ProgramGym：面向长周期程序复现智能体的可扩展训练环境
原标题：Paper page - ProgramGym: Scaling Training Environments for Long-Horizon Program Reproduction Agents
AI 摘要
该论文提出 ProgramGym 框架，包含 Program-Env 和 Program-Agent 两部分，用于为长周期程序复现智能体构建可扩展、可验证的训练环境。Program-Env 通过反馈驱动循环从代码仓库构建环境并修复运行时沙箱与黑盒测试质量，共获得覆盖 C、C++、Go、Rust 的 1,648 个训练环境；Program-Agent 结合分析、实现、审查的角色隔离工作流与修订管理。实验显示，用这些环境轨迹微调可将 Qwen3.8-27B 从 34.87% 提升至 42.84%，Program-Agent 进一步将基础模型和 SFT 模型提升至 55.71% 和 56.25%，超过部分旗舰模型。
正文节选
ProgramGym: Scaling Training Environments for Long-Horizon Program Reproduction Agents Abstract Coding agents are increasingly used to build programs from scratch as software engineers do rather than to resolve issues within a repository or to generate a repository from a given requirements document. However, scalable verifiable training environments that force agents to discover what is required and to decide how the system should be designed remain scarce. Moreover, existing agent workflows al