truss

Why we included this project

Truss is the command-line tool you reach for when a Hugging Face model needs to become a live API without writing a Dockerfile or touching Kubernetes. You describe the model, hardware, and engine in a config file, and `uvx truss push` builds the serving container and deploys it, so the same packaged setup behaves in local development and in production. It covers most major Python frameworks, from transformers and diffusers to PyTorch, TensorFlow, vLLM, SGLang, and TensorRT-LLM, which keeps one workflow usable across LLMs, image, and audio models. It is built around Baseten and its GPU-backed autoscaling, but the resulting container can run on your own infrastructure too. For teams that want a short path from a trained model to an OpenAI-compatible endpoint, this is a practical way to get there.