{
  "schema_version": 1,
  "protocol": {
    "scope": "complete mixer preparation, recurrence, input and parameter gradients",
    "dtype": "BF16 autocast and FP32 recurrent carry",
    "input_gradient": true,
    "order": "AB/BA counterbalanced",
    "compile_excluded": true,
    "baseline_policy": "strongest previously calibrated legal RG-LRU backend per shape",
    "samu_policy": "strongest previously calibrated legal SAMU backend per shape",
    "resets": "independent sequences reset at first token; no internal packed reset"
  },
  "environment": {
    "torch": "2.1.2+cu118",
    "cuda": "11.8",
    "gpu": "NVIDIA H800 PCIe"
  },
  "rows": [
    {
      "shape": {
        "batch": 1,
        "length": 32768,
        "width": 1536
      },
      "architecture": "rglru",
      "backend": "fused_grouped_prefix32_hierarchical_chunk16",
      "controller": null,
      "forward": {
        "samples_ms": [
          2.389702945947647,
          2.372976392507553,
          2.3865364491939545,
          2.3757703602313995,
          2.371106296777725,
          2.3801252245903015,
          2.368360757827759,
          2.375815063714981,
          2.3639388382434845,
          2.367895096540451
        ],
        "median_ms": 2.3743733763694763,
        "mean_ms": 2.3752227425575256,
        "std_ms": 0.00782598395847477,
        "minimum_ms": 2.3639388382434845,
        "maximum_ms": 2.389702945947647
      },
      "backward": {
        "samples_ms": [
          4.514019936323166,
          4.530556499958038,
          4.532966762781143,
          4.512939602136612,
          4.510402679443359,
          4.543960094451904,
          4.541311413049698,
          4.536394029855728,
          4.5271143317222595,
          4.525508731603622
        ],
        "median_ms": 4.528835415840149,
        "mean_ms": 4.527517408132553,
        "std_ms": 0.011270172254551397,
        "minimum_ms": 4.510402679443359,
        "maximum_ms": 4.543960094451904
      },
      "forward_backward": {
        "samples_ms": [
          6.962928920984268,
          6.952270865440369,
          6.960820406675339,
          6.9717057049274445,
          6.951257586479187,
          6.983276456594467,
          6.966836750507355,
          6.991229951381683,
          6.986215710639954,
          6.972014904022217
        ],
        "median_ms": 6.9692712277174,
        "mean_ms": 6.969855725765228,
        "std_ms": 0.013063525711409058,
        "minimum_ms": 6.951257586479187,
        "maximum_ms": 6.991229951381683
      },
      "rounds": [
        {
          "round": 0,
          "order_index": 0,
          "forward_ms": [
            2.389702945947647,
            2.372976392507553,
            2.3865364491939545,
            2.3757703602313995,
            2.371106296777725
          ],
          "backward_ms": [
            4.514019936323166,
            4.530556499958038,
            4.532966762781143,
            4.512939602136612,
            4.510402679443359
          ],
          "forward_backward_ms": [
            6.962928920984268,
            6.952270865440369,
            6.960820406675339,
            6.9717057049274445,
            6.951257586479187
          ]
        },
        {
          "round": 1,
          "order_index": 1,
          "forward_ms": [
            2.3801252245903015,
            2.368360757827759,
            2.375815063714981,
            2.3639388382434845,
            2.367895096540451
          ],
          "backward_ms": [
            4.543960094451904,
            4.541311413049698,
            4.536394029855728,
            4.5271143317222595,
            4.525508731603622
          ],
          "forward_backward_ms": [
            6.983276456594467,
            6.966836750507355,
            6.991229951381683,
            6.986215710639954,
            6.972014904022217
          ]
        }
      ],
      "peak_allocated_delta_bytes": 1184583168,
      "peak_reserved_delta_bytes": 1033895936,
      "peak_allocated_bytes": 1454257152,
      "peak_reserved_bytes": 1472200704,
      "baseline_allocated_bytes": 269673984,
      "baseline_reserved_bytes": 438304768,
      "parameters": 299520,
      "parameter_bytes": 1198080,
      "dynamic_control_values_per_token": 3072,
      "materialized_state_sized_transition": false
    },
    {
      "shape": {
        "batch": 1,
        "length": 32768,
        "width": 1536
      },
      "architecture": "samu",
      "backend": "fused_output_fused_write_shared_sfu_grouped_prefix64_hierarchical_chunk32",
      "controller": "fp32_fused_coords",
      "forward": {
        "samples_ms": [
          1.1837296187877655,
          1.1120252311229706,
          1.119907945394516,
          1.1608824133872986,
          1.1084489524364471,
          1.1698156595230103,
          1.1411979794502258,
          1.131061464548111,
          1.1484809219837189,
          1.1093243956565857
        ],
        "median_ms": 1.1361297219991684,
        "mean_ms": 1.138487458229065,
        "std_ms": 0.02547635911893841,
        "minimum_ms": 1.1084489524364471,
        "maximum_ms": 1.1837296187877655
      },
