{
  "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": 65536,
        "width": 1024
      },
      "architecture": "rglru",
      "backend": "fused_grouped_prefix32_hierarchical_chunk16",
      "controller": null,
      "forward": {
        "samples_ms": [
          3.271147608757019,
          3.170877695083618,
          3.1821057200431824,
          3.159966319799423,
          3.1504742801189423,
          3.16513329744339,
          3.1564682722091675,
          3.1535476446151733,
          3.148280084133148,
          3.1451433897018433
        ],
        "median_ms": 3.1582172960042953,
        "mean_ms": 3.1703144311904907,
        "std_ms": 0.03525078103880543,
        "minimum_ms": 3.1451433897018433,
        "maximum_ms": 3.271147608757019
      },
      "backward": {
        "samples_ms": [
          5.8892108500003815,
          5.8857761323452,
          5.890596657991409,
          5.9063620865345,
          5.894307047128677,
          5.893893539905548,
          5.888529121875763,
          5.88606670498848,
          5.891323089599609,
          5.893636494874954
        ],
        "median_ms": 5.890959873795509,
        "mean_ms": 5.891970172524452,
        "std_ms": 0.005603937504828212,
        "minimum_ms": 5.8857761323452,
        "maximum_ms": 5.9063620865345
      },
      "forward_backward": {
        "samples_ms": [
          9.16542112827301,
          9.11073386669159,
          9.175024926662445,
          9.188909083604813,
          9.949568659067154,
          9.179148823022842,
          9.127739816904068,
          9.135991334915161,
          9.141877293586731,
          9.14262980222702
        ],
        "median_ms": 9.154025465250015,
        "mean_ms": 9.231704473495483,
        "std_ms": 0.24044445207606024,
        "minimum_ms": 9.11073386669159,
        "maximum_ms": 9.949568659067154
      },
      "rounds": [
        {
          "round": 0,
          "order_index": 0,
          "forward_ms": [
            3.271147608757019,
            3.170877695083618,
            3.1821057200431824,
            3.159966319799423,
            3.1504742801189423
          ],
          "backward_ms": [
            5.8892108500003815,
            5.8857761323452,
            5.890596657991409,
            5.9063620865345,
            5.894307047128677
          ],
          "forward_backward_ms": [
            9.16542112827301,
            9.11073386669159,
            9.175024926662445,
            9.188909083604813,
            9.949568659067154
          ]
        },
        {
          "round": 1,
          "order_index": 1,
          "forward_ms": [
            3.16513329744339,
            3.1564682722091675,
            3.1535476446151733,
            3.148280084133148,
            3.1451433897018433
          ],
          "backward_ms": [
            5.893893539905548,
            5.888529121875763,
            5.88606670498848,
            5.891323089599609,
            5.893636494874954
          ],
          "forward_backward_ms": [
            9.179148823022842,
            9.127739816904068,
            9.135991334915161,
            9.141877293586731,
            9.14262980222702
          ]
        }
      ],
      "peak_allocated_delta_bytes": 1578907136,
      "peak_reserved_delta_bytes": 1579155456,
      "peak_allocated_bytes": 1915016192,
      "peak_reserved_bytes": 1916796928,
      "baseline_allocated_bytes": 336109056,
      "baseline_reserved_bytes": 337641472,
      "parameters": 134144,
      "parameter_bytes": 536576,
      "dynamic_control_values_per_token": 2048,
      "materialized_state_sized_transition": false
    },
    {
      "shape": {
        "batch": 1,
        "length": 65536,
        "width": 1024
      },
      "architecture": "samu",
      "backend": "fused_output_fused_write_shared_sfu_serial_forward_prefix_grouped_prefix64_hierarchical_chunk32",
      "controller": "fp32_fused_coords",
      "forward": {
        "samples_ms": [
          2.787332981824875,
          2.783697098493576,
          2.774178981781006,
          2.7714483439922333,
          2.778671681880951,
          2.764914184808731,
          2.762462943792343,
          2.765856683254242,
          2.7714818716049194,
          2.7801133692264557
        ],
        "median_ms": 2.7728304266929626,
        "mean_ms": 2.7740158140659332,
        "std_ms": 0.007911073140287678,
        "minimum_ms": 2.762462943792343,
        "maximum_ms": 2.787332981824875
      },
      "backward": {
        "samples_ms": [
          3.839004784822464,
          3.8904324173927307,
          3.8809552788734436,
          3.8333795964717865,
          3.8421154022216797,
          3.883209079504013,
          3.8285814225673676,
          3.8494542241096497,
          3.8368217647075653,
          3.8881823420524597
        ],
        "median_ms": 3.8457848131656647,
        "mean_ms": 3.857213631272316,
        "std_ms": 0.023927803450145595,
