Running MCP 33 33 Serverless TextGen Hub ♨ Now supports inference providers and multimodal. No GPU req.