VanishD/CodeGym

CodeGym是一个合成环境生成框架，用于大型语言模型代理在多轮工具使用任务上进行强化学习。该框架将静态代码问题转换为交互式且可验证的环境，代理可以在其中学习使用多种工具集来解决复杂任务，增强在分布外任务上的泛化能力。数据集提供英语和中文两种语言的环境，以及针对不同环境的英语和中文提示的任务实例。

CodeGym is a synthetic environment generation framework for LLM agent reinforcement learning on multi-turn tool-use tasks. It automatically converts static code problems into interactive and verifiable environments where agents can learn to use diverse tool sets to solve complex tasks, enhancing their generalization ability on out-of-distribution tasks. The dataset provides environments in both English and Chinese, as well as task instances with English and Chinese prompts for different environments.