使用SWE-Gym训练软件工程代理程序和验证器

Training Software Engineering Agents and Verifiers with SWE-Gym

December 30, 2024

作者: Jiayi Pan, Xingyao Wang, Graham Neubig, Navdeep Jaitly, Heng Ji, Alane Suhr, Yizhe Zhang

cs.AI

摘要

我们提出了SWE-Gym，这是用于训练真实世界软件工程（SWE）代理的第一个环境。SWE-Gym包含2,438个真实世界的Python任务实例，每个实例包括一个带有可执行运行环境、单元测试和自然语言任务描述的代码库。我们使用SWE-Gym来训练基于语言模型的SWE代理，在流行的SWE-Bench Verified和Lite测试集上实现高达19%的绝对解决率提升。我们还尝试通过在从SWE-Gym中采样的代理轨迹上训练的验证器进行推理时间缩放。与我们微调的SWE代理相结合时，在SWE-Bench Verified和Lite上分别实现32.0%和26.0%，体现了开放权重SWE代理的最新技术水平。为了促进进一步研究，我们公开发布了SWE-Gym、模型和代理轨迹。

English

We present SWE-Gym, the first environment for training real-world software engineering (SWE) agents. SWE-Gym contains 2,438 real-world Python task instances, each comprising a codebase with an executable runtime environment, unit tests, and a task specified in natural language. We use SWE-Gym to train language model based SWE agents , achieving up to 19% absolute gains in resolve rate on the popular SWE-Bench Verified and Lite test sets. We also experiment with inference-time scaling through verifiers trained on agent trajectories sampled from SWE-Gym. When combined with our fine-tuned SWE agents, we achieve 32.0% and 26.0% on SWE-Bench Verified and Lite, respectively, reflecting a new state-of-the-art for open-weight SWE agents. To facilitate further research, we publicly release SWE-Gym, models, and agent trajectories.