{
  "schema_version": 1,
  "protocol": {
    "scope": "complete recurrent block including input gradient",
    "components": "norm, x/y projections, depthwise conv, mixer, join/output, norm and gated MLP",
    "dtype": "BF16 autocast, FP32 recurrent carry",
    "order": "AB/BA",
    "compile_excluded": true
  },
  "environment": {
    "torch": "2.1.2+cu118",
    "cuda": "11.8",
    "gpu": "NVIDIA H800 PCIe"
  },
  "cases": [
    {
      "case": "state2048_short",
      "batch": 4,
      "length": 2048,
      "config": {
        "width": 2048,
        "rnn_width": 2048,
        "depth": 1
      },
      "shared_parameters_copied": 50366464,
      "rows": [
        {
          "architecture": "rglru",
          "backend": "fused_grouped_prefix32_hierarchical_chunk16",
          "controller": null,
          "forward": {
            "samples_ms": [
              5.347307771444321,
              5.335729569196701,
              5.339097231626511,
              5.342915654182434,
              5.344334989786148,
              5.455613136291504,
              5.446966737508774,
              5.459748208522797,
              5.465805530548096,
              5.465641617774963
            ],
            "median_ms": 5.397137254476547,
            "mean_ms": 5.400316044688225,
            "std_ms": 0.05871989194076264,
            "mad_ms": 0.05825795233249664,
            "minimum_ms": 5.335729569196701,
            "maximum_ms": 5.465805530548096
          },
          "backward": {
            "samples_ms": [
              10.145779699087143,
              10.149993002414703,
              10.102394968271255,
              10.104957967996597,
              10.161563754081726,
              10.161180049180984,
              10.168328881263733,
              10.115295648574829,
              10.089796036481857,
              10.088670998811722
            ],
            "median_ms": 10.130537673830986,
            "mean_ms": 10.128796100616455,
            "std_ms": 0.0300026642240251,
            "mad_ms": 0.029392540454864502,
            "minimum_ms": 10.088670998811722,
            "maximum_ms": 10.168328881263733
          },
          "forward_backward": {
            "samples_ms": [
              15.721060335636139,
              15.757139772176743,
              15.745483338832855,
              15.761561691761017,
              15.76395332813263,
              15.791594982147217,
              15.76320081949234,
              15.878196805715561,
              15.895150601863861,
              15.838410705327988
            ],
            "median_ms": 15.763577073812485,
            "mean_ms": 15.791575238108635,
            "std_ms": 0.05583566942518686,
            "mad_ms": 0.023055821657180786,
            "minimum_ms": 15.721060335636139,
            "maximum_ms": 15.895150601863861
          },
          "baseline_allocated_bytes": 740789760,
          "baseline_reserved_bytes": 859832320,
          "peak_allocated_bytes": 2430198784,
          "peak_reserved_bytes": 2707423232,
          "peak_allocated_delta_bytes": 1689409024,
          "peak_reserved_delta_bytes": 1847590912,
          "parameters": 50896896
        },
        {
          "architecture": "samu",
          "backend": "fused_output_fused_write_shared_sfu_chunk32",
          "controller": "fp32_fused_coords",
          "forward": {
            "samples_ms": [
              4.9464404582977295,
              4.948820918798447,
              4.935767501592636,
              4.94685024023056,
              4.935722798109055,
              4.9369484186172485,
              4.921190440654755,
              4.921916872262955,
              4.916999489068985,
              4.916984587907791
            ],
            "median_ms": 4.935745149850845,
            "mean_ms": 4.932764172554016,
            "std_ms": 0.011948616198404614,
            "mad_ms": 0.012090429663658142,
            "minimum_ms": 4.916984587907791,
            "maximum_ms": 4.948820918798447
          },
          "backward": {
            "samples_ms": [
              9.63488221168518,
              9.701088070869446,
              9.711313992738724,
              9.66602936387062,
              9.666655212640762,
              9.689494967460632,
              9.631503373384476,
              9.635131806135178,