      "backward": {
        "samples_ms": [
          2.9172562062740326,
          2.976786345243454,
          2.9165931046009064,
          2.920381724834442,
          2.9150694608688354,
          2.943318337202072,
          2.978898584842682,
          2.9166191816329956,
          2.948898822069168,
          2.94668972492218
        ],
        "median_ms": 2.931850031018257,
        "mean_ms": 2.938051149249077,
        "std_ms": 0.023625286869806188,
        "minimum_ms": 2.9150694608688354,
        "maximum_ms": 2.978898584842682
      },
      "forward_backward": {
        "samples_ms": [
          4.095505923032761,
          4.15310263633728,
          4.0818266570568085,
          4.1765570640563965,
          4.074271768331528,
          4.149794578552246,
          4.066061228513718,
          4.1770972311496735,
          4.096284508705139,
          4.163149744272232
        ],
        "median_ms": 4.123039543628693,
        "mean_ms": 4.123365134000778,
        "std_ms": 0.04220308826592776,
        "minimum_ms": 4.066061228513718,
        "maximum_ms": 4.1770972311496735
      },
      "rounds": [
        {
          "round": 0,
          "order_index": 1,
          "forward_ms": [
            1.1837296187877655,
            1.1120252311229706,
            1.119907945394516,
            1.1608824133872986,
            1.1084489524364471
          ],
          "backward_ms": [
            2.9172562062740326,
            2.976786345243454,
            2.9165931046009064,
            2.920381724834442,
            2.9150694608688354
          ],
          "forward_backward_ms": [
            4.095505923032761,
            4.15310263633728,
            4.0818266570568085,
            4.1765570640563965,
            4.074271768331528
          ]
        },
        {
          "round": 1,
          "order_index": 0,
          "forward_ms": [
            1.1698156595230103,
            1.1411979794502258,
            1.131061464548111,
            1.1484809219837189,
            1.1093243956565857
          ],
          "backward_ms": [
            2.943318337202072,
            2.978898584842682,
            2.9166191816329956,
            2.948898822069168,
            2.94668972492218
          ],
          "forward_backward_ms": [
            4.149794578552246,
            4.066061228513718,
            4.1770972311496735,
            4.096284508705139,
            4.163149744272232
          ]
        }
      ],
      "peak_allocated_delta_bytes": 906910208,
      "peak_reserved_delta_bytes": 807403520,
      "peak_allocated_bytes": 1177762304,
      "peak_reserved_bytes": 1247805440,
      "baseline_allocated_bytes": 270852096,
      "baseline_reserved_bytes": 440401920,
      "parameters": 4612,
      "parameter_bytes": 18448,
      "dynamic_control_values_per_token": 2,
      "materialized_state_sized_transition": false
    },
    {
      "shape": {
        "batch": 1,
        "length": 32768,
        "width": 2560
      },
      "architecture": "rglru",
      "backend": "fused_grouped_prefix32_hierarchical_chunk16",
      "controller": null,
      "forward": {
        "samples_ms": [
          4.28888201713562,
          3.712274134159088,
          3.71343269944191,
          3.703836351633072,
          3.7102028727531433,
          3.715142607688904,
          3.707483410835266,
          3.7059634923934937,
          3.7149041891098022,
          3.7122368812561035
        ],
        "median_ms": 3.712255507707596,
        "mean_ms": 3.7684358656406403,
        "std_ms": 0.17351945147405634,
        "minimum_ms": 3.703836351633072,
        "maximum_ms": 4.28888201713562
      },
      "backward": {
        "samples_ms": [
          7.240545004606247,
          7.33591616153717,
          7.251538336277008,
          7.251013070344925,
          7.265616208314896,
          7.24266842007637,
          7.275287061929703,
          7.257889956235886,
          7.321204990148544,
          7.259495556354523
        ],
        "median_ms": 7.258692756295204,
        "mean_ms": 7.270117476582527,
        "std_ms": 0.030962669539308943,
        "minimum_ms": 7.240545004606247,
        "maximum_ms": 7.33591616153717
      },
      "forward_backward": {
        "samples_ms": [
          11.084601283073425,
          11.066894978284836,
          11.071357876062393,
          11.075679212808609,
          11.075370013713837,
          11.256683617830276,
          11.069491505622864,
          11.07911765575409,
          11.056050658226013,
          11.061359196901321
        ],
        "median_ms": 11.073363944888115,
        "mean_ms": 11.089660599827766,
        "std_ms": 0.05623503542941832,
        "minimum_ms": 11.056050658226013,
        "maximum_ms": 11.256683617830276
      },
      "rounds": [
        {
          "round": 0,
          "order_index": 1,
          "forward_ms": [
            4.28888201713562,
            3.712274134159088,
            3.71343269944191,
            3.703836351633072,
            3.7102028727531433
          ],
          "backward_ms": [
            7.240545004606247,
            7.33591616153717,
