| { |
| "step": 286103, |
| "tokens_seen": 75000184832, |
| "metric_for_index": "acc_norm", |
| "intelligence_index": 29.36669386806301, |
| "intelligence_index_raw": 24.006542494002783, |
| "benchmarks": { |
| "hellaswag": { |
| "acc": 36.695877315275844, |
| "acc_norm": 47.63991236805417, |
| "correct": 3685, |
| "correct_norm": 4784, |
| "examples": 10042 |
| }, |
| "arc_easy": { |
| "acc": 54.629629629629626, |
| "acc_norm": 55.513468013468014, |
| "correct": 1298, |
| "correct_norm": 1319, |
| "examples": 2376 |
| }, |
| "arc_challenge": { |
| "acc": 28.92491467576792, |
| "acc_norm": 32.42320819112628, |
| "correct": 339, |
| "correct_norm": 380, |
| "examples": 1172 |
| }, |
| "piqa": { |
| "acc": 69.58650707290533, |
| "acc_norm": 71.21871599564744, |
| "correct": 1279, |
| "correct_norm": 1309, |
| "examples": 1838 |
| }, |
| "arithmark3": { |
| "acc": 37.1, |
| "acc_norm": 35.7, |
| "correct": 371, |
| "correct_norm": 357, |
| "examples": 1000 |
| } |
| }, |
| "examples": { |
| "hellaswag": 10042, |
| "arc_easy": 2376, |
| "arc_challenge": 1172, |
| "piqa": 1838, |
| "arithmark3": 1000 |
| }, |
| "full_split_evaluation": true, |
| "validation_loss": { |
| "timestamp_utc": "2026-10-08T00:13:38.271610+00:00", |
| "step": 286103, |
| "tokens_seen": 75000184832, |
| "metric": "validation_cross_entropy", |
| "validation_ce_loss": 2.7677009465308164, |
| "dataset": "Salesforce/wikitext", |
| "config": "wikitext-2-raw-v1", |
| "split": "validation", |
| "validation_tokens": 366592, |
| "sequences": 179, |
| "full_split": true, |
| "artifact_uri": "hf://buckets/j0no12/sol-pro-checkpoints/logs/validation_loss.jsonl" |
| }, |
| "model_stage": "supervised_multiple_choice_adaptation", |
| "adaptation_step": 2500, |
| "baseline_index": 19.652741613772147 |
| } |