              9.632434695959091,
              9.670697152614594
            ],
            "median_ms": 9.666342288255692,
            "mean_ms": 9.66392308473587,
            "std_ms": 0.02837906061679067,
            "mad_ms": 0.03133527934551239,
            "minimum_ms": 9.631503373384476,
            "maximum_ms": 9.711313992738724
          },
          "forward_backward": {
            "samples_ms": [
              14.698401093482971,
              14.703880995512009,
              14.706213027238846,
              14.646068215370178,
              14.700077474117279,
              14.762669801712036,
              14.698773622512817,
              14.635603874921799,
              14.66449350118637,
              14.678794890642166
            ],
            "median_ms": 14.698587357997894,
            "mean_ms": 14.689497649669647,
            "std_ms": 0.034129323463850066,
            "mad_ms": 0.013709068298339844,
            "minimum_ms": 14.635603874921799,
            "maximum_ms": 14.762669801712036
          },
          "baseline_allocated_bytes": 742885376,
          "baseline_reserved_bytes": 853540864,
          "peak_allocated_bytes": 2297287680,
          "peak_reserved_bytes": 2571108352,
          "peak_allocated_delta_bytes": 1554402304,
          "peak_reserved_delta_bytes": 1717567488,
          "parameters": 50372612
        }
      ]
    },
    {
      "case": "state2560_medium",
      "batch": 1,
      "length": 8192,
      "config": {
        "width": 2560,
        "rnn_width": 2560,
        "depth": 1
      },
      "shared_parameters_copied": 78686720,
      "rows": [
        {
          "architecture": "rglru",
          "backend": "fused_grouped_prefix32_hierarchical_chunk16",
          "controller": null,
          "forward": {
            "samples_ms": [
              7.069878280162811,
              7.067002356052399,
              7.060721516609192,
              7.054764777421951,
              7.05951452255249,
              7.164750248193741,
              7.148034870624542,
              7.160164415836334,
              7.15840607881546,
              7.168307900428772
            ],
            "median_ms": 7.108956575393677,
            "mean_ms": 7.111154496669769,
            "std_ms": 0.0491700851426799,
            "mad_ms": 0.049445778131484985,
            "minimum_ms": 7.054764777421951,
            "maximum_ms": 7.168307900428772
          },
          "backward": {
            "samples_ms": [
              13.527054339647293,
              13.523668050765991,
              13.489913195371628,
              15.153605490922928,
              13.441618531942368,
              13.825159519910812,
              13.96278291940689,
              13.969838619232178,
              13.853773474693298,
              13.72375339269638
            ],
            "median_ms": 13.774456456303596,
            "mean_ms": 13.847116753458977,
            "std_ms": 0.4743605476636468,
            "mad_ms": 0.22139213979244232,
            "minimum_ms": 13.441618531942368,
            "maximum_ms": 15.153605490922928
          },
          "forward_backward": {
            "samples_ms": [
              20.788218826055527,
              20.737268030643463,
              20.75762301683426,
              20.754892379045486,
              20.830392837524414,
              20.764991641044617,
              20.594246685504913,
              20.556144416332245,
              20.540442317724228,
              20.65485715866089
            ],
            "median_ms": 20.746080204844475,
            "mean_ms": 20.697907730937004,
            "std_ms": 0.09801132836897199,
            "mad_ms": 0.0632256269454956,
            "minimum_ms": 20.540442317724228,
            "maximum_ms": 20.830392837524414
          },
          "baseline_allocated_bytes": 1107914240,
          "baseline_reserved_bytes": 1457520640,
          "peak_allocated_bytes": 3194254848,
          "peak_reserved_bytes": 3678404608,
          "peak_allocated_delta_bytes": 2086340608,
          "peak_reserved_delta_bytes": 2220883968,
          "parameters": 79513600
        },
        {
          "architecture": "samu",
          "backend": "fused_output_fused_write_shared_sfu_grouped_prefix64_hierarchical_chunk32",
          "controller": "fp32_fused_coords",
          "forward": {
