Source: Hugging Face Inference Endpoints docs

Hugging Face's managed dedicated deployment service: Hub model + chosen engine (vLLM, SGLang, TGI, llama.cpp, TEI) packaged as a container with autoscaling and scale-to-zero.

Contents

Summary

Inference Endpoints deploys a Hub model to dedicated, managed infrastructure. You choose the model, the inference engine and the hardware. Hugging Face builds the container, runs it, scales it and monitors it.

Key claims

  • A deployment has three parts: model weights versioned on the Hub, an inference engine, and the managed production infrastructure, which is the product itself.
  • Natively supported engines are vLLM, TGI, SGLang, llama.cpp and Text Embeddings Inference (TEI). Custom containers are also possible.
  • The chosen engine runs as a prebuilt Docker container with the weights pulled from the Hub. Hugging Face manages its lifecycle, including autoscaling and scale-to-zero.
  • Logs and metrics provide observability.
  • Pricing is shown per hour but charged per minute, for the time an endpoint is initializing or running, multiplied by replica count (minimum replicas plus any autoscaled extras). Scaled-to-zero endpoints still count against instance quota; pausing frees it.
  • Team and Enterprise plans add higher GPU quotas, SSO, audit logs, resource groups and private storage.

Feeds

Hugging Face ecosystem

Suggest changes →