- Switched from large-v3 (1.56B) to distil-large-v3 (~756M) — 2x faster transcription
- Increased CPU threads from 4 to 8 (Ryzen 5800X3D has 8c/16t)
- French language quality preserved (~5% accuracy tradeoff for 2x speed)
- Expected pipeline latency: 4-5s total (down from ~15s)
- effi-stt: Déchargement Whisper après 300s inactivité (INACTIVITY_TIMEOUT)
* Nouveaux endpoints: /unload, /reload, /stats
* Modelo rechargé à la première requête
- ComfyUI: Script monitor qui décharge VRAM après inactivité
* Vérifie usage VRAM toutes les 60s
* Exécute workflow UnloadAllModelsNode après 300s inactivité
- Kreuzberg: Manager qui redémarre le container après seuil mémoire + inactivité
* Restart si > 2GB pendant > 10min inactif
- Documentation: docs/MEMOIRE-GESTION.md
- FastAPI server with OpenAI-compatible API
- Integration with faster-whisper (large-v3)
- Full Docker support with docker-compose
- Support for multiple languages (FR/EN/Auto)
- Clean, maintainable codebase (~100 lines)
- Health endpoint and error handling
- Volume caching for models
Folzo