{
  "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": 8192,
        "width": 1024
      },
      "architecture": "rglru",
      "backend": "fused_grouped_prefix32_hierarchical_chunk16",
      "controller": null,
      "forward": {
        "samples_ms": [
          0.5974024534225464,
          0.5765333771705627,
          0.5573146045207977,
          0.5576349794864655,
          0.5695521831512451,
          0.5522221326828003,
          0.5697868764400482,
          0.5691647529602051,
          0.5590580403804779,
          0.5523860454559326
        ],
        "median_ms": 0.5641113966703415,
        "mean_ms": 0.5661055445671082,
        "std_ms": 0.01304874963721331,
        "minimum_ms": 0.5522221326828003,
        "maximum_ms": 0.5974024534225464
      },
      "backward": {
        "samples_ms": [
          0.9922124445438385,
          0.9772256016731262,
          0.9884759783744812,
          0.9790100157260895,
          0.9777732193470001,
          0.982385128736496,
          0.9699389338493347,
          0.9612590074539185,
          0.9773150086402893,
          0.9684748947620392
        ],
        "median_ms": 0.9775441139936447,
        "mean_ms": 0.9774070233106613,
        "std_ms": 0.008749643188036806,
        "minimum_ms": 0.9612590074539185,
        "maximum_ms": 0.9922124445438385
      },
      "forward_backward": {
        "samples_ms": [
          1.6032829880714417,
          1.5832781791687012,
          1.6130805015563965,
          1.6160868108272552,
          1.5862546861171722,
          1.604236662387848,
          1.6018524765968323,
          1.6023293137550354,
          1.5725158154964447,
          1.6032606363296509
        ],
        "median_ms": 1.6027949750423431,
        "mean_ms": 1.5986178070306778,
        "std_ms": 0.012972780015257398,
        "minimum_ms": 1.5725158154964447,
        "maximum_ms": 1.6160868108272552
      },
      "rounds": [
        {
          "round": 0,
          "order_index": 0,
          "forward_ms": [
            0.5974024534225464,
            0.5765333771705627,
            0.5573146045207977,
            0.5576349794864655,
            0.5695521831512451
          ],
          "backward_ms": [
            0.9922124445438385,
            0.9772256016731262,
            0.9884759783744812,
            0.9790100157260895,
            0.9777732193470001
          ],
          "forward_backward_ms": [
            1.6032829880714417,
            1.5832781791687012,
            1.6130805015563965,
            1.6160868108272552,
            1.5862546861171722
          ]
        },
        {
          "round": 1,
          "order_index": 1,
          "forward_ms": [
            0.5522221326828003,
            0.5697868764400482,
            0.5691647529602051,
            0.5590580403804779,
            0.5523860454559326
          ],
          "backward_ms": [
            0.982385128736496,
            0.9699389338493347,
            0.9612590074539185,
            0.9773150086402893,
            0.9684748947620392
          ],
          "forward_backward_ms": [
            1.604236662387848,
            1.6018524765968323,
            1.6023293137550354,
            1.5725158154964447,
            1.6032606363296509
          ]
        }
      ],
      "peak_allocated_delta_bytes": 193410048,
      "peak_reserved_delta_bytes": 272629760,
      "peak_allocated_bytes": 294638080,
      "peak_reserved_bytes": 375390208,
      "baseline_allocated_bytes": 101228032,
      "baseline_reserved_bytes": 102760448,
      "parameters": 134144,
      "parameter_bytes": 536576,
      "dynamic_control_values_per_token": 2048,
      "materialized_state_sized_transition": false
    },
    {
      "shape": {
        "batch": 1,
        "length": 8192,
        "width": 1024
      },
      "architecture": "samu",
      "backend": "fused_output_fused_write_shared_sfu_chunk32",
      "controller": "fp32_fused_coords",
      "forward": {
        "samples_ms": [
          0.7529482245445251,
          0.745866447687149,
          0.7543750107288361,
          0.7424280047416687,
          0.7356069982051849,
          0.7375739514827728,
          0.7360577583312988,
          0.7340386509895325,
          0.7426105439662933,
          0.7162131369113922
        ],
        "median_ms": 0.7400009781122208,
        "mean_ms": 0.7397718727588654,
        "std_ms": 0.01031009393939792,
        "minimum_ms": 0.7162131369113922,
        "maximum_ms": 0.7543750107288361
      },
      "backward": {
        "samples_ms": [
          1.0211244225502014,
          1.0221339762210846,
          1.0201893746852875,
          1.0088421404361725,
          1.0022670030593872,
          0.9894147515296936,
          1.024387776851654,
          0.9711198508739471,
          1.0090135037899017,
          1.0117962956428528
        ],
        "median_ms": 1.0104048997163773,
        "mean_ms": 1.0080289095640182,
        "std_ms": 0.015949055909557038,
        "minimum_ms": 0.9711198508739471,
        "maximum_ms": 1.024387776851654
      },
      "forward_backward": {
        "samples_ms": [
          1.7963461577892303,
          1.7928481101989746,
          1.8761418759822845,
          1.7926841974258423,
          1.7863921821117401,
          1.843772828578949,
          1.7978698015213013,
          1.8157362937927246,
          1.8483996391296387,
          1.8049851059913635
        ],
        "median_ms": 1.8014274537563324,
        "mean_ms": 1.815517619252205,
        "std_ms": 0.02868522185806646,
        "minimum_ms": 1.7863921821117401,
        "maximum_ms": 1.8761418759822845
      },
      "rounds": [
        {
          "round": 0,
          "order_index": 1,
          "forward_ms": [
            0.7529482245445251,
            0.745866447687149,
            0.7543750107288361,
            0.7424280047416687,
            0.7356069982051849
          ],
          "backward_ms": [
            1.0211244225502014,
            1.0221339762210846,
            1.0201893746852875,
            1.0088421404361725,
            1.0022670030593872
          ],
          "forward_backward_ms": [
            1.7963461577892303,
            1.7928481101989746,
            1.8761418759822845,
            1.7926841974258423,
            1.7863921821117401
          ]
        },
        {
          "round": 1,
          "order_index": 0,
          "forward_ms": [
            0.7375739514827728,
            0.7360577583312988,
            0.7340386509895325,
            0.7426105439662933,
            0.7162131369113922
          ],
          "backward_ms": [
            0.9894147515296936,
            1.024387776851654,
            0.9711198508739471,
            1.0090135037899017,
            1.0117962956428528
          ],
          "forward_backward_ms": [
            1.843772828578949,
            1.7978698015213013,
            1.8157362937927246,
            1.8483996391296387,
            1.8049851059913635
          ]
        }
      ],
      "peak_allocated_delta_bytes": 151240192,
      "peak_reserved_delta_bytes": 155189248,
      "peak_allocated_bytes": 252990976,
      "peak_reserved_bytes": 257949696,
      "baseline_allocated_bytes": 101750784,
      "baseline_reserved_bytes": 102760448,
      "parameters": 3076,
      "parameter_bytes": 12304,
      "dynamic_control_values_per_token": 2,
      "materialized_state_sized_transition": false
    }
  ]
}