        "minimum_ms": 3.8285814225673676,
        "maximum_ms": 3.8904324173927307
      },
      "forward_backward": {
        "samples_ms": [
          6.718818098306656,
          6.65910542011261,
          6.656579673290253,
          6.762240082025528,
          6.748136132955551,
          6.6666379570961,
          6.662208586931229,
          6.664544343948364,
          6.665181368589401,
          6.67368620634079
        ],
        "median_ms": 6.6659096628427505,
        "mean_ms": 6.687713786959648,
        "std_ms": 0.03781043538748372,
        "minimum_ms": 6.656579673290253,
        "maximum_ms": 6.762240082025528
      },
      "rounds": [
        {
          "round": 0,
          "order_index": 1,
          "forward_ms": [
            2.787332981824875,
            2.783697098493576,
            2.774178981781006,
            2.7714483439922333,
            2.778671681880951
          ],
          "backward_ms": [
            3.839004784822464,
            3.8904324173927307,
            3.8809552788734436,
            3.8333795964717865,
            3.8421154022216797
          ],
          "forward_backward_ms": [
            6.718818098306656,
            6.65910542011261,
            6.656579673290253,
            6.762240082025528,
            6.748136132955551
          ]
        },
        {
          "round": 1,
          "order_index": 0,
          "forward_ms": [
            2.764914184808731,
            2.762462943792343,
            2.765856683254242,
            2.7714818716049194,
            2.7801133692264557
          ],
          "backward_ms": [
            3.883209079504013,
            3.8285814225673676,
            3.8494542241096497,
            3.8368217647075653,
            3.8881823420524597
          ],
          "forward_backward_ms": [
            6.6666379570961,
            6.662208586931229,
            6.664544343948364,
            6.665181368589401,
            6.67368620634079
          ]
        }
      ],
      "peak_allocated_delta_bytes": 1209810432,
      "peak_reserved_delta_bytes": 1346371584,
      "peak_allocated_bytes": 1546442240,
      "peak_reserved_bytes": 1684013056,
      "baseline_allocated_bytes": 336631808,
      "baseline_reserved_bytes": 337641472,
      "parameters": 3076,
      "parameter_bytes": 12304,
      "dynamic_control_values_per_token": 2,
      "materialized_state_sized_transition": false
    },
    {
      "shape": {
        "batch": 1,
        "length": 131072,
        "width": 1024
      },
      "architecture": "rglru",
      "backend": "fused_grouped_prefix32_hierarchical_chunk16",
      "controller": null,
      "forward": {
        "samples_ms": [
          6.1570703983306885,
          6.108373403549194,
          6.105821579694748,
          6.113626062870026,
          6.11099973320961,
          6.134595721960068,
          6.118878722190857,
          6.102543324232101,
          6.110183894634247,
          6.119489669799805
        ],
        "median_ms": 6.112312898039818,
        "mean_ms": 6.118158251047134,
        "std_ms": 0.015536330063276356,
        "minimum_ms": 6.102543324232101,
        "maximum_ms": 6.1570703983306885
      },
      "backward": {
        "samples_ms": [
          11.513426899909973,
          11.514082551002502,
          11.517025530338287,
          11.52268797159195,
          11.514566838741302,
          11.50330901145935,
          11.50362566113472,
          11.516258120536804,
          11.514715850353241,
          11.518418788909912
        ],
        "median_ms": 11.514641344547272,
        "mean_ms": 11.513811722397804,
        "std_ms": 0.005761928085117605,
        "minimum_ms": 11.50330901145935,
        "maximum_ms": 11.52268797159195
      },
      "forward_backward": {
        "samples_ms": [
          17.787225544452667,
          17.684005200862885,
          17.684079706668854,
          17.692402005195618,
          17.719894647598267,
          17.804257571697235,
          17.76835322380066,
          17.704181373119354,
          17.731361091136932,
          17.684943974018097
        ],
        "median_ms": 17.71203801035881,
        "mean_ms": 17.726070433855057,
        "std_ms": 0.04306265964315891,
        "minimum_ms": 17.684005200862885,
        "maximum_ms": 17.804257571697235
      },
      "rounds": [
        {
          "round": 0,
          "order_index": 1,
          "forward_ms": [
            6.1570703983306885,
            6.108373403549194,
            6.105821579694748,
            6.113626062870026,
            6.11099973320961
          ],
          "backward_ms": [
            11.513426899909973,
            11.514082551002502,
            11.517025530338287,
            11.52268797159195,
            11.514566838741302
          ],
          "forward_backward_ms": [
            17.787225544452667,
            17.684005200862885,
            17.684079706668854,