            7.251538336277008,
            7.251013070344925,
            7.265616208314896
          ],
          "forward_backward_ms": [
            11.084601283073425,
            11.066894978284836,
            11.071357876062393,
            11.075679212808609,
            11.075370013713837
          ]
        },
        {
          "round": 1,
          "order_index": 0,
          "forward_ms": [
            3.715142607688904,
            3.707483410835266,
            3.7059634923934937,
            3.7149041891098022,
            3.7122368812561035
          ],
          "backward_ms": [
            7.24266842007637,
            7.275287061929703,
            7.257889956235886,
            7.321204990148544,
            7.259495556354523
          ],
          "forward_backward_ms": [
            11.256683617830276,
            11.069491505622864,
            11.07911765575409,
            11.056050658226013,
            11.061359196901321
          ]
        }
      ],
      "peak_allocated_delta_bytes": 1974959616,
      "peak_reserved_delta_bytes": 1973420032,
      "peak_allocated_bytes": 2380985344,
      "peak_reserved_bytes": 2399141888,
      "baseline_allocated_bytes": 406025728,
      "baseline_reserved_bytes": 425721856,
      "parameters": 826880,
      "parameter_bytes": 3307520,
      "dynamic_control_values_per_token": 5120,
      "materialized_state_sized_transition": false
    },
    {
      "shape": {
        "batch": 1,
        "length": 32768,
        "width": 2560
      },
      "architecture": "samu",
      "backend": "fused_output_fused_write_shared_sfu_grouped_prefix64_hierarchical_chunk32",
      "controller": "fp32_fused_coords",
      "forward": {
        "samples_ms": [
          1.7830990254878998,
          1.7849691212177277,
          1.8691010773181915,
          1.8581636250019073,
          1.8573403358459473,
          1.8583685159683228,
          1.8496662378311157,
          1.8552429974079132,
          1.8550865352153778,
          1.8640197813510895
        ],
        "median_ms": 1.8562916666269302,
        "mean_ms": 1.8435057252645493,
        "std_ms": 0.03014703233957401,
        "minimum_ms": 1.7830990254878998,
        "maximum_ms": 1.8691010773181915
      },
      "backward": {
        "samples_ms": [
          4.745423793792725,
          4.816796630620956,
          4.805214703083038,
          4.739426076412201,
          4.792366176843643,
          4.744425415992737,
          4.749830812215805,
          4.746049642562866,
          4.7463104128837585,
          4.795148968696594
        ],
        "median_ms": 4.748070612549782,
        "mean_ms": 4.768099263310432,
        "std_ms": 0.028741981144993715,
        "minimum_ms": 4.739426076412201,
        "maximum_ms": 4.816796630620956
      },
      "forward_backward": {
        "samples_ms": [
          6.575681269168854,
          6.648968905210495,
          6.649278104305267,
          6.642796099185944,
          6.612848490476608,
          6.63086399435997,
          6.619419902563095,
          6.645131856203079,
          6.591968238353729,
          6.666526198387146
        ],
        "median_ms": 6.636830046772957,
        "mean_ms": 6.628348305821419,
        "std_ms": 0.02689769278027944,
        "minimum_ms": 6.575681269168854,
        "maximum_ms": 6.666526198387146
      },
      "rounds": [
        {
          "round": 0,
          "order_index": 0,
          "forward_ms": [
            1.7830990254878998,
            1.7849691212177277,
            1.8691010773181915,
            1.8581636250019073,
            1.8573403358459473
          ],
          "backward_ms": [
            4.745423793792725,
            4.816796630620956,
            4.805214703083038,
            4.739426076412201,
            4.792366176843643
          ],
          "forward_backward_ms": [
            6.575681269168854,
            6.648968905210495,
            6.649278104305267,
            6.642796099185944,
            6.612848490476608
          ]
        },
        {
          "round": 1,
          "order_index": 1,
          "forward_ms": [
            1.8583685159683228,
            1.8496662378311157,
            1.8552429974079132,
            1.8550865352153778,
            1.8640197813510895
          ],
          "backward_ms": [
            4.744425415992737,
            4.749830812215805,
            4.746049642562866,
            4.7463104128837585,
            4.795148968696594
          ],
          "forward_backward_ms": [
            6.63086399435997,
            6.619419902563095,
            6.645131856203079,
            6.591968238353729,
            6.666526198387146
          ]
        }
      ],
      "peak_allocated_delta_bytes": 1510904320,
      "peak_reserved_delta_bytes": 1679818752,
      "peak_allocated_bytes": 1920205312,
      "peak_reserved_bytes": 2105540608,
      "baseline_allocated_bytes": 409300992,
      "baseline_reserved_bytes": 425721856,
      "parameters": 7684,
      "parameter_bytes": 30736,
      "dynamic_control_values_per_token": 2,
      "materialized_state_sized_transition": false
    }
  ]
}
