loganrobbins/parallel-decoder-transformer
1
1{2 "agreement_threshold": 0.15,3 "best_eval_loss": 21.752999266554905,4 "eval_history_length": 4,5 "eval_metrics": {6 "avg_margin_violation": null,7 "contradiction_rate": null,8 "coverage_f1": 0.09190465965213657,9 "coverage_precision": 0.7157190635451505,10 "coverage_recall": 0.04910509407985315,11 "coverage_source": "logits",12 "coverage_support": 117723.0,13 "eval_loss": 70.75614915129924,14 "nli_pair_count": 0,15 "redundancy_index": 0.00012577735405615995,16 "redundancy_pair_count": 8586617 },18 "generated_at": "2025-12-07T15:14:33.566756+00:00",19 "global_step": 50000,20 "stage": 3,21 "train_history_length": 1100,22 "train_metrics": {23 "coverage_f1": {24 "count": 1000,25 "last": 0.0,26 "max": 0.5,27 "mean": 0.0034451770451770455,28 "min": 0.029 },30 "coverage_precision": {31 "count": 1000,32 "last": 0.0,33 "max": 1.0,34 "mean": 0.0029482174688057043,35 "min": 0.036 },37 "coverage_recall": {38 "count": 1000,39 "last": 0.0,40 "max": 1.0,41 "mean": 0.010866666666666667,42 "min": 0.043 },44 "kd_ce_ratio": {45 "count": 1100,46 "last": 6.771889053472952e-07,47 "max": 1.993265237894144e-06,48 "mean": 2.7301640674022355e-07,49 "min": -2.044742390260858e-0650 },51 "loss": {52 "count": 1100,53 "last": 432.0,54 "max": 604.0336303710938,55 "mean": 405.4652520197088,56 "min": 216.201828002929757 },58 "repair_error_rate": {59 "count": 3,60 "last": 0.0,61 "max": 0.0,62 "mean": 0.0,63 "min": 0.064 },65 "repair_margin": {66 "count": 3,67 "last": 224.0,68 "max": 294.0,69 "mean": 261.3333333333333,70 "min": 224.071 },72 "rollback_kl": {73 "count": 3,74 "last": 0.0,75 "max": 0.0,76 "mean": 0.0,77 "min": 0.078 },79 "stability_error_rate": {80 "count": 3,81 "last": 0.0,82 "max": 0.0,83 "mean": 0.0,84 "min": 0.085 },86 "stability_kl": {87 "count": 3,88 "last": 0.0,89 "max": 2.837623469531536e-09,90 "mean": -1.8531864043325186e-07,91 "min": -5.587935447692871e-0792 },93 "stability_margin": {94 "count": 3,95 "last": 235.0,96 "max": 258.0,97 "mean": 246.33333333333334,98 "min": 235.099 },100 "stage": {101 "last": 3.0102 },103 "usage_loss": {104 "count": 1100,105 "last": 0.0,106 "max": 0.0,107 "mean": 0.0,108 "min": 0.0109 }110 }111}