[
  {
    "phase": 62,
    "mode": "original",
    "label": "op Some(21) LayoutRearrangement: TensorShape([2, 577, 4096]) F8F143 { scale_exponent: -4 } Amp(Left) -> TensorShape([2, 577, 4096]) Amp(Left)",
    "transfers": 147712,
    "receive_fragments": 2363392,
    "maximum_endpoint_fragments": 2020,
    "transmitted_bytes": 4726784,
    "received_bytes": 75628544,
    "maximum_outgoing_bytes": 8192,
    "maximum_incoming_bytes": 56448,
    "source_staging_max_bytes": 0,
    "destination_staging_max_bytes": 0,
    "combined_staging_max_bytes": 0,
    "source_copy_max_bytes": 0,
    "destination_copy_max_bytes": 0,
    "local_copy_runs": 0,
    "affine_copy_loops": 0,
    "copy_cycles_floor": 0,
    "replay_address_ranges_valid": true,
    "scheduled_row_max_bytes": 16776,
    "scheduled_horizon_cycles": 15937,
    "read_write_overlap": false
  },
  {
    "phase": 62,
    "mode": "source",
    "label": "op Some(21) LayoutRearrangement: TensorShape([2, 577, 4096]) F8F143 { scale_exponent: -4 } Amp(Left) -> TensorShape([2, 577, 4096]) Amp(Left)",
    "transfers": 147712,
    "receive_fragments": 2363392,
    "maximum_endpoint_fragments": 2020,
    "transmitted_bytes": 4726784,
    "received_bytes": 75628544,
    "maximum_outgoing_bytes": 8192,
    "maximum_incoming_bytes": 56448,
    "source_staging_max_bytes": 8192,
    "destination_staging_max_bytes": 0,
    "combined_staging_max_bytes": 8192,
    "source_copy_max_bytes": 8192,
    "destination_copy_max_bytes": 0,
    "local_copy_runs": 577,
    "affine_copy_loops": 577,
    "copy_cycles_floor": 1024,
    "replay_address_ranges_valid": true,
    "scheduled_row_max_bytes": 16776,
    "scheduled_horizon_cycles": 15937,
    "read_write_overlap": false
  },
  {
    "phase": 62,
    "mode": "destination",
    "label": "op Some(21) LayoutRearrangement: TensorShape([2, 577, 4096]) F8F143 { scale_exponent: -4 } Amp(Left) -> TensorShape([2, 577, 4096]) Amp(Left)",
    "transfers": 8655,
    "receive_fragments": 138480,
    "maximum_endpoint_fragments": 113,
    "transmitted_bytes": 4726784,
    "received_bytes": 75628544,
    "maximum_outgoing_bytes": 8192,
    "maximum_incoming_bytes": 56448,
    "source_staging_max_bytes": 0,
    "destination_staging_max_bytes": 56448,
    "combined_staging_max_bytes": 56448,
    "source_copy_max_bytes": 0,
    "destination_copy_max_bytes": 56448,
    "local_copy_runs": 2363392,
    "affine_copy_loops": 138480,
    "copy_cycles_floor": 7056,
    "replay_address_ranges_valid": true,
    "scheduled_row_max_bytes": 976,
    "scheduled_horizon_cycles": 14376,
    "read_write_overlap": false
  },
  {
    "phase": 62,
    "mode": "both",
    "label": "op Some(21) LayoutRearrangement: TensorShape([2, 577, 4096]) F8F143 { scale_exponent: -4 } Amp(Left) -> TensorShape([2, 577, 4096]) Amp(Left)",
    "transfers": 8655,
    "receive_fragments": 138480,
    "maximum_endpoint_fragments": 113,
    "transmitted_bytes": 4726784,
    "received_bytes": 75628544,
    "maximum_outgoing_bytes": 8192,
    "maximum_incoming_bytes": 56448,
    "source_staging_max_bytes": 8192,
    "destination_staging_max_bytes": 56448,
    "combined_staging_max_bytes": 64640,
    "source_copy_max_bytes": 8192,
    "destination_copy_max_bytes": 56448,
    "local_copy_runs": 2363969,
    "affine_copy_loops": 139057,
    "copy_cycles_floor": 8080,
    "replay_address_ranges_valid": true,
    "scheduled_row_max_bytes": 976,
    "scheduled_horizon_cycles": 14376,