            "samples_ms": [
              6.327122449874878,
              6.30984827876091,
              6.350520998239517,
              6.403379142284393,
              6.398376077413559,
              6.550628691911697,
              6.620246917009354,
              6.620582193136215,
              6.7004673182964325,
              9.04301181435585
            ],
            "median_ms": 6.477003917098045,
            "mean_ms": 6.732418388128281,
            "std_ms": 0.7814427119908226,
            "mad_ms": 0.14341063797473907,
            "minimum_ms": 6.30984827876091,
            "maximum_ms": 9.04301181435585
          },
          "backward": {
            "samples_ms": [
              12.950733304023743,
              12.944359332323074,
              12.90886104106903,
              12.896575033664703,
              12.897256761789322,
              13.005301356315613,
              12.985900044441223,
              15.745528042316437,
              12.832503765821457,
              12.818004935979843
            ],
            "median_ms": 12.926610186696053,
            "mean_ms": 13.198502361774445,
            "std_ms": 0.8509041519166118,
            "mad_ms": 0.04466250538825989,
            "minimum_ms": 12.818004935979843,
            "maximum_ms": 15.745528042316437
          },
          "forward_backward": {
            "samples_ms": [
              19.52766627073288,
              19.476141780614853,
              19.451536238193512,
              19.453946501016617,
              19.49303224682808,
              19.562609493732452,
              19.560981541872025,
              19.58225667476654,
              19.627805799245834,
              19.633740186691284
            ],
            "median_ms": 19.544323906302452,
            "mean_ms": 19.536971673369408,
            "std_ms": 0.06388225024819053,
            "mad_ms": 0.05973689258098602,
            "minimum_ms": 19.451536238193512,
            "maximum_ms": 19.633740186691284
          },
          "baseline_allocated_bytes": 1111567360,
          "baseline_reserved_bytes": 1474297856,
          "peak_allocated_bytes": 3070722560,
          "peak_reserved_bytes": 3569352704,
          "peak_allocated_delta_bytes": 1959155200,
          "peak_reserved_delta_bytes": 2095054848,
          "parameters": 78694404
        }
      ]
    },
    {
      "case": "state1024_long",
      "batch": 1,
      "length": 32768,
      "config": {
        "width": 1024,
        "rnn_width": 1024,
        "depth": 1
      },
      "shared_parameters_copied": 12600320,
      "rows": [
        {
          "architecture": "rglru",
          "backend": "fused_grouped_prefix32_hierarchical_chunk16",
          "controller": null,
          "forward": {
            "samples_ms": [
              7.998153567314148,
              7.980726659297943,
              7.983766496181488,
              7.983855903148651,
              7.9807378351688385,
              8.052334189414978,
              8.039973676204681,
              8.084297180175781,
              8.216466754674911,
              8.042663335800171
            ],
            "median_ms": 8.019063621759415,
            "mean_ms": 8.03629755973816,
            "std_ms": 0.0694375438124772,
            "mad_ms": 0.03525242209434509,
            "minimum_ms": 7.980726659297943,
            "maximum_ms": 8.216466754674911
          },
          "backward": {
            "samples_ms": [
              15.958499163389206,
              15.952810645103455,
              15.961125493049622,
              15.955563634634018,
              15.958957374095917,
              15.956249088048935,
              16.085200011730194,
              15.962120145559311,
              15.969861298799515,
              15.995770692825317
            ],
            "median_ms": 15.96004143357277,
            "mean_ms": 15.975615754723549,
            "std_ms": 0.03835081680180496,
            "mad_ms": 0.004135072231292725,
            "minimum_ms": 15.952810645103455,
            "maximum_ms": 16.085200011730194
          },
          "forward_backward": {
            "samples_ms": [
              24.130430072546005,
              24.094510823488235,
              24.08967912197113,
              24.121027439832687,
              24.117056280374527,
              24.119406938552856,
