{ "problems": 256, "samples": 512, "acc": 0.607421875, "acc_se": 0.02843871830979484, "cut_share": 0.140625, "mean_tokens": 14588.779296875, "closed_share": 0.845703125, "p_finish": 0.859375, "p_correct_given_finished": 0.7068181818181818, "p_correct_given_cut": 0.0, "no_answer_finished_share": 0.072265625, "loop_end_share_of_cut": 0.2638888888888889, "tail_dup8_cut_mean": 0.2930152777777778, "median_tokens_correct": 7714, "by_source": { "gsm8k": { "problems": 128, "samples": 256, "acc": 0.71875, "acc_se": 0.037100065528307684, "cut_share": 0.046875, "mean_tokens": 10288.83984375 }, "math": { "problems": 128, "samples": 256, "acc": 0.49609375, "acc_se": 0.04094122599425784, "cut_share": 0.234375, "mean_tokens": 18888.71875 } }, "run": { "name": "dev/a8", "model": "qwen35-a8", "data": "dev256", "n": 2, "shards": 1, "est": 8 }, "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": { "name": "a8", "spec": { "about": "stronger than released: flip and multiply by 7", "layers": { "15": { "alpha": 8 }, "19": { "alpha": 8 } } }, "built": { "15": { "alpha": 8.0, "n_cols": 64, "dir": "released", "relative_change": 0.1417999267578125 }, "19": { "alpha": 8.0, "n_cols": 64, "dir": "released", "relative_change": 0.10807694494724274 } }, "edited_sha256": { "model.language_model.layers.15.mlp.down_proj.weight": "aff5c48746038da1d27e71977e548b151078ae7c40c7dc83c46de0c119bb7517", "model.language_model.layers.19.mlp.down_proj.weight": "d34e317e46979afa1fa56c2e0c2d4feca5ef9cc366172a700adeed661ee176c8" } }, "tokens_total": 7469455, "gpu_seconds": 1088, "tok_per_s_per_gpu": 7263 }