BCIT：自主 LLM 后训练中的条件经验迁移方法
原标题：Paper page - Knowing When Not to Reuse: Conditional Experience Transfer in Autonomous LLM Post-Training
AI 摘要
Hugging Face 每日论文页面介绍了一篇关于自主 LLM 后训练中条件经验迁移的论文。论文提出边界校准干预迁移（BCIT）方法，通过检查上下文适用性并运行有界试验来选择性地重用过去的训练证据，减少有害更新并提高最终模型质量。在 Qwen3-4B 模型上的金融推理、文本到 SQL 和函数调用等任务中，BCIT 相比基线方法在相同计算预算下取得了更高的最终模型质量，平均任务分数提升 2.63 分。
正文节选
Knowing When Not to Reuse: Conditional Experience Transfer in Autonomous LLM Post-Training Abstract Boundary-Calibrated Intervention Transfer selectively reuses past training evidence by checking contextual applicability and running bounded trials, reducing harmful updates and improving final model quality in autonomous post-training. Large language models offer broad capabilities, but adapting them to evolving domains, tools, and requirements often entails repeated post-training. Autonomous sys