SWE-Serve 如何揭示本地测试与线上服务之间的鸿沟
How SWE-Serve Exposes the Gap Between Local Tests and Live Serving
NVIDIA 发布 SWE-Serve，用于评估推理服务软件改动——AI 编程智能体的补丁能通过测试，却可能在服务器加载真实模型并处理请求时失败。该基准由 SGLang 团队参与开发，包含 53 个任务，检查系统能否通过公开接口返回正确结果。
An AI coding agent’s patch can pass tests yet fail when the server loads a real model and handles requests. Evaluating changes to inference-serving software...
An AI coding agent’s patch can pass tests yet fail when the server loads a real model and handles requests. Evaluating changes to inference-serving software therefore requires checking the full serving path, including whether the system returns correct results through its public interface. Developed with input from the SGLang team, SWE-Serve evaluates this gap with 53 tasks derived from…
来源：NVIDIA Developer Blog · developer.nvidia.com