            17.692402005195618,
            17.719894647598267
          ]
        },
        {
          "round": 1,
          "order_index": 0,
          "forward_ms": [
            6.134595721960068,
            6.118878722190857,
            6.102543324232101,
            6.110183894634247,
            6.119489669799805
          ],
          "backward_ms": [
            11.50330901145935,
            11.50362566113472,
            11.516258120536804,
            11.514715850353241,
            11.518418788909912
          ],
          "forward_backward_ms": [
            17.804257571697235,
            17.76835322380066,
            17.704181373119354,
            17.731361091136932,
            17.684943974018097
          ]
        }
      ],
      "peak_allocated_delta_bytes": 3157538304,
      "peak_reserved_delta_bytes": 3156213760,
      "peak_allocated_bytes": 3762082816,
      "peak_reserved_bytes": 3762290688,
      "baseline_allocated_bytes": 604544512,
      "baseline_reserved_bytes": 606076928,
      "parameters": 134144,
      "parameter_bytes": 536576,
      "dynamic_control_values_per_token": 2048,
      "materialized_state_sized_transition": false
    },
    {
      "shape": {
        "batch": 1,
        "length": 131072,
        "width": 1024
      },
      "architecture": "samu",
      "backend": "fused_output_fused_write_shared_sfu_serial_forward_prefix_grouped_prefix64_hierarchical_chunk32",
      "controller": "fp32_fused_coords",
      "forward": {
        "samples_ms": [
          5.640197545289993,
          5.6227147579193115,
          5.625523626804352,
          5.628306418657303,
          5.642469972372055,
          5.630843341350555,
          5.625154823064804,
          5.626849830150604,
          5.625668913125992,
          5.6240856647491455
        ],
        "median_ms": 5.626259371638298,
        "mean_ms": 5.629181489348412,
        "std_ms": 0.006454607967766797,
        "minimum_ms": 5.6227147579193115,
        "maximum_ms": 5.642469972372055
      },
      "backward": {
        "samples_ms": [
          7.480110973119736,
          7.471136748790741,
          7.465150207281113,
          7.468856871128082,
          7.495954632759094,
          7.461942732334137,
          7.477235049009323,
          7.472682744264603,
          7.4798427522182465,
          7.471449673175812
        ],
        "median_ms": 7.472066208720207,
        "mean_ms": 7.474436238408089,
        "std_ms": 0.009102944548568232,
        "minimum_ms": 7.461942732334137,
        "maximum_ms": 7.495954632759094
      },
      "forward_backward": {
        "samples_ms": [
          13.221655040979385,
          13.16460594534874,
          13.16148042678833,
          13.16608116030693,
          13.208240270614624,
          13.254549354314804,
          13.14820721745491,
          13.138830661773682,
          13.158369809389114,
          13.147063553333282
        ],
        "median_ms": 13.163043186068535,
        "mean_ms": 13.17690834403038,
        "std_ms": 0.03609256695163937,
        "minimum_ms": 13.138830661773682,
        "maximum_ms": 13.254549354314804
      },
      "rounds": [
        {
          "round": 0,
          "order_index": 0,
          "forward_ms": [
            5.640197545289993,
            5.6227147579193115,
            5.625523626804352,
            5.628306418657303,
            5.642469972372055
          ],
          "backward_ms": [
            7.480110973119736,
            7.471136748790741,
            7.465150207281113,
            7.468856871128082,
            7.495954632759094
          ],
          "forward_backward_ms": [
            13.221655040979385,
            13.16460594534874,
            13.16148042678833,
            13.16608116030693,
            13.208240270614624
          ]
        },
        {
          "round": 1,
          "order_index": 1,
          "forward_ms": [
            5.630843341350555,
            5.625154823064804,
            5.626849830150604,
            5.625668913125992,
            5.6240856647491455
          ],
          "backward_ms": [
            7.461942732334137,
            7.477235049009323,
            7.472682744264603,
            7.4798427522182465,
            7.471449673175812
          ],
          "forward_backward_ms": [
            13.254549354314804,
            13.14820721745491,
            13.138830661773682,
            13.158369809389114,
            13.147063553333282
          ]
        }
      ],
      "peak_allocated_delta_bytes": 2419604992,
      "peak_reserved_delta_bytes": 2692743168,
      "peak_allocated_bytes": 3024672256,
      "peak_reserved_bytes": 3298820096,
      "baseline_allocated_bytes": 605067264,
      "baseline_reserved_bytes": 606076928,
      "parameters": 3076,
      "parameter_bytes": 12304,
      "dynamic_control_values_per_token": 2,
      "materialized_state_sized_transition": false
    }
  ]
}
