| { | |
| "problems": 16, | |
| "samples": 32, | |
| "acc": 0.1875, | |
| "acc_se": 0.07739239842086129, | |
| "cut_share": 0.84375, | |
| "mean_tokens": 28374.84375, | |
| "closed_share": 0.15625, | |
| "p_finish": 0.15625, | |
| "p_correct_given_finished": 1.0, | |
| "p_correct_given_cut": 0.037037037037037035, | |
| "no_answer_finished_share": 0.0, | |
| "loop_end_share_of_cut": 1.0, | |
| "tail_dup8_cut_mean": 0.9051074074074074, | |
| "median_tokens_correct": 5698.5, | |
| "run": { | |
| "name": "quick/base", | |
| "model": "base", | |
| "data": "dev256", | |
| "n": 2, | |
| "shards": 1, | |
| "est": 1, | |
| "args": [ | |
| "--limit", | |
| "16" | |
| ] | |
| }, | |
| "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": 907995, | |
| "gpu_seconds": 274, | |
| "tok_per_s_per_gpu": 3948 | |
| } |