cs.LG提交于 2026-10-23已译

SWE-Verifier：面向成本高效编码智能体的选择性验证

SWE-VERIFIER: Selective Verification for Cost-Efficient Coding Agents

Kaijian Zou · Vaibhav Chennareddy · Yunxiang Zhang · Naihao Deng · Longju Bai · Jiaxin Pei · et al.

中文摘要

编码智能体需要耗费大量计算来生成动作并处理长交互历史，不良动作可能使其进入低价值状态，造成计算浪费和 API 成本上升。对替代动作进行采样与验证可改善决策，但每一步都执行会带来显著开销。SWE-Verifier 是一种轻量级生成模型，用于选择性触发候选动作采样与验证：先为智能体提出的动作评分，判断是否需要额外采样，再对所得动作排序后执行。该模型基于 Qwen3.5-4B 训练，偏好数据将差异动作及其环境观察和基于评分准则的教师分数组织成组。在 mini-swe-agent 框架中，MiniMax-M2.7 的问题解决率由 75.47% 提升至 77.20%，成本为 1.09 倍；未在其动作上训练时，模型也可泛化至 DeepSeek-V4-Flash，问题解决率由 72.60% 提升至 74.27%，成本为 1.13 倍。两个模型上，SWE-Verifier 验证的步骤均不足 4%，以显著更低成本取得与轨迹基线和全步骤基线相当或更高的解决率。模型、代码和数据已公开。

关键要点

01问题：全步骤采样与验证开销高昂，低质量动作还会浪费计算和 API 成本。
02方法：SWE-Verifier 评分当前动作，仅在必要时触发候选采样、执行前排序并验证。
03训练：基于 Qwen3.5-4B，使用对比动作、环境观察和准则引导教师分数训练。
04结果：MiniMax-M2.7 解决率从 75.47% 升至 77.20%，成本为 1.09 倍。
05泛化与效率：DeepSeek-V4-Flash 解决率升至 74.27%，且仅验证不足 4% 的步骤。

解读

尚无解读。

原始英文摘要

arXiv:2610.30284v1 Announce Type: new Abstract: Coding agents spend substantial computation generating actions and processing long interaction histories. Poor action choices can steer agents into low-value states, wasting computation and increasing API costs. Sampling and verifying alternative actions can improve the agent's decisions, but applying this procedure at every step adds substantial overhead. We introduce SWE-Verifier, a lightweight generative model that selectively triggers candidate sampling and verification. It scores the coding agent's proposed action to determine when additional candidate sampling is needed, then ranks the resulting actions before execution. We train SWE-Verifier from Qwen3.5-4B using preference data that groups contrasting actions with their resulting environment observations and rubric-guided teacher scores. Using the mini-swe-agent harness, SWE-Verifier improves resolution from 75.47% to 77.20% on MiniMax-M2.7 at 1.09 times cost and generalizes to DeepSeek-V4-Flash without training on its actions, improving resolution from 72.60% to 74.27% at 1.13 times cost. Across both models, SWE-Verifier verifies fewer than 4% of steps and achieves comparable or higher resolution than the trajectory and all-step baselines at substantially lower cost. These results demonstrate that selective candidate sampling and lightweight verification provide complementary gains in coding-agent performance and inference efficiency. Our model, code, and data are available at https://github.com/zkjzou/SWE-Verifier.