Post by Xinference

Xinference v2.12.0 unlocks model autonomy — your models now self-recover after any system restart, no human in the loop. What this means for you: 🔁 No more restart anxiety — Models marked with autostart come back on their own. A 10-model cluster recovers in seconds, not the 30 minutes it took you to manually relaunch each one. 🔐 Access control that actually exists — API Key lifecycle management, user/role permissions, resource-scoped isolation. Multiple teams sharing one cluster without running naked. 📊 See what's broken, fast — Grafana split into 6 focused sub-panels. Worker / Supervisor / Replica metrics unified in Prometheus. New replica_index tells you exactly which replica is lagging. 🧠 New models, ready to run — VibeThinker 3B (reasoning-chain small model), Nex-N2 Pro (FP8 quantized), Ornith-1.0-35B, Unlimited-OCR. 🖥 Test without code — "Try to API" button on every running model, right in the Web UI. 📦 Upgrade: pip install 'xinference==2.12.0' 🔗 GitHub: https://lnkd.in/gbcqWtR3 📖 Docs: https://lnkd.in/gmzJEXq8 🌐 https://xinference.co #OpenSource #AIInfrastructure #LLM #ModelServing #Xinference #MLOps