{
  "run_id": "p4v2_abl_L4_seed01_20261030095138",
  "name": "p4v2_abl_L4",
  "seed": 1,
  "code_commit": "unknown",
  "config_sha256": "4f419f90c2a80c304c5e4442f65d32bc4f2124a82a7d06a1aac4bbf61f341b3d",
  "started_at": "2026-10-30T09:51:38-0400",
  "host": "MacStudnpierre6",
  "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 1 --device mps",
  "state": "completed",
  "wall_seconds": 3212.56,
  "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": 38.47460174560547,
  "finished_at": "2026-10-30T10:45:10-0400",
  "train_tokens": 496298,
  "peak_memory_bytes": 15399682048,
  "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": 0.397,
        "n": 1000,
        "by_family": {
          "fsm": 0.5375,
          "bind": 0.26976744186046514,
          "arith": 0.10714285714285714,
          "excl": 0.28703703703703703,
          "sched": 0.7981220657276995
        },
        "edits_per_token": 1.0
      },
      "2": {
        "accuracy": 0.406,
        "n": 1000,
        "by_family": {
          "fsm": 0.575,
          "bind": 0.2558139534883721,
          "arith": 0.11224489795918367,
          "excl": 0.2962962962962963,
          "sched": 0.812206572769953
        },
        "edits_per_token": 2.0
      },
      "8": {
        "accuracy": 0.388,
        "n": 1000,
        "by_family": {
          "fsm": 0.53125,
          "bind": 0.25116279069767444,
          "arith": 0.11734693877551021,
          "excl": 0.26851851851851855,
          "sched": 0.7887323943661971
        },
        "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": 0.278,
        "n": 1000,
        "by_family": {
          "fsm": 0.4270833333333333,
          "arith": 0.039603960396039604,
          "bind": 0.13861386138613863,
          "sched": 0.5495049504950495,
          "excl": 0.24257425742574257
        },
        "edits_per_token": 1.0
      },
      "2": {
        "accuracy": 0.278,
        "n": 1000,
        "by_family": {
          "fsm": 0.4166666666666667,
          "arith": 0.034653465346534656,
          "bind": 0.13366336633663367,
          "sched": 0.5742574257425742,
          "excl": 0.2376237623762376
        },
        "edits_per_token": 2.0
      },
      "8": {
        "accuracy": 0.28,
        "n": 1000,
        "by_family": {
          "fsm": 0.4114583333333333,
          "arith": 0.019801980198019802,
          "bind": 0.12871287128712872,
          "sched": 0.599009900990099,
          "excl": 0.24752475247524752
        },
        "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": 0.292,
        "n": 1000,
        "by_family": {
          "arith": 0.025,
          "bind": 0.015,
          "fsm": 0.385,
          "excl": 0.75,
          "sched": 0.285
        },
        "edits_per_token": 1.0
      },
      "2": {
        "accuracy": 0.298,
        "n": 1000,
        "by_family": {
          "arith": 0.03,
          "bind": 0.02,
          "fsm": 0.395,
          "excl": 0.75,
          "sched": 0.295
        },
        "edits_per_token": 2.0
      },
      "8": {
        "accuracy": 0.302,
        "n": 1000,
        "by_family": {
          "arith": 0.035,
          "bind": 0.015,
          "fsm": 0.415,
          "excl": 0.745,
          "sched": 0.3
        },
        "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": 0.027777777777777776,
        "n": 540,
        "by_family": {
          "compose": 0.027777777777777776
        },
        "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_abl_L4_seed01_20261030095138/metrics.jsonl"
}