1_ms": 3.0309857000247575,
      "q3_ms": 3.1799712332334216
    },
    {
      "suite": "lse",
      "impl": "xla",
      "mode": "fwd",
      "n": 4096,
      "m": 4096,
      "d": 32,
      "dtype": "float32",
      "xla_temp_mb": 16.53154754638672,
      "xla_arg_mb": 1.015625,
      "xla_out_mb": 0.015625,
      "compile_s": 0.0070826390001457185,
      "inner": 6,
      "median_ms": 7.157474500369669,
      "q1_ms": 7.121992416311211,
      "q3_ms": 7.326476750070773
    },
    {
      "suite": "lse",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 4096,
      "m": 4096,
      "d": 32,
      "dtype": "float32",
      "xla_temp_mb": 17.555313110351562,
      "xla_arg_mb": 1.015625,
      "xla_out_mb": 1.0156478881835938,
      "compile_s": 0.03319759700389113,
      "inner": 1,
      "median_ms": 31.86119800375309,
      "q1_ms": 31.646682502469048,
      "q3_ms": 32.04437450040132
    },
    {
      "suite": "lse",
      "impl": "dense",
      "mode": "fwd",
      "n": 4096,
      "m": 4096,
      "d": 32,
      "dtype": "float32",
      "xla_temp_mb": 64.015625,
      "xla_arg_mb": 1.015625,
      "xla_out_mb": 0.015625,
      "compile_s": 0.01823972800048068,
      "inner": 2,
      "median_ms": 19.173193999449722,
      "q1_ms": 17.708958748698933,
      "q3_ms": 21.956447249976918
    },
    {
      "suite": "lse",
      "impl": "dense",
      "mode": "fwd_bwd",
      "n": 4096,
      "m": 4096,
      "d": 32,
      "dtype": "float32",
      "xla_temp_mb": 193.0,
      "xla_arg_mb": 1.015625,
      "xla_out_mb": 1.0156478881835938,
      "compile_s": 0.07953334900230402,
      "inner": 1,
      "median_ms": 78.33037399541354,
      "q1_ms": 77.63169699683203,
      "q3_ms": 82.05919849569909
    },
    {
      "suite": "lse",
      "impl": "xla",
      "mode": "fwd",
      "n": 16384,
      "m": 16384,
      "d": 32,
      "dtype": "float32",
      "xla_temp_mb": 16.53154754638672,
      "xla_arg_mb": 4.0625,
      "xla_out_mb": 0.0625,
      "compile_s": 0.10849915000289911,
      "inner": 1,
      "median_ms": 104.5729020042927,
      "q1_ms": 103.78711550220032,
      "q3_ms": 105.85014850221341
    },
    {
      "suite": "lse",
      "impl": "xla",
      "mode": "fwd_bwd",
      "n": 16384,
      "m": 16384,
      "d": 32,
      "dtype": "float32",
      "xla_temp_mb": 17.649063110351562,
      "xla_arg_mb": 4.0625,
      "xla_out_mb": 4.062522888183594,
      "compile_s": 0.48630854000657564,
      "inner": 1,
      "median_ms": 482.8754350019153,
      "q1_ms": 482.2413055007928,
      "q3_ms": 490.7335015013814
    },
    {
      "suite": "lse",
      "impl": "dense",
      "mode": "fwd",
      "n": 16384,
      "m": 16384,
      "d": 32,
      "dtype": "float32",
      "xla_temp_mb": 1024.0625,
      "xla_arg_mb": 4.0625,
      "xla_out_mb": 0.0625,
      "compile_s": 0.2990920739975991,
      "inner": 1,
      "median_ms": 303.0065100028878,
      "q1_ms": 300.68296899844427,
      "q3_ms": 305.30375100352103
    },
    {
      "suite": "lse",
      "impl": "dense",
      "mode": "fwd_bwd",
      "n": 16384,
      "m": 16384,
      "d": 32,
      "dtype": "float32",
      "xla_temp_mb": 3076.0,
      "xla_arg_mb": 4.0625,
      "xla_out_mb": 4.062522888183594,
      "compile_s": 1.2040903570014052,
