协作头脑风暴
brainstorming
在实现前澄清需求、探索设计选择，并形成清晰的开发方向。
探索技能
开发与工程 · SKILL.md
serving-runtime
serving-runtime【开发与工程】Launch and debug SGLang serving runtime, OpenAI-compatible APIs, server arguments, cache/scheduler behavior, parallelism, and production serving caveats.
独立 GitHub 快照 · 自动分类 · 以原作者文档为准
技能说明
GitHubname: serving-runtime description: "Launch and debug SGLang serving runtime, OpenAI-compatible APIs, server arguments, cache/scheduler behavior, parallelism, and production serving caveats." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0
Serving Runtime
Use this sub-skill for sglang serve, launch_server.py, ServerArgs, HTTP/OpenAI-compatible endpoints, tokenizer/engine settings, tensor/data/expert parallelism, radix/KV cache, speculative decoding, disaggregation, and server runtime failures.
Start by separating argument parsing, model/tokenizer loading, CUDA visibility, backend dependency availability, and request-shape errors. Avoid recommending a full server run unless the environment has the pinned SGLang runtime dependencies installed.
Cross-Links
- Root routing:
../../SKILL.md - Workflow map:
../../references/workflow-map.md - Cross-cutting troubleshooting:
../../references/troubleshooting.md