              24.155713617801666,
              24.148281663656235,
              24.13257583975792,
              24.103134870529175
            ],
            "median_ms": 24.120217189192772,
            "mean_ms": 24.121181666851044,
            "std_ms": 0.020474005432722667,
            "mad_ms": 0.014720484614372253,
            "minimum_ms": 24.08967912197113,
            "maximum_ms": 24.155713617801666
          },
          "baseline_allocated_bytes": 355192320,
          "baseline_reserved_bytes": 830472192,
          "peak_allocated_bytes": 3408148992,
          "peak_reserved_bytes": 4055891968,
          "peak_allocated_delta_bytes": 3052956672,
          "peak_reserved_delta_bytes": 3225419776,
          "parameters": 12734464
        },
        {
          "architecture": "samu",
          "backend": "fused_output_fused_write_shared_sfu_grouped_prefix64_hierarchical_chunk32",
          "controller": "fp32_fused_coords",
          "forward": {
            "samples_ms": [
              7.088728249073029,
              7.060360163450241,
              7.071424275636673,
              7.067013531923294,
              7.070552557706833,
              7.075183093547821,
              7.066022604703903,
              7.071830332279205,
              7.0686303079128265,
              7.09080696105957
            ],
            "median_ms": 7.070988416671753,
            "mean_ms": 7.07305520772934,
            "std_ms": 0.009178363826425721,
            "mad_ms": 0.004084780812263489,
            "minimum_ms": 7.060360163450241,
            "maximum_ms": 7.09080696105957
          },
          "backward": {
            "samples_ms": [
              15.06197452545166,
              15.094000846147537,
              15.064351260662079,
              15.073299407958984,
              15.043679624795914,
              15.074390918016434,
              15.068594366312027,
              15.079747885465622,
              15.079066157341003,
              15.045754611492157
            ],
            "median_ms": 15.070946887135506,
            "mean_ms": 15.068485960364342,
            "std_ms": 0.014626665745078735,
            "mad_ms": 0.008460134267807007,
            "minimum_ms": 15.043679624795914,
            "maximum_ms": 15.094000846147537
          },
          "forward_backward": {
            "samples_ms": [
              22.319979965686798,
              22.276848554611206,
              22.254392504692078,
              22.23297953605652,
              22.230494767427444,
              22.27066457271576,
              22.27519080042839,
              22.289808839559555,
              22.313274443149567,
              22.30846881866455
            ],
            "median_ms": 22.2760196775198,
            "mean_ms": 22.277210280299187,
            "std_ms": 0.029953453793187507,
            "mad_ms": 0.027038156986236572,
            "minimum_ms": 22.230494767427444,
            "maximum_ms": 22.319979965686798
          },
          "baseline_allocated_bytes": 357812224,
          "baseline_reserved_bytes": 629145600,
          "peak_allocated_bytes": 3210113536,
          "peak_reserved_bytes": 3787456512,
          "peak_allocated_delta_bytes": 2852301312,
          "peak_reserved_delta_bytes": 3158310912,
          "parameters": 12603396
        }
      ]
    },
    {
      "case": "400m_block_long",
      "batch": 1,
      "length": 32768,
      "config": {
        "width": 1536,
        "rnn_width": 2048,
        "depth": 12
      },
      "shared_parameters_copied": 30700544,
      "rows": [
        {
          "architecture": "rglru",
          "backend": "fused_grouped_prefix32_hierarchical_chunk16",
          "controller": null,
          "forward": {
            "samples_ms": [
              14.75922018289566,
              14.771386981010437,
              14.874733984470367,
              14.831308275461197,
              14.788936823606491,
              14.87557590007782,
              14.8075670003891,
              14.79056105017662,
              14.898911118507385,
              14.907259494066238
            ],
            "median_ms": 14.819437637925148,
            "mean_ms": 14.830546081066132,
            "std_ms": 0.05194700804263495,
            "mad_ms": 0.05167350172996521,
            "minimum_ms": 14.75922018289566,