      "inner": 1,
      "median_ms": 1199.815068001044,
      "q1_ms": 1191.813359499065,
      "q3_ms": 1205.391231997055
    },
    {
      "suite": "sinkhorn",
      "impl": "kernel",
      "mode": "fwd",
      "n": 4096,
      "m": 4096,
      "d": 8,
      "iters": 50,
      "eps": 0.8975364685058594,
      "xla_temp_mb": 17.00078582763672,
      "xla_arg_mb": 0.25,
      "xla_out_mb": 3.814697265625e-06,
      "compile_s": 0.5325614260000293,
      "inner": 1,
      "median_ms": 535.5183200008469,
      "q1_ms": 533.6475309995876,
      "q3_ms": 535.7087660013349
    },
    {
      "suite": "sinkhorn",
      "impl": "dense",
      "mode": "fwd",
      "n": 4096,
      "m": 4096,
      "d": 8,
      "iters": 50,
      "eps": 0.8975364685058594,
      "xla_temp_mb": 128.82821655273438,
      "xla_arg_mb": 0.25,
      "xla_out_mb": 3.814697265625e-06,
      "compile_s": 1.0967473429991514,
      "inner": 1,
      "median_ms": 1110.2454100036994,
      "q1_ms": 1101.2647099996684,
      "q3_ms": 1123.27438450302
    },
    {
      "suite": "sinkhorn",
      "impl": "kernel",
      "mode": "fwd_bwd_implicit",
      "n": 4096,
      "m": 4096,
      "d": 8,
      "iters": 50,
      "eps": 0.8975364685058594,
      "xla_temp_mb": 17.0010986328125,
      "xla_arg_mb": 0.25,
      "xla_out_mb": 0.125,
      "compile_s": 0.5469624719989952,
      "inner": 1,
      "median_ms": 537.8922510062694,
      "q1_ms": 537.8245535030146,
      "q3_ms": 542.4304270018183
    },
    {
      "suite": "sinkhorn",
      "impl": "agreement",
      "n": 4096,
      "d": 8,
      "cost_kernel": 14.728632926940918,
      "cost_dense": 14.728632926940918,
      "abs_diff": 0.0
    },
    {
      "suite": "sinkhorn",
      "impl": "kernel",
      "mode": "fwd",
      "n": 16384,
      "m": 16384,
      "d": 8,
      "iters": 50,
      "eps": 0.8991072654724122,
      "xla_temp_mb": 19.53203582763672,
      "xla_arg_mb": 1.0,
      "xla_out_mb": 3.814697265625e-06,
      "compile_s": 8.519671039997775,
      "inner": 1,
      "median_ms": 8475.268413996673,
      "q1_ms": 8471.344951500214,
      "q3_ms": 8477.599433997966
    },
    {
      "suite": "sinkhorn",
      "impl": "dense",
      "mode": "fwd",
      "n": 16384,
      "m": 16384,
      "d": 8,
      "iters": 50,
      "eps": 0.8991072654724122,
      "xla_temp_mb": 2051.3125915527344,
      "xla_arg_mb": 1.0,
      "xla_out_mb": 3.814697265625e-06,
      "compile_s": -0.07201984700077446,
      "median_ms": 19860.808801997337,
      "q1_ms": 19860.808801997337,
      "q3_ms": 19860.808801997337,
      "inner": 1,
      "repeats": 1
    },
    {
      "suite": "sinkhorn",
      "impl": "kernel",
      "mode": "fwd_bwd_implicit",
      "n": 16384,
      "m": 16384,
      "d": 8,
      "iters": 50,
      "eps": 0.8991072654724122,
      "xla_temp_mb": 19.5323486328125,
      "xla_arg_mb": 1.0,
      "xla_out_mb": 0.5,
      "compile_s": 8.976917567997589,
      "inner": 1,
      "median_ms": 8718.903438995767,
      "q1_ms": 8647.111139496701,
      "q3_ms": 8833.185569998022
    },
    {
      "suite": "sinkhorn",
      "impl": "agreement",
      "n": 16384,
      "d": 8,
      "cost_kernel": 14.778953552246094,
      "cost_dense": 14.77895450592041,
      "abs_diff": 9.5367431640625e-07
    }
  ]
}