    "read_write_overlap": false
  },
  {
    "phase": 58,
    "mode": "original",
    "label": "op Some(18) OperatorInputs: TensorShape([2, 577, 1024]) F8F143 { scale_exponent: -4 } Amp(Left) -> TensorShape([2, 577, 1024]) Amp(Left)",
    "transfers": 45375,
    "receive_fragments": 1229447,
    "maximum_endpoint_fragments": 1000,
    "transmitted_bytes": 9565696,
    "received_bytes": 85128704,
    "maximum_outgoing_bytes": 37888,
    "maximum_incoming_bytes": 60416,
    "source_staging_max_bytes": 0,
    "destination_staging_max_bytes": 0,
    "combined_staging_max_bytes": 0,
    "source_copy_max_bytes": 0,
    "destination_copy_max_bytes": 0,
    "local_copy_runs": 0,
    "affine_copy_loops": 0,
    "copy_cycles_floor": 0,
    "replay_address_ranges_valid": true,
    "scheduled_row_max_bytes": 8408,
    "scheduled_horizon_cycles": 27446,
    "read_write_overlap": true
  },
  {
    "phase": 58,
    "mode": "source",
    "label": "op Some(18) OperatorInputs: TensorShape([2, 577, 1024]) F8F143 { scale_exponent: -4 } Amp(Left) -> TensorShape([2, 577, 1024]) Amp(Left)",
    "transfers": 41538,
    "receive_fragments": 1225610,
    "maximum_endpoint_fragments": 973,
    "transmitted_bytes": 9565696,
    "received_bytes": 85128704,
    "maximum_outgoing_bytes": 37888,
    "maximum_incoming_bytes": 60416,
    "source_staging_max_bytes": 37888,
    "destination_staging_max_bytes": 0,
    "combined_staging_max_bytes": 37888,
    "source_copy_max_bytes": 37888,
    "destination_copy_max_bytes": 0,
    "local_copy_runs": 8656,
    "affine_copy_loops": 4801,
    "copy_cycles_floor": 4736,
    "replay_address_ranges_valid": true,
    "read_write_overlap": true,
    "invalid_counterfactual_reason": "Receive/forward or alias dependencies were not preserved"
  },
  {
    "phase": 58,
    "mode": "destination",
    "label": "op Some(18) OperatorInputs: TensorShape([2, 577, 1024]) F8F143 { scale_exponent: -4 } Amp(Left) -> TensorShape([2, 577, 1024]) Amp(Left)",
    "transfers": 5018,
    "receive_fragments": 81250,
    "maximum_endpoint_fragments": 88,
    "transmitted_bytes": 9565696,
    "received_bytes": 85128704,
    "maximum_outgoing_bytes": 37888,
    "maximum_incoming_bytes": 60416,
    "source_staging_max_bytes": 0,
    "destination_staging_max_bytes": 60416,
    "combined_staging_max_bytes": 60416,
    "source_copy_max_bytes": 0,
    "destination_copy_max_bytes": 60416,
    "local_copy_runs": 1226927,
    "affine_copy_loops": 80530,
    "copy_cycles_floor": 7552,
    "replay_address_ranges_valid": true,
    "read_write_overlap": true,
    "invalid_counterfactual_reason": "Receive/forward or alias dependencies were not preserved"
  },
  {
    "phase": 58,
    "mode": "both",
    "label": "op Some(18) OperatorInputs: TensorShape([2, 577, 1024]) F8F143 { scale_exponent: -4 } Amp(Left) -> TensorShape([2, 577, 1024]) Amp(Left)",
    "transfers": 4388,
    "receive_fragments": 80620,
    "maximum_endpoint_fragments": 81,
    "transmitted_bytes": 9565696,
    "received_bytes": 85128704,
    "maximum_outgoing_bytes": 37888,
    "maximum_incoming_bytes": 60416,
    "source_staging_max_bytes": 37888,
    "destination_staging_max_bytes": 60416,
    "combined_staging_max_bytes": 98304,
    "source_copy_max_bytes": 37888,
    "destination_copy_max_bytes": 60416,
    "local_copy_runs": 1230143,
    "affine_copy_loops": 83102,
    "copy_cycles_floor": 12288,
    "replay_address_ranges_valid": true,
    "read_write_overlap": true,
