Can uv integrate with e.g. pytorch prebuilt docker env?

Question

So, pytorch requires a rather large bundle of packages. The prebuilt docker pytorch gpu images (https://docs.nvidia.com/deeplearning/frameworks/pytorch-release-notes/running.html) are quite helpful in this regard.

On the other hand, uv and its lockfile are also a very handy utility.

How to marry those two? I.e. I am looking for a dockerfile, which roughly looks like

FROM nvcr.io/nvidia/pytorch:<xx.xx>-py3

# more preparation ...

RUN uv sync --locked # does NOT fetch pytorch again

# some more preparation ...

CMD ["python", "-m", ""src.main:train_model"]

or, alternatively: Is it possible to tell uv somehow that some deps, albeit mentioned in the uv.lock, are provided by the global environment which provides python and pytorch?

colonel · Accepted Answer · 2025-10-23 18:25:22Z

2

You can explicitly skip packages in the lockