Penn State 与 Microsoft Research 提出 Behavioral Self-Rewarding 改进 SWE 智能体训练
训练编程智能体只看测试通过与否，组内全过或全败就没信号。这篇论文让模型自己 pairwise 比较轨迹当奖励，Qwen3-14B 白拿 2.6 个点。
Penn State 和 Microsoft Research 的研究人员在论文中指出，SWE 智能体通常基于二值测试奖励训练，当一组轨迹全部通过或全部失败时，组内没有对比信号，学不到东西。他们提出 Behavioral Self-Rewarding 方法：策略对自己的轨迹两两比较，将偏好转化为有界小奖励叠加在测试结果之上。由于奖励有界，自我评判只能区分两条都通过的轨迹，不会让失败轨迹排在通过轨迹之前，且无需新的环境交互。在 SWE-bench Verified 上，Qwen3-14B 在 400 步预算下平均准确率从 32.1% 提升到 34.7%，在 SWE-bench-Live 上从 6.33% 提升到 8.11%。
// Behavioral Self-Rewarding for SWE agents //
It's standard practice to train SWE agents on binary test rewards.
This paper shows the problem with that. When every trajectory in a group passes, or every one fails, the reward has no contrast, and the group teaches nothing.
Researchers from Penn State and Microsoft Research propose Behavioral Self-Rewarding. The policy compares its own trajectories in pairs, and those preferences become small bounded rewards added on top of the test result.
The added rewards are bounded, so a self-judged preference can rank two passing trajectories but can never lift a failing one above a passing one.
No new environment interactions are needed. On SWE-bench Verified, Qwen3-14B improves from 32.1% to 34.7% average accuracy at a 400-step budget, and on SWE-bench-Live from 6.33% to 8.11%.
https://t.co/ekD2h7ya4y
Chat with Paper: https://t.co/6vUbYmIXRR
