serving
Launch and reason about OpenChat's OpenAI-compatible vLLM/FastAPI server.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- serving
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Launch and reason about OpenChat's OpenAI-compatible vLLM/FastAPI server.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/api-reference.md7.2 KB
- 参考文档references/deployment.md7.3 KB
- 参考文档references/troubleshooting.md8.8 KB
- 脚本scripts/run_openchat_server.sh1.2 KB
03 SKILL.md 正文
OpenChat serving sub-skill
Use this sub-skill when the task is to launch, secure, call, or debug OpenChat's OpenAI-compatible API server implemented by python -m ochat.serving.openai_api_server.
Route by task
- Launching the server, choosing
--model,--model-type, GPU/Ray/tensor-parallel flags, CORS, API keys, or logging: read references/deployment.md. - Calling
/v1/modelsor/v1/chat/completions, constructing chat payloads, streaming,n, sampling,condition, context limits, and model aliases: read references/api-reference.md. - Debugging missing dependencies, CUDA/vLLM/Ray errors, 404 model names, 400 context length failures, API-key failures, ignored system prompts, or unsupported request fields: read references/troubleshooting.md.
- Use scripts/run_openchat_server.sh as the safe installed-package wrapper; it forwards to the module and refuses non-help launches without
--model.
Boundaries
- In scope: serving, deployment, security/logging knobs, OpenAI-compatible API usage, vLLM/Ray/GPU prerequisite reasoning.
- Route prompt-template internals, tokenization details, and
conditionsemantics beyond serving behavior to ../prompting/SKILL.md. - Route benchmark harness, MT-Bench/AlpacaEval/HumanEval evaluation flows, and answer matching to ../evaluation/SKILL.md.
- Treat Docker, SSH, and cloudflared files as reference-only deployment evidence, not bundled helpers to run from this skill.
- Source training, data generation, scripts, and experimental notebooks are outside this sub-skill.
Safety defaults
Actual serving requires model weights, compatible CUDA/PyTorch/vLLM/Ray packages, and enough GPU memory for the selected model and tensor-parallel plan. Do not start a long-running listener unless the user explicitly asks for a launch and has supplied the model source, host/port exposure intent, and security requirements.
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).