{
  "run_id": "p4v2_dense_seed05_20261030004544",
  "name": "p4v2_dense",
  "seed": 5,
  "code_commit": "unknown",
  "config_sha256": "3380d74a688a64f553180135f59017be80822d18d8f866636228cab887fe72f6",
  "started_at": "2026-10-30T00:45:44-0400",
  "host": "simonpirresMBP2",
  "python": "3.12.13",
  "torch": "2.15.0",
  "command": "/Users/simon-pierreboucher/wp31/prisme-slm/training/train_editor.py --config configs/experiments/p4v2_dense.yaml --seed 5 --device mps",
  "state": "completed",
  "wall_seconds": 2620.45,
  "device": "mps",
  "backend": "mps",
  "precision": "fp32",
  "n_params": 4937413,
  "n_trainable": 137025,
  "checkpoint_parent": "experiments/runs/p4_tiny_backbone_v2_seed01_20261029233624/checkpoint",
  "mode": "dense",
  "freeze_backbone": true,
  "readout_scale": 39.13882827758789,
  "finished_at": "2026-10-30T01:29:25-0400",
  "train_tokens": 497078,
  "peak_memory_bytes": 28204498944,
  "final_exact_match": {
    "dev_iid": {
      "0": {
        "accuracy": 0.377,
        "n": 1000,
        "by_family": {
          "fsm": 0.48125,
          "bind": 0.24651162790697675,
          "arith": 0.09693877551020408,
          "excl": 0.2777777777777778,
          "sched": 0.7887323943661971
        },
        "edits_per_token": 0.0
      },
      "1": {
        "accuracy": 0.395,
        "n": 1000,
        "by_family": {
          "fsm": 0.53125,
          "bind": 0.26976744186046514,
          "arith": 0.09693877551020408,
          "excl": 0.2777777777777778,
          "sched": 0.812206572769953
        },
        "edits_per_token": 1.0
      },
      "2": {
        "accuracy": 0.392,
        "n": 1000,
        "by_family": {
          "fsm": 0.5375,
          "bind": 0.2558139534883721,
          "arith": 0.08673469387755102,
          "excl": 0.2824074074074074,
          "sched": 0.812206572769953
        },
        "edits_per_token": 2.0
      },
      "8": {
        "accuracy": 0.398,
        "n": 1000,
        "by_family": {
          "fsm": 0.58125,
          "bind": 0.24186046511627907,
          "arith": 0.11734693877551021,
          "excl": 0.26851851851851855,
          "sched": 0.8075117370892019
        },
        "edits_per_token": 8.0
      }
    },
    "dev_template": {
      "0": {
        "accuracy": 0.281,
        "n": 1000,
        "by_family": {
          "fsm": 0.453125,
          "arith": 0.04950495049504951,
          "bind": 0.12871287128712872,
          "sched": 0.5445544554455446,
          "excl": 0.2376237623762376
        },
        "edits_per_token": 0.0
      },
      "1": {
        "accuracy": 0.282,
        "n": 1000,
        "by_family": {
          "fsm": 0.453125,
          "arith": 0.054455445544554455,
          "bind": 0.12871287128712872,
          "sched": 0.5396039603960396,
          "excl": 0.24257425742574257
        },
        "edits_per_token": 1.0
      },
      "2": {
        "accuracy": 0.28,
        "n": 1000,
        "by_family": {
          "fsm": 0.4479166666666667,
          "arith": 0.04950495049504951,
          "bind": 0.11386138613861387,
          "sched": 0.5594059405940595,
          "excl": 0.2376237623762376
        },
        "edits_per_token": 2.0
      },
      "8": {
        "accuracy": 0.279,
        "n": 1000,
        "by_family": {
          "fsm": 0.4270833333333333,
          "arith": 0.0297029702970297,
          "bind": 0.11386138613861387,
          "sched": 0.594059405940594,
          "excl": 0.2376237623762376
        },
        "edits_per_token": 8.0
      }
    },
    "dev_depth": {
      "0": {
        "accuracy": 0.29,
        "n": 1000,
        "by_family": {
          "arith": 0.03,
          "bind": 0.02,
          "fsm": 0.37,
          "excl": 0.75,
          "sched": 0.28
        },
        "edits_per_token": 0.0
      },
      "1": {
        "accuracy": 0.294,
        "n": 1000,
        "by_family": {
          "arith": 0.03,
          "bind": 0.015,
          "fsm": 0.385,
          "excl": 0.75,
          "sched": 0.29
        },
        "edits_per_token": 1.0
      },
      "2": {
        "accuracy": 0.294,
        "n": 1000,
        "by_family": {
          "arith": 0.03,
          "bind": 0.015,
          "fsm": 0.38,
          "excl": 0.755,
          "sched": 0.29
        },
        "edits_per_token": 2.0
      },
      "8": {
        "accuracy": 0.3,
        "n": 1000,
        "by_family": {
          "arith": 0.02,
          "bind": 0.02,
          "fsm": 0.39,
          "excl": 0.755,
          "sched": 0.315
        },
        "edits_per_token": 8.0
      }
    },
    "dev_compose": {
      "0": {
        "accuracy": 0.03518518518518519,
        "n": 540,
        "by_family": {
          "compose": 0.03518518518518519
        },
        "edits_per_token": 0.0
      },
      "1": {
        "accuracy": 0.03148148148148148,
        "n": 540,
        "by_family": {
          "compose": 0.03148148148148148
        },
        "edits_per_token": 1.0
      },
      "2": {
        "accuracy": 0.03148148148148148,
        "n": 540,
        "by_family": {
          "compose": 0.03148148148148148
        },
        "edits_per_token": 2.0
      },
      "8": {
        "accuracy": 0.03148148148148148,
        "n": 540,
        "by_family": {
          "compose": 0.03148148148148148
        },
        "edits_per_token": 8.0
      }
    }
  },
  "metrics_file": "experiments/runs/p4v2_dense_seed05_20261030004544/metrics.jsonl"
}