1{2 "run_id": "p4v2_abl_M8_seed01_20261030095137",3 "name": "p4v2_abl_M8",4 "seed": 1,5 "code_commit": "unknown",6 "config_sha256": "5823212ed888f7d9f5d85ec0de87aed177dc1c8ce4b90dc8c5eef760275b84c1",7 "started_at": "2026-10-30T09:51:37-0400",8 "host": "M5M36",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/p4v2_abl_M8.yaml --seed 1 --device mps",12 "state": "completed",13 "wall_seconds": 1989.04,14 "device": "mps",15 "backend": "mps",16 "precision": "fp32",17 "n_params": 4936901,18 "n_trainable": 136513,19 "checkpoint_parent": "experiments/runs/p4_tiny_backbone_v2_seed01_20261029233624/checkpoint",20 "mode": "prisme",21 "freeze_backbone": true,22 "readout_scale": 38.47460174560547,23 "finished_at": "2026-10-30T10:24:46-0400",24 "train_tokens": 496298,25 "peak_memory_bytes": 7056687104,26 "final_exact_match": {27 "dev_iid": {28 "0": {29 "accuracy": 0.377,30 "n": 1000,31 "by_family": {32 "fsm": 0.48125,33 "bind": 0.24651162790697675,34 "arith": 0.09693877551020408,35 "excl": 0.2777777777777778,36 "sched": 0.788732394366197137 },38 "edits_per_token": 0.039 },40 "1": {41 "accuracy": 0.397,42 "n": 1000,43 "by_family": {44 "fsm": 0.5375,45 "bind": 0.2744186046511628,46 "arith": 0.11224489795918367,47 "excl": 0.2824074074074074,48 "sched": 0.793427230046948449 },50 "edits_per_token": 1.051 },52 "2": {53 "accuracy": 0.407,54 "n": 1000,55 "by_family": {56 "fsm": 0.56875,57 "bind": 0.2837209302325581,58 "arith": 0.12244897959183673,59 "excl": 0.2824074074074074,60 "sched": 0.798122065727699561 },62 "edits_per_token": 2.063 },64 "8": {65 "accuracy": 0.38,66 "n": 1000,67 "by_family": {68 "fsm": 0.54375,69 "bind": 0.2,70 "arith": 0.12244897959183673,71 "excl": 0.25925925925925924,72 "sched": 0.798122065727699573 },74 "edits_per_token": 8.075 }76 },77 "dev_template": {78 "0": {79 "accuracy": 0.281,80 "n": 1000,81 "by_family": {82 "fsm": 0.453125,83 "arith": 0.04950495049504951,84 "bind": 0.12871287128712872,85 "sched": 0.5445544554455446,86 "excl": 0.237623762376237687 },88 "edits_per_token": 0.089 },90 "1": {91 "accuracy": 0.28,92 "n": 1000,93 "by_family": {94 "fsm": 0.453125,95 "arith": 0.04455445544554455,96 "bind": 0.12376237623762376,97 "sched": 0.5445544554455446,98 "excl": 0.2425742574257425799 },100 "edits_per_token": 1.0101 },102 "2": {103 "accuracy": 0.28,104 "n": 1000,105 "by_family": {106 "fsm": 0.4479166666666667,107 "arith": 0.04950495049504951,108 "bind": 0.12376237623762376,109 "sched": 0.5544554455445545,110 "excl": 0.23267326732673269111 },112 "edits_per_token": 2.0113 },114 "8": {115 "accuracy": 0.274,116 "n": 1000,117 "by_family": {118 "fsm": 0.421875,119 "arith": 0.019801980198019802,120 "bind": 0.1188118811881188,121 "sched": 0.5742574257425742,122 "excl": 0.24257425742574257123 },124 "edits_per_token": 8.0125 }126 },127 "dev_depth": {128 "0": {129 "accuracy": 0.29,130 "n": 1000,131 "by_family": {132 "arith": 0.03,133 "bind": 0.02,134 "fsm": 0.37,135 "excl": 0.75,136 "sched": 0.28137 },138 "edits_per_token": 0.0139 },140 "1": {141 "accuracy": 0.293,142 "n": 1000,143 "by_family": {144 "arith": 0.025,145 "bind": 0.015,146 "fsm": 0.39,147 "excl": 0.75,148 "sched": 0.285149 },150 "edits_per_token": 1.0151 },152 "2": {153 "accuracy": 0.294,154 "n": 1000,155 "by_family": {156 "arith": 0.03,157 "bind": 0.015,158 "fsm": 0.385,159 "excl": 0.75,160 "sched": 0.29161 },162 "edits_per_token": 2.0163 },164 "8": {165 "accuracy": 0.297,166 "n": 1000,167 "by_family": {168 "arith": 0.015,169 "bind": 0.02,170 "fsm": 0.41,171 "excl": 0.75,172 "sched": 0.29173 },174 "edits_per_token": 8.0175 }176 },177 "dev_compose": {178 "0": {179 "accuracy": 0.03518518518518519,180 "n": 540,181 "by_family": {182 "compose": 0.03518518518518519183 },184 "edits_per_token": 0.0185 },186 "1": {187 "accuracy": 0.03148148148148148,188 "n": 540,189 "by_family": {190 "compose": 0.03148148148148148191 },192 "edits_per_token": 1.0193 },194 "2": {195 "accuracy": 0.027777777777777776,196 "n": 540,197 "by_family": {198 "compose": 0.027777777777777776199 },200 "edits_per_token": 2.0201 },202 "8": {203 "accuracy": 0.03148148148148148,204 "n": 540,205 "by_family": {206 "compose": 0.03148148148148148207 },208 "edits_per_token": 8.0209 }210 }211 },212 "metrics_file": "experiments/runs/p4v2_abl_M8_seed01_20261030095137/metrics.jsonl"213}