GitHub PR：新增 Hugging Face Inference Provider 的落地验证清单

适用于“想把 OpenClaw 模型后端扩展到 Hugging Face/Together”的场景：按配置、路由、回归测试逐步上线，降低模型切换风险。

GITHUBDiscovered 2026-02-13Author Josephrp

Prerequisites

You can run a version containing PR #13472 changes and have provider API credentials.
A baseline test suite exists for model quality/latency/cost regression.

Steps

Enable provider config in staging first; keep existing default model as fallback path.
Run representative prompts across old/new providers and compare output stability + tool-call behavior.
Check auth injection and error handling for 401/429/timeout cases before production rollout.
Roll out gradually by traffic slice or selected workflows; monitor failures and cost drift.

Commands

openclaw status

openclaw gateway status

openclaw gateway restart

Verify

New provider completes the same workflow set with acceptable quality and no elevated tool-call failure rate.

Caveats

Provider-side model behavior may differ even under identical prompts; keep per-model prompt tuning.
PR is still open; final config keys/contracts may change（需验证）。

Source attribution

This tip is aggregated from community/public sources and preserved with attribution.