Powered by
A Docker registry for embedding machine learning models in your Dockerfile.
Usage Model list About Source Code


The registry serves model content from Hugging Face as highly-optimized Docker images, which you can embed in your own container images with just a COPY.
# Copy all files from the model repo to the current WORKDIR
COPY --link / .

# COPY just one file from the model repo to the current WORKDIR
COPY --link /v1-5-pruned.ckpt .

Compared to alternatives like git lfs clone, images build faster, are cached between builds, and only store the model layer once in your registry, reducing storage and bandwidth.

Each model image includes an eStargz index, which allows Docker's BuildKit to download only the files copied from the model during the build. If you build images with Depot, this index is automatically used. Otherwise you will need to enable eStargz in your image builder.

More information →


Missing a model? Open a pull request!
© 2024 Depot Technologies Inc