inference-providers
Operate the TaskingAI inference microservice provider/model catalog, credential validation, chat completion, text embedding, rerank, provider icons, custom/local providers, proxy safety, and provider troubleshooting.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- inference-providers
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Operate the TaskingAI inference microservice provider/model catalog, credential validation, chat completion, text embedding, rerank, provider icons, custom/local providers, proxy safety, and provider troubleshooting.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/api-reference.md13.0 KB
- 参考文档references/native-testing.md7.8 KB
- 参考文档references/provider-catalog.md12.0 KB
- 参考文档references/troubleshooting.md11.1 KB
- 脚本scripts/inspect_inference_catalog.py14.4 KB
03 SKILL.md 正文
TaskingAI Inference Providers
Use this sub-skill when a task is about TaskingAI's inference microservice: provider and model schema discovery, provider credentials, credential verification, chat completion, text embedding, rerank, provider icons, OpenAI-compatible/custom hosts, local model providers, proxy/URL blacklist behavior, or provider-specific failure diagnosis.
Load this first
- For provider/model selection, credential fields, wildcard schemas, local/custom-host choices, and catalog counts, read references/provider-catalog.md.
- For HTTP route contracts, request/response bodies, validation order, streaming, fallbacks, proxy/custom-header behavior, and icon routes, read references/api-reference.md.
- For source-backed native-test behavior, credential-sensitive checks, and safe skip criteria, read references/native-testing.md.
- For symptoms, likely causes, and recovery steps, especially
PROVIDER_ERRORversusREQUEST_VALIDATION_ERROR, read references/troubleshooting.md. - To inspect a user-supplied TaskingAI source tree without network calls or credentials, use scripts/inspect_inference_catalog.py.
Source-backed facts to preserve
- The verified inference catalog snapshot contains 35 provider directories.
- Model schema types are exactly
chat_completion,text_embedding,rerank, andwildcard. - Provider and model metadata are loaded from bundled YAML resources into runtime caches before route handlers use them.
- Static import probes passed in Python 3.10 for
APIRouter,Provider, andDebugChatCompletionModel. - Credential verification can be schema-only at provider level or can call a default chat, embedding, or rerank model depending on provider metadata.
Quick routing
- Backend object lifecycle, model-instance APIs exposed by the main backend, backend proxy endpoints, authentication, and database/API object semantics belong to
../backend-api/. - Deployment variables, container startup, service URLs, image tags, port wiring,
TASKINGAI_INFERENCE_URL, global outbound proxy setup, and storage configuration belong to../deployment-configuration/. - Plugin tool bundles, plugin vision/tool schemas, plugin execution, and plugin image generation belong to
../plugin-bundles/. - Stay here for provider catalog semantics, inference route payloads, credentials,
custom_host, Ollama/LM Studio/LocalAI, provider icons, and provider-specific troubleshooting.
Operating workflow
- Classify the requested model capability. Use
chat_completion,text_embedding,rerank, orwildcard; do not assume a wildcard schema has enough information withoutprovider_model_id,model_type, and any requiredproperties. - Choose the provider path. Use provider selection guidance for hosted providers, hosted OpenAI-compatible routers,
custom_host, and local OpenAI-compatible servers. - Validate credentials before invocation. Compare the request to the provider
credentials_schema; pass either plaintextcredentialsorencrypted_credentials, never both. Use credential validation semantics when the user needs encrypted credentials for later reuse. - Build the inference payload. Use chat completion, text embedding, or rerank contracts. Check model properties and configs before provider calls.
- Handle proxy and URL safety deliberately. Request-level
proxyvalues must be HTTPS and are rejected if they contain a configured provider URL blacklist entry.custom_hostendpoint URLs are checked against the same blacklist before a provider call. - Diagnose by error class. Treat schema/model/property/proxy failures as request validation first. Treat upstream HTTP/API/auth/quota failures as provider failures. Use troubleshooting before retrying with new credentials.
Hard usability cases covered here
- Choose between
custom_host, Ollama/LM Studio/LocalAI, OpenRouter, and a named hosted provider for an OpenAI-compatible user task while preserving the difference between actual provider model IDs and TaskingAI model schema IDs. - Diagnose
PROVIDER_ERRORversusREQUEST_VALIDATION_ERRORduring credential validation by separating local schema/model/proxy checks from upstream provider responses.
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).