arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2610.27290cs.AIcs.CLcs.LG
收藏

TEX:基于执行的交叉验证测试时缩放测试智能体

TEX: Test-Time Scaling Testing Agents via Execution-based Cross-Validation

Recursive(递归公司);Salesforce Research(Salesforce研究院);NVIDIA(英伟达公司)

Akash Gokul, Srijan Bansal, Yingbo Zhou, Semih Yavuz

AI总结:

本研究提出TEX协作测试时缩放框架,通过执行交叉验证提升长视野智能体测试用例生成性能,在SWT-Bench Verified上达86.4%准确率,刷新最优水平。

摘要:

测试用例生成对语言模型至关重要,但仍具挑战性。现有测试时缩放方法要么孤立处理候选样本,要么聚合轨迹,无法扩展到长视野任务。我们提出TEX,一种基于执行驱动交叉验证的协作测试时缩放框架。TEX并行运行智能体以生成测试用例和补丁,将每个测试用例与每个补丁交叉执行,形成K×K反馈矩阵。智能体利用自身的矩阵切片进行优化,无需繁琐的轨迹共享即可实现客观验证。在SWT-Bench Verified数据集上,TEX达到86.4%的准确率,创下新的最优性能,较单智能体基线提升9.0%。此外,从TEX输出中随机选择的结果(84.4%)也超过了此前的最优水平(84.0%),表明协作执行反馈可提升解决方案质量,而非仅得到最优样本。我们的结果凸显基于执行的反馈是长视野智能体任务的有效缩放机制。

英文摘要:

Test-case generation is critical but remains challenging for language models. Existing test-time scaling methods either isolate candidates or aggregate traces, which do not scale to long-horizon tasks. We present TEX, a collaborative test-time scaling framework based on execution-driven cross-validation. TEX parallelizes agents to generate tests and patches, cross-executing each test against every patch to form a $K \times K$ feedback matrix. Agents use their matrix slices for refinement, enabling objective validation without verbose trace sharing. On SWT-Bench Verified, TEX achieves 86.4%, establishing a new state-of-the-art and improving +9.0% over single-agent baselines. Moreover, even random selection from TEX outputs (84.4%) surpasses the previous SOTA (84.0%), showing collaborative execution feedback improves solution quality, not just best-case samples. Our results highlight execution-grounded feedback as an effective scaling mechanism for long-horizon agentic tasks.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