meshapplied / goethe Serve your models fast on your own infra in 3 steps. Created Oct 2026 gpu inference llm nvfp4 nvidia Total80