1{2 "run_id": "p4_editor_v1_Ustd02_seed01_20261029235050",3 "name": "p4_editor_v1_Ustd02",4 "seed": 1,5 "code_commit": "unknown",6 "config_sha256": "d677af0752c565d76a56b78c1fa0003de1fbd0347957f8df0a21dd0dae86f639",7 "started_at": "2026-10-29T23:50:50-0400",8 "host": "M4M64c",9 "python": "3.12.14",10 "torch": "2.15.0",11 "command": "/Users/simon-pierreboucher/wp31/prisme-slm/training/train_editor.py --config configs/experiments/p4_editor_v1_Ustd02.yaml --device mps",12 "state": "completed",13 "wall_seconds": 650.11,14 "device": "mps",15 "backend": "mps",16 "precision": "fp32",17 "n_params": 1765445,18 "n_trainable": 120577,19 "checkpoint_parent": "experiments/runs/p4_tiny_backbone_seed01_20261029232626/checkpoint",20 "mode": "prisme",21 "freeze_backbone": true,22 "readout_scale": 5.251033782958984,23 "finished_at": "2026-10-30T00:01:40-0400",24 "train_tokens": 331026,25 "peak_memory_bytes": 8829190144,26 "final_exact_match": {27 "dev_iid": {28 "0": {29 "accuracy": 0.4375,30 "n": 400,31 "by_family": {32 "fsm": 0.640625,33 "bind": 0.38372093023255816,34 "arith": 0.22784810126582278,35 "excl": 0.27906976744186046,36 "sched": 0.694117647058823537 },38 "edits_per_token": 0.039 },40 "2": {41 "accuracy": 0.4575,42 "n": 400,43 "by_family": {44 "fsm": 0.671875,45 "bind": 0.3953488372093023,46 "arith": 0.27848101265822783,47 "excl": 0.27906976744186046,48 "sched": 0.705882352941176549 },50 "edits_per_token": 2.051 },52 "8": {53 "accuracy": 0.4425,54 "n": 400,55 "by_family": {56 "fsm": 0.671875,57 "bind": 0.38372093023255816,58 "arith": 0.21518987341772153,59 "excl": 0.26744186046511625,60 "sched": 0.717647058823529461 },62 "edits_per_token": 8.063 }64 },65 "dev_template": {66 "0": {67 "accuracy": 0.2675,68 "n": 400,69 "by_family": {70 "fsm": 0.4025974025974026,71 "bind": 0.2222222222222222,72 "arith": 0.08641975308641975,73 "sched": 0.475,74 "excl": 0.1604938271604938275 },76 "edits_per_token": 0.077 },78 "2": {79 "accuracy": 0.26,80 "n": 400,81 "by_family": {82 "fsm": 0.4025974025974026,83 "bind": 0.2345679012345679,84 "arith": 0.07407407407407407,85 "sched": 0.45,86 "excl": 0.1481481481481481487 },88 "edits_per_token": 2.089 },90 "8": {91 "accuracy": 0.23,92 "n": 400,93 "by_family": {94 "fsm": 0.36363636363636365,95 "bind": 0.19753086419753085,96 "arith": 0.07407407407407407,97 "sched": 0.4125,98 "excl": 0.111111111111111199 },100 "edits_per_token": 8.0101 }102 },103 "dev_depth": {104 "0": {105 "accuracy": 0.3125,106 "n": 400,107 "by_family": {108 "arith": 0.0125,109 "bind": 0.0875,110 "fsm": 0.375,111 "excl": 0.825,112 "sched": 0.2625113 },114 "edits_per_token": 0.0115 },116 "2": {117 "accuracy": 0.3175,118 "n": 400,119 "by_family": {120 "arith": 0.0125,121 "bind": 0.0875,122 "fsm": 0.3875,123 "excl": 0.825,124 "sched": 0.275125 },126 "edits_per_token": 2.0127 },128 "8": {129 "accuracy": 0.3075,130 "n": 400,131 "by_family": {132 "arith": 0.0125,133 "bind": 0.0875,134 "fsm": 0.375,135 "excl": 0.8,136 "sched": 0.2625137 },138 "edits_per_token": 8.0139 }140 },141 "dev_compose": {142 "0": {143 "accuracy": 0.0225,144 "n": 400,145 "by_family": {146 "compose": 0.0225147 },148 "edits_per_token": 0.0149 },150 "2": {151 "accuracy": 0.025,152 "n": 400,153 "by_family": {154 "compose": 0.025155 },156 "edits_per_token": 2.0157 },158 "8": {159 "accuracy": 0.03,160 "n": 400,161 "by_family": {162 "compose": 0.03163 },164 "edits_per_token": 8.0165 }166 }167 },168 "metrics_file": "experiments/runs/p4_editor_v1_Ustd02_seed01_20261029235050/metrics.jsonl"169}