用大规模训练与测试数据打造高效开源编程代理
SWE-Dev: Building Software Engineering Agents with Training and Inference Scaling
- 自动生成高质量测试用例，用于代码修复评估
- 7B和32B模型在SWE-bench-Verified上分别达23.4%和36.6%成功率
- 适合研究编程代理、自动化开发的开发者与研究人员
大语言模型（LLM）已从对话求解演进到支持工具调用的真实任务，如软件工程（SWE）。尽管OpenAI Codex和Cursor等工具包实现了开发流程的端到端自动化，但构建高效SWE代理仍面临高质量训练数据与有效测试用例不足的挑战。为此，我们提出基于开源LLM的SWE-Dev代理。首先，构建稳健的流水线以合成用于补丁评估的测试用例；其次，通过扩展代理轨迹来构建训练数据。在SWE-bench-Verified基准上的实验表明，SWE-Dev模型在所有开源SWE代理中表现领先：7B和32B参数模型的成功率分别为23.4%和36.6%，显著优于现有开源模型。所有代码、模型与数据集均公开于https://github.com/THUDM/SWE-Dev。
原文摘要 · Abstract (English)
Large language models (LLMs) have advanced rapidly from conversational problem solving to addressing real-world tasks involving tool use, such as software engineering (SWE). Recent LLM-powered toolkits, such as OpenAI Codex and Cursor, have offered end-to-end automation of the software development process. However, building effective SWE agents remains challenging due to the lack of high-quality training data and effective test cases. To address this issue, we present SWE-Dev, an SWE agent built upon open-source LLMs. First, we develop a robust pipeline to synthesize test cases for patch evaluation. Second, we scale up agent trajectories to construct the training data for building SWE-Dev. Experiments on the SWE-bench-Verified benchmark show that the SWE-Dev models can achieve top performance among all open SWE agents. Specifically, the success rates of the SWE-Dev 7B and 32B parameter models reach 23.4% and 36.6%, respectively, outperforming state-of-the-art open-source models. All code, models, and datasets are publicly available at https://github.com/THUDM/SWE-Dev.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。