{
  "run_id": "p4v2_abl_L4_seed03_20261030095138",
  "name": "p4v2_abl_L4",
  "seed": 3,
  "code_commit": "unknown",
  "config_sha256": "4f419f90c2a80c304c5e4442f65d32bc4f2124a82a7d06a1aac4bbf61f341b3d",
  "started_at": "2026-10-30T09:51:38-0400",
  "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_abl_L4.yaml --seed 3 --device mps",
  "state": "completed",
  "wall_seconds": 2156.44,
  "device": "mps",
  "backend": "mps",
  "precision": "fp32",
  "n_params": 4937413,
  "n_trainable": 137025,
  "checkpoint_parent": "experiments/runs/p4_tiny_backbone_v2_seed01_20261029233624/checkpoint",
  "mode": "prisme",
  "freeze_backbone": true,
  "readout_scale": 38.98434829711914,
  "finished_at": "2026-10-30T10:27:35-0400",
  "train_tokens": 499017,
  "peak_memory_bytes": 16439672832,
  "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": 0.398,
        "n": 1000,
        "by_family": {
          "fsm": 0.525,
          "bind": 0.27906976744186046,
          "arith": 0.10714285714285714,
          "excl": 0.2777777777777778,
          "sched": 0.812206572769953
        },
        "edits_per_token": 1.0
      },
      "2": {
        "accuracy": 0.399,
        "n": 1000,
        "by_family": {
          "fsm": 0.55625,
          "bind": 0.26976744186046514,
          "arith": 0.10714285714285714,
          "excl": 0.27314814814814814,
          "sched": 0.8075117370892019
        },
        "edits_per_token": 2.0
      },
      "8": {
        "accuracy": 0.406,
        "n": 1000,
        "by_family": {
          "fsm": 0.5375,
          "bind": 0.3023255813953488,
          "arith": 0.1377551020408163,
          "excl": 0.2638888888888889,
          "sched": 0.8028169014084507
        },
        "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": 0.281,
        "n": 1000,
        "by_family": {
          "fsm": 0.4427083333333333,
          "arith": 0.04950495049504951,
          "bind": 0.12871287128712872,
          "sched": 0.5495049504950495,
          "excl": 0.24257425742574257
        },
        "edits_per_token": 1.0
      },
      "2": {
        "accuracy": 0.272,
        "n": 1000,
        "by_family": {
          "fsm": 0.421875,
          "arith": 0.039603960396039604,
          "bind": 0.1188118811881188,
          "sched": 0.5594059405940595,
          "excl": 0.22772277227722773
        },
        "edits_per_token": 2.0
      },
      "8": {
        "accuracy": 0.274,
        "n": 1000,
        "by_family": {
          "fsm": 0.4010416666666667,
          "arith": 0.039603960396039604,
          "bind": 0.12376237623762376,
          "sched": 0.5841584158415841,
          "excl": 0.22772277227722773
        },
        "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": 0.297,
        "n": 1000,
        "by_family": {
          "arith": 0.025,
          "bind": 0.01,
          "fsm": 0.4,
          "excl": 0.75,
          "sched": 0.3
        },
        "edits_per_token": 1.0
      },
      "2": {
        "accuracy": 0.295,
        "n": 1000,
        "by_family": {
          "arith": 0.03,
          "bind": 0.01,
          "fsm": 0.385,
          "excl": 0.75,
          "sched": 0.3
        },
        "edits_per_token": 2.0
      },
      "8": {
        "accuracy": 0.3,
        "n": 1000,
        "by_family": {
          "arith": 0.03,
          "bind": 0.015,
          "fsm": 0.42,
          "excl": 0.725,
          "sched": 0.31
        },
        "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": 0.03518518518518519,
        "n": 540,
        "by_family": {
          "compose": 0.03518518518518519
        },
        "edits_per_token": 1.0
      },
      "2": {
        "accuracy": 0.03518518518518519,
        "n": 540,
        "by_family": {
          "compose": 0.03518518518518519
        },
        "edits_per_token": 2.0
      },
      "8": {
        "accuracy": 0.024074074074074074,
        "n": 540,
        "by_family": {
          "compose": 0.024074074074074074
        },
        "edits_per_token": 8.0
      }
    }
  },
  "metrics_file": "experiments/runs/p4v2_abl_L4_seed03_20261030095138/metrics.jsonl"
}