[backryun]

Summary

• Updated the HuggingFace provider to use the current Hugging Face router OpenAI-compatible chat endpoint.
• Added HuggingFace /v1/models discovery support so “Import from /models” fetches chat models from the Hugging Face router instead of falling back to unrelated local audio models.
• Refreshed HuggingFace ASR/TTS model catalog entries.

Validation

• npm run lint
• npm run test:unit
• npm run test:coverage
• Coverage is still >= 60% for statements, lines, functions, and branches
• SonarQube PR analysis is green or any remaining issues are explicitly documented below

Validation performed:

• npx prettier --write open-sse/config/audioRegistry.ts open-sse/config/providerRegistry.ts 'src/app/api/providers/[id]/models/route.ts'
• node --import tsx/esm -e "await import('./open-sse/config/audioRegistry.ts'); await import('./open-sse/config/providerRegistry.ts'); console.log('changed registry imports ok')"
• Commit hook passed prettier, eslint --fix, docs sync, any-budget, env/docs sync, CLI i18n, and UI i18n coverage checks.
• npm run typecheck:core -- --pretty false was attempted but failed on existing open-sse/services/contextManager.ts never type errors unrelated to this change.

Tests Added Or Updated

• No automated test files changed.

Coverage Notes

• This PR changes open-sse/ provider registry data and a provider model-discovery route under src/.
• No production logic tests were added; validation was manual/runtime-oriented around registry importability and the dashboard “Import from /models” behavior.
• Coverage was not re-run, so no coverage movement was measured.

Reviewer Notes

• HuggingFace chat now targets https://router.huggingface.co/v1/chat/completions.
• HuggingFace model import now targets https://router.huggingface.co/v1/models.
• HuggingFace audio ASR/TTS remains on the separate HF Inference task API path, which is distinct from the OpenAI-compatible chat router.

[backryun]

The existing huggingface configuration was completely incorrect, and we confirmed that it works very well after modification.
Some model lists were reconfigured to include only supported models by manually querying them via the API.

[gemini-code-assist]

Code Review

This pull request updates the HuggingFace provider configuration to support dynamic model discovery and refreshes the audio registry with newer ASR and TTS models. Feedback suggests retaining a few 'seed' models in the configuration to provide a fallback if remote discovery fails and highlights the need for automated tests to comply with the repository's style guide for production code changes.

[diegosouzapw]

Obrigado @backryun pela contribuição! 🙌

O fix do HuggingFace foi integrado na release/v3.8.0. A migração para router.huggingface.co/v1/chat/completions com discovery dinâmico de modelos via /v1/models é uma melhoria significativa — agora os usuários podem importar modelos diretamente em vez de depender de uma lista estática desatualizada.

Este fix estará no OmniRoute v3.8.0. 🚀