{ "problems": 256, "samples": 512, "acc": 0.3515625, "acc_se": 0.025920367008548377, "cut_share": 0.626953125, "mean_tokens": 23911.46875, "closed_share": 0.40234375, "p_finish": 0.373046875, "p_correct_given_finished": 0.9214659685863874, "p_correct_given_cut": 0.012461059190031152, "no_answer_finished_share": 0.001953125, "loop_end_share_of_cut": 0.9470404984423676, "tail_dup8_cut_mean": 0.8846445482866043, "median_tokens_correct": 8115.0, "by_source": { "gsm8k": { "problems": 128, "samples": 256, "acc": 0.45703125, "acc_se": 0.036799286573390676, "cut_share": 0.5078125, "mean_tokens": 20868.8515625 }, "math": { "problems": 128, "samples": 256, "acc": 0.24609375, "acc_se": 0.03418594835093874, "cut_share": 0.74609375, "mean_tokens": 26954.0859375 } }, "run": { "name": "dev/a2", "model": "qwen35-a2", "data": "dev256", "n": 2, "shards": 1, "est": 10 }, "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": "a2", "spec": { "about": "mirror the loop component (reflection): same column sizes, opposite push", "layers": { "15": { "alpha": 2 }, "19": { "alpha": 2 } } }, "built": { "15": { "alpha": 2.0, "n_cols": 64, "dir": "released", "relative_change": 0.03544837236404419 }, "19": { "alpha": 2.0, "n_cols": 64, "dir": "released", "relative_change": 0.02702036127448082 } }, "edited_sha256": { "model.language_model.layers.15.mlp.down_proj.weight": "8790286e9f8319f6ce1ca1ac10b58a7473cb1730e70a11eb5014de458bacdca7", "model.language_model.layers.19.mlp.down_proj.weight": "03c8e5afa093a0931d3cb0cf6251ce5c321aebe4340dd894488839bf242bd447" } }, "tokens_total": 12242672, "gpu_seconds": 2056, "tok_per_s_per_gpu": 6090 }