| { |
| "problems": 256, |
| "samples": 512, |
| "acc": 0.162109375, |
| "acc_se": 0.018982408061987328, |
| "cut_share": 0.845703125, |
| "mean_tokens": 28614.287109375, |
| "closed_share": 0.154296875, |
| "p_finish": 0.154296875, |
| "p_correct_given_finished": 0.9873417721518988, |
| "p_correct_given_cut": 0.011547344110854504, |
| "no_answer_finished_share": 0.0, |
| "loop_end_share_of_cut": 0.9676674364896074, |
| "tail_dup8_cut_mean": 0.9179789838337182, |
| "median_tokens_correct": 5353, |
| "by_source": { |
| "gsm8k": { |
| "problems": 128, |
| "samples": 256, |
| "acc": 0.20703125, |
| "acc_se": 0.029621621243045373, |
| "cut_share": 0.80078125, |
| "mean_tokens": 27368.25 |
| }, |
| "math": { |
| "problems": 128, |
| "samples": 256, |
| "acc": 0.1171875, |
| "acc_se": 0.02319013159717069, |
| "cut_share": 0.890625, |
| "mean_tokens": 29860.32421875 |
| } |
| }, |
| "run": { |
| "name": "dev/base+pp2", |
| "model": "base", |
| "data": "dev256", |
| "n": 2, |
| "shards": 1, |
| "est": 15, |
| "set": [ |
| "presence_penalty=2.0" |
| ] |
| }, |
| "sampling": { |
| "temperature": 0.6, |
| "top_p": 0.95, |
| "top_k": 20, |
| "presence_penalty": 2.0, |
| "repetition_penalty": 1.0, |
| "max_tokens": 32768, |
| "thinking": true |
| }, |
| "edit_note": null, |
| "tokens_total": 14650515, |
| "gpu_seconds": 2576, |
| "tok_per_s_per_gpu": 5786 |
| } |