    "invalid_counterfactual_reason": "Receive/forward or alias dependencies were not preserved"
  },
  {
    "phase": 84,
    "mode": "original",
    "label": "op Some(35) OperatorInputs: TensorShape([2, 577, 1024]) F16 RowMajor -> TensorShape([16, 577, 128]) RowMajor",
    "transfers": 82504,
    "receive_fragments": 83079,
    "maximum_endpoint_fragments": 749,
    "transmitted_bytes": 4579072,
    "received_bytes": 4726272,
    "maximum_outgoing_bytes": 12000,
    "maximum_incoming_bytes": 22528,
    "source_staging_max_bytes": 0,
    "destination_staging_max_bytes": 0,
    "combined_staging_max_bytes": 0,
    "source_copy_max_bytes": 0,
    "destination_copy_max_bytes": 0,
    "local_copy_runs": 0,
    "affine_copy_loops": 0,
    "copy_cycles_floor": 0,
    "replay_address_ranges_valid": true,
    "scheduled_row_max_bytes": 6008,
    "scheduled_horizon_cycles": 9256,
    "read_write_overlap": false
  },
  {
    "phase": 84,
    "mode": "source",
    "label": "op Some(35) OperatorInputs: TensorShape([2, 577, 1024]) F16 RowMajor -> TensorShape([16, 577, 128]) RowMajor",
    "transfers": 82504,
    "receive_fragments": 83079,
    "maximum_endpoint_fragments": 749,
    "transmitted_bytes": 4579072,
    "received_bytes": 4726272,
    "maximum_outgoing_bytes": 12000,
    "maximum_incoming_bytes": 22528,
    "source_staging_max_bytes": 12000,
    "destination_staging_max_bytes": 0,
    "combined_staging_max_bytes": 12000,
    "source_copy_max_bytes": 12000,
    "destination_copy_max_bytes": 0,
    "local_copy_runs": 76505,
    "affine_copy_loops": 9172,
    "copy_cycles_floor": 1500,
    "replay_address_ranges_valid": true,
    "scheduled_row_max_bytes": 6048,
    "scheduled_horizon_cycles": 6228,
    "read_write_overlap": false
  },
  {
    "phase": 84,
    "mode": "destination",
    "label": "op Some(35) OperatorInputs: TensorShape([2, 577, 1024]) F16 RowMajor -> TensorShape([16, 577, 128]) RowMajor",
    "transfers": 82399,
    "receive_fragments": 82974,
    "maximum_endpoint_fragments": 749,
    "transmitted_bytes": 4579072,
    "received_bytes": 4726272,
    "maximum_outgoing_bytes": 12000,
    "maximum_incoming_bytes": 22528,
    "source_staging_max_bytes": 0,
    "destination_staging_max_bytes": 22528,
    "combined_staging_max_bytes": 22528,
    "source_copy_max_bytes": 0,
    "destination_copy_max_bytes": 22528,
    "local_copy_runs": 5512,
    "affine_copy_loops": 5364,
    "copy_cycles_floor": 2816,
    "replay_address_ranges_valid": true,
    "scheduled_row_max_bytes": 5936,
    "scheduled_horizon_cycles": 6220,
    "read_write_overlap": false
  },
  {
    "phase": 84,
    "mode": "both",
    "label": "op Some(35) OperatorInputs: TensorShape([2, 577, 1024]) F16 RowMajor -> TensorShape([16, 577, 128]) RowMajor",
    "transfers": 80479,
    "receive_fragments": 81054,
    "maximum_endpoint_fragments": 719,
    "transmitted_bytes": 4579072,
    "received_bytes": 4726272,
    "maximum_outgoing_bytes": 12000,
    "maximum_incoming_bytes": 22528,
    "source_staging_max_bytes": 12000,
    "destination_staging_max_bytes": 22528,
    "combined_staging_max_bytes": 34272,
    "source_copy_max_bytes": 12000,
    "destination_copy_max_bytes": 22528,
    "local_copy_runs": 82017,
    "affine_copy_loops": 14536,
    "copy_cycles_floor": 4316,
    "replay_address_ranges_valid": true,
    "scheduled_row_max_bytes": 5760,
    "scheduled_horizon_cycles": 6218,
    "read_write_overlap": false
  }
]