arXiv cs.CL· Haoyi Yang, Kshitij Mishra, Nils Lukas, Salem Lahlou·· 2 天前AI 评分24
HeirCommit 与 EvoHeir：自进化语言模型如何选择继承哪一代模型
Selection Is Not Learning: Choosing Model Futures
AI 导读
针对 R-Zero、Agent0、Skill Self-Play 等自进化循环中"任务看似可学却未必带来有效更新"的选择-更新缺口，HeirCommit 冻结更新后的模型并用全新评测题决定是否继承，在相同接受率下将有害更新占比相对随机继承降低 68-83%。EvoHeir 进一步分离参数继承与经验继承，在四个骨干模型上取得最佳数学与通用推理均值，数学均值较最强基线 Agent0 提升 5.3%。
正文 · 原文
Abstract:Self-evolving language models train on tasks they generate, so every round must decide which updated model the next round inherits. Current loops such as R-Zero, Agent0, and Skill Self-Play make this decision through the task, admitting tasks that look learnable and inheriting every resulting update. We identify a selection-to-update gap in these loops, where a task that looks learnable need not produce a useful update. Most updates on verified tasks fail to make correct answers more likely, yet these loops inherit them all. HeirCommit freezes the updated model and uses fresh evaluation questions to decide whether to inherit it or retain the current model. EvoHeir separates parameter inheritance from experience inheritance, using current and archived experience to repair failures and preserve successes before final confirmation. Our analysis characterizes selection-induced optimism and decomposes regret into selection, absorption, and confirmation errors. HeirCommit cuts the harmful share of accepted updates by 68-83% relative to random inheritance at the same acceptance rate. Across four backbones, EvoHeir achieves the best mathematics and general-reasoning averages, raising the math average by 5.3% relative to the strongest baseline, Agent0. The framework lets useful experience persist even when the model that produced it is not inherited.
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.22029 [cs.LG] |
| (or arXiv:2610.22029v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.22029 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Haoyi Yang [view email]
[v1]
Thu, 15 Oct 2026 15:07:09 UTC (298 KB)
来源：arXiv cs.CL · arxiv.org