Running MCP 39 Serverless TextGen Hub ♨ 39 Now supports inference providers and multimodal. No GPU req.