| { |
| "cycle": 3, |
| "timestamp": 1776959087.1466334, |
| "scores": { |
| "pre_training": 0.44171779141104295, |
| "post_training": 0.44171779141104295, |
| "held_out_eval": 0.4843205574912892, |
| "improvement": 0.0, |
| "improvement_ema": 0.0, |
| "best_score": 0.0, |
| "best_checkpoint_cycle": null |
| }, |
| "domain_scores": { |
| "pre": { |
| "code": 0.5396825396825397, |
| "math": 0.4027777777777778, |
| "logic": 0.32142857142857145 |
| }, |
| "post": {}, |
| "eval": { |
| "code": 0.4222222222222222, |
| "math": 0.4793388429752066, |
| "logic": 0.512396694214876 |
| } |
| }, |
| "subdomain_scores": { |
| "pre": { |
| "code/bit_manipulation": 0.6923076923076923, |
| "code/implementation": 0.4074074074074074, |
| "code/prediction": 0.3333333333333333, |
| "code/debugging": 0.6666666666666666, |
| "code/computing": 1.0, |
| "math/word_problem": 0.5555555555555556, |
| "math/symbolic": 0.14285714285714285, |
| "math/algebra": 0.0, |
| "math/number_theory": 0.7, |
| "math/arithmetic": 1.0, |
| "math/math": 1.0, |
| "math/combinatorics": 0.0, |
| "math/probability": 0.3333333333333333, |
| "math/calculus": 0.75, |
| "logic/fallacies": 0.0, |
| "logic/syllogism": 0.15384615384615385, |
| "logic/propositional": 0.375, |
| "logic/reasoning": 1.0, |
| "logic/modal": 0.0 |
| }, |
| "post": {}, |
| "eval": { |
| "code/computing": 1.0, |
| "code/implementation": 0.36585365853658536, |
| "math/symbolic": 0.23684210526315788, |
| "math/word_problem": 0.5813953488372093, |
| "math/math": 1.0, |
| "math/probability": 0.0, |
| "math/calculus": 0.0, |
| "math/combinatorics": 1.0, |
| "math/number_theory": 0.6923076923076923, |
| "math/percentage": 1.0, |
| "math/arithmetic": 1.0, |
| "logic/reasoning": 0.9411764705882353, |
| "logic/syllogism": 0.30952380952380953, |
| "logic/propositional": 1.0 |
| } |
| }, |
| "samples": { |
| "generated": 6, |
| "verified": 5, |
| "rejected": 1, |
| "pass_rate": 0.8333333333333334, |
| "diversity": { |
| "mean": 0.4479910306688751, |
| "variance": 0.02499077202728213, |
| "n_problems": 6, |
| "n_collapsed": 0, |
| "fraction_collapsed": 0.0, |
| "mode_collapse_alarm": false, |
| "embedding_backend": "hash-fallback" |
| } |
| }, |
| "training": { |
| "avg_loss": 0.8987893462181091, |
| "final_loss": 0.8295454382896423, |
| "steps": 2, |
| "learning_rate": 7.28e-06, |
| "lora_layers": 448 |
| }, |
| "calibration": { |
| "ece": null, |
| "brier": null, |
| "samples": 0 |
| }, |
| "timing": { |
| "rsi_step0_diagnose": 621.4804835319519, |
| "rsi_step1_propose": 425.96521759033203, |
| "rsi_step2_admit": 0.00039839744567871094, |
| "rsi_step3_solve": 139.55653166770935, |
| "rsi_step4_verify": 1.8809049129486084, |
| "rsi_step7_calibration": 0.0003170967102050781, |
| "rsi_step8_train": 120.59938359260559, |
| "eval": 851.7243919372559 |
| }, |
| "escalations": { |
| "verification": false, |
| "diagnosis": false, |
| "generation": false |
| }, |
| "degradation_count": 0, |
| "plateau_count": 2, |
| "errors": [], |
| "history_summary": [ |
| { |
| "cycle": 1, |
| "pre": 0.5, |
| "post": 0.5, |
| "improvement": 0.0, |
| "eval": 0.49825783972125437, |
| "eval_subdomain": { |
| "code/computing": 1.0, |
| "code/implementation": 0.36585365853658536, |
| "math/symbolic": 0.21052631578947367, |
| "math/word_problem": 0.627906976744186, |
| "math/math": 1.0, |
| "math/probability": 0.0, |
| "math/calculus": 0.0, |
| "math/combinatorics": 1.0, |
| "math/number_theory": 0.7692307692307693, |
| "math/percentage": 1.0, |
| "math/arithmetic": 1.0, |
| "logic/reasoning": 1.0, |
| "logic/syllogism": 0.32142857142857145, |
| "logic/propositional": 1.0 |
| }, |
| "pass_rate": 0.3333333333333333, |
| "had_errors": false |
| }, |
| { |
| "cycle": 2, |
| "pre": 0.46706586826347307, |
| "post": 0.46706586826347307, |
| "improvement": 0.0, |
| "eval": 0.5121951219512195, |
| "eval_subdomain": { |
| "code/computing": 1.0, |
| "code/implementation": 0.34146341463414637, |
| "math/symbolic": 0.23684210526315788, |
| "math/word_problem": 0.627906976744186, |
| "math/math": 1.0, |
| "math/probability": 0.0, |
| "math/calculus": 0.0, |
| "math/combinatorics": 1.0, |
| "math/number_theory": 0.7692307692307693, |
| "math/percentage": 1.0, |
| "math/arithmetic": 1.0, |
| "logic/reasoning": 0.9411764705882353, |
| "logic/syllogism": 0.38095238095238093, |
| "logic/propositional": 1.0 |
| }, |
| "pass_rate": 1.0, |
| "had_errors": false |
| }, |
| { |
| "cycle": 3, |
| "pre": 0.44171779141104295, |
| "post": 0.44171779141104295, |
| "improvement": 0.0, |
| "eval": 0.4843205574912892, |
| "eval_subdomain": { |
| "code/computing": 1.0, |
| "code/implementation": 0.36585365853658536, |
| "math/symbolic": 0.23684210526315788, |
| "math/word_problem": 0.5813953488372093, |
| "math/math": 1.0, |
| "math/probability": 0.0, |
| "math/calculus": 0.0, |
| "math/combinatorics": 1.0, |
| "math/number_theory": 0.6923076923076923, |
| "math/percentage": 1.0, |
| "math/arithmetic": 1.0, |
| "logic/reasoning": 0.9411764705882353, |
| "logic/syllogism": 0.30952380952380953, |
| "logic/propositional": 1.0 |
| }, |
| "pass_rate": 0.8333333333333334, |
| "had_errors": false |
| } |
| ] |
| } |