            "maximum_ms": 14.907259494066238
          },
          "backward": {
            "samples_ms": [
              27.740228921175003,
              27.912277728319168,
              28.123505413532257,
              28.078585863113403,
              27.972780168056488,
              27.983568608760834,
              27.98117697238922,
              27.83338353037834,
              27.753297239542007,
              27.898170053958893
            ],
            "median_ms": 27.942528948187828,
            "mean_ms": 27.92769744992256,
            "std_ms": 0.12028582710936198,
            "mad_ms": 0.07675215601921082,
            "minimum_ms": 27.740228921175003,
            "maximum_ms": 28.123505413532257
          },
          "forward_backward": {
            "samples_ms": [
              43.28601434826851,
              43.5812771320343,
              43.57847571372986,
              43.10421273112297,
              42.788129299879074,
              43.11739280819893,
              42.823903262615204,
              42.99258068203926,
              43.43096911907196,
              43.847497552633286
            ],
            "median_ms": 43.20170357823372,
            "mean_ms": 43.255045264959335,
            "std_ms": 0.3329298334573352,
            "mad_ms": 0.30301883816719055,
            "minimum_ms": 42.788129299879074,
            "maximum_ms": 43.847497552633286
          },
          "baseline_allocated_bytes": 639016448,
          "baseline_reserved_bytes": 1140850688,
          "peak_allocated_bytes": 5614819840,
          "peak_reserved_bytes": 6614417408,
          "peak_allocated_delta_bytes": 4975803392,
          "peak_reserved_delta_bytes": 5473566720,
          "parameters": 31230976
        },
        {
          "architecture": "samu",
          "backend": "fused_output_fused_write_shared_sfu_grouped_prefix64_hierarchical_chunk32",
          "controller": "fp32_fused_coords",
          "forward": {
            "samples_ms": [
              12.931425124406815,
              13.054441660642624,
              13.235893100500107,
              13.307947665452957,
              13.27565684914589,
              13.332057744264603,
              13.45871016383171,
              13.480190187692642,
              13.322822749614716,
              13.389568775892258
            ],
            "median_ms": 13.315385207533836,
            "mean_ms": 13.278871402144432,
            "std_ms": 0.16238108625777667,
            "mad_ms": 0.07683783769607544,
            "minimum_ms": 12.931425124406815,
            "maximum_ms": 13.480190187692642
          },
          "backward": {
            "samples_ms": [
              26.38090029358864,
              26.342429220676422,
              26.305213570594788,
              26.302967220544815,
              26.38551965355873,
              26.22101828455925,
              26.1659175157547,
              26.20123326778412,
              26.121828705072403,
              26.21176466345787
            ],
            "median_ms": 26.261992752552032,
            "mean_ms": 26.263879239559174,
            "std_ms": 0.08726349527806719,
            "mad_ms": 0.07059797644615173,
            "minimum_ms": 26.121828705072403,
            "maximum_ms": 26.38551965355873
          },
          "forward_backward": {
            "samples_ms": [
              39.86521437764168,
              39.86257314682007,
              40.000081062316895,
              39.89472612738609,
              40.39924964308739,
              39.91178423166275,
              39.89989310503006,
              39.985980838537216,
              40.05413129925728,
              39.99108448624611
            ],
            "median_ms": 39.94888253509998,
            "mean_ms": 39.98647183179855,
            "std_ms": 0.150570812573014,
            "mad_ms": 0.052677467465400696,
            "minimum_ms": 39.86257314682007,
            "maximum_ms": 40.39924964308739
          },
          "baseline_allocated_bytes": 641112064,
          "baseline_reserved_bytes": 807403520,
          "peak_allocated_bytes": 5214161408,
          "peak_reserved_bytes": 6146752512,
          "peak_allocated_delta_bytes": 4573049344,
          "peak_reserved_delta_bytes": 5339348992,
          "parameters": 30706692
        }
      ]
    }
  ]
}
