Ship current Tabby Tavern compose (TabbyAPI EXL3, SillyTavern cards, Open WebUI dual backend, Ollama 11435, SearXNG JSON, MCPO). Hugging Face model card README + DEVLOG aligned with the live lab. Placeholders only.
Add in-container CUDA/GPU tuning: env vars for TabbyAPI EXL3 + Ollama, shared_memory 16g, Dockerfile ENV defaults, README tuning docs