{
  "run_id": "p4v2_prisme_seed04_20261030001749",
  "name": "p4v2_prisme",
  "seed": 4,
  "code_commit": "unknown",
  "config_sha256": "4588d88acc48828c1409cf3d75e6dd080e44f944e5dffc04fa5ebf6b869e1fc5",
  "started_at": "2026-10-30T00:17:49-0400",
  "host": "M5M36",
  "python": "3.12.14",
  "torch": "2.15.0",
  "command": "/Users/simon-pierreboucher/wp31/prisme-slm/training/train_editor.py --config configs/experiments/p4v2_prisme.yaml --seed 4 --device mps",
  "state": "completed",
  "wall_seconds": 1504.41,
  "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": 39.0492057800293,
  "finished_at": "2026-10-30T00:42:53-0400",
  "train_tokens": 497821,
  "peak_memory_bytes": 9288859648,
  "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": 0.392,
        "n": 1000,
        "by_family": {
          "fsm": 0.525,
          "bind": 0.27906976744186046,
          "arith": 0.10204081632653061,
          "excl": 0.27314814814814814,
          "sched": 0.7934272300469484
        },
        "edits_per_token": 1.0
      },
      "2": {
        "accuracy": 0.395,
        "n": 1000,
        "by_family": {
          "fsm": 0.54375,
          "bind": 0.26976744186046514,
          "arith": 0.11224489795918367,
          "excl": 0.26851851851851855,
          "sched": 0.7981220657276995
        },
        "edits_per_token": 2.0
      },
      "8": {
        "accuracy": 0.39,
        "n": 1000,
        "by_family": {
          "fsm": 0.55,
          "bind": 0.24651162790697675,
          "arith": 0.10714285714285714,
          "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": 0.284,
        "n": 1000,
        "by_family": {
          "fsm": 0.4479166666666667,
          "arith": 0.04950495049504951,
          "bind": 0.13861386138613863,
          "sched": 0.5445544554455446,
          "excl": 0.24752475247524752
        },
        "edits_per_token": 1.0
      },
      "2": {
        "accuracy": 0.279,
        "n": 1000,
        "by_family": {
          "fsm": 0.4479166666666667,
          "arith": 0.039603960396039604,
          "bind": 0.1188118811881188,
          "sched": 0.5544554455445545,
          "excl": 0.24257425742574257
        },
        "edits_per_token": 2.0
      },
      "8": {
        "accuracy": 0.271,
        "n": 1000,
        "by_family": {
          "fsm": 0.40625,
          "arith": 0.019801980198019802,
          "bind": 0.1188118811881188,
          "sched": 0.5742574257425742,
          "excl": 0.24257425742574257
        },
        "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": 0.297,
        "n": 1000,
        "by_family": {
          "arith": 0.03,
          "bind": 0.025,
          "fsm": 0.385,
          "excl": 0.75,
          "sched": 0.295
        },
        "edits_per_token": 2.0
      },
      "8": {
        "accuracy": 0.296,
        "n": 1000,
        "by_family": {
          "arith": 0.035,
          "bind": 0.02,
          "fsm": 0.39,
          "excl": 0.74,
          "sched": 0.295
        },
        "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": 0.03333333333333333,
        "n": 540,
        "by_family": {
          "compose": 0.03333333333333333
        },
        "edits_per_token": 8.0
      }
    }
  },
  "metrics_file": "experiments/runs/p4v2_prisme_seed04_20261030001749/metrics.jsonl"
}