inference
Operate and troubleshoot OpenAssistant inference server, workers, text client, safety server, model configs, websocket work protocol, and SSE chat flows.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- inference
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Operate and troubleshoot OpenAssistant inference server, workers, text client, safety server, model configs, websocket work protocol, and SSE chat flows.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/api-reference.md6.6 KB
- 参考文档references/model-configs.md3.6 KB
- 参考文档references/troubleshooting.md4.5 KB
- 参考文档references/workflows.md4.3 KB
- 脚本scripts/check_inference_config.py4.0 KB
03 SKILL.md 正文
OpenAssistant inference sub-skill
Use this sub-skill when a task is about the OpenAssistant local inference stack: the FastAPI inference server, websocket workers, text client, safety server, model configuration selection, chat streaming, plugin calls, GPU sizing, or lightweight _lorem/distilgpt2 smoke checks.
Route the request
- Server/worker/chat protocol, debug auth, route map, websocket request/response types, SSE events, plugins, or safety parameters: read
references/api-reference.md. - Model config names,
_lorem,distilgpt2, OpenAssistant SFT/RLHF variants, quantization, or GPU memory sizing: readreferences/model-configs.mdand runscripts/check_inference_config.pywhen a checkout is available. - Local Docker compose inference stack, text client, worker startup, full dev tmux flow, safety server, or load testing: read
references/workflows.md. - Failures involving API keys, websocket status, protocol upgrade, Redis/Postgres, TGI connection, model downloads, GPU OOM, SSE parsing, plugin OpenAPI, or safety server availability: read
references/troubleshooting.md.
Safe checks
These checks do not download models or start services:
python scripts/check_inference_config.py --repo-root <repo-root> --model-config _lorem
python scripts/check_inference_config.py --repo-root <repo-root> --model-config distilgpt2 --json
python scripts/check_inference_config.py --repo-root <repo-root> --list
Use _lorem when the user needs a CPU-only logical smoke path. Use distilgpt2 only when a tiny Hugging Face model download/cache is acceptable or already available. Real OpenAssistant model configs may require large downloads and GPU memory.
Boundaries
- Route website chat components, browser-side API calls, and frontend SSE rendering to the
websitesub-skill. - Route the data-collection backend, task lifecycle, shared REST API client, and OA JSONL exports to the
backendsub-skill. - Model training/evaluation/pretokenization, production deployment, Ansible, and infrastructure operations are excluded from this generated skill run.
- Do not run a real model worker, download model weights, use credentials, or start Docker services unless the user explicitly requests it and accepts resource/network 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).