| { |
| "problems": 256, |
| "samples": 512, |
| "acc": 0.12109375, |
| "acc_se": 0.01577554619307842, |
| "cut_share": 0.88671875, |
| "mean_tokens": 29682.626953125, |
| "closed_share": 0.115234375, |
| "p_finish": 0.11328125, |
| "p_correct_given_finished": 1.0, |
| "p_correct_given_cut": 0.00881057268722467, |
| "no_answer_finished_share": 0.0, |
| "loop_end_share_of_cut": 0.9691629955947136, |
| "tail_dup8_cut_mean": 0.8995321585903084, |
| "median_tokens_correct": 5600.5, |
| "by_source": { |
| "gsm8k": { |
| "problems": 128, |
| "samples": 256, |
| "acc": 0.15234375, |
| "acc_se": 0.02514634439642162, |
| "cut_share": 0.86328125, |
| "mean_tokens": 28956.9765625 |
| }, |
| "math": { |
| "problems": 128, |
| "samples": 256, |
| "acc": 0.08984375, |
| "acc_se": 0.01875294739793572, |
| "cut_share": 0.91015625, |
| "mean_tokens": 30408.27734375 |
| } |
| }, |
| "run": { |
| "name": "dev/base", |
| "model": "base", |
| "data": "dev256", |
| "n": 2, |
| "shards": 1, |
| "est": 15, |
| "priority": 1 |
| }, |
| "sampling": { |
| "temperature": 0.6, |
| "top_p": 0.95, |
| "top_k": 20, |
| "presence_penalty": 1.5, |
| "repetition_penalty": 1.0, |
| "max_tokens": 32768, |
| "thinking": true |
| }, |
| "edit_note": null, |
| "tokens_total": 15197505, |
| "gpu_seconds": 2691, |
| "tok_per_s_per_gpu": 5770 |
| } |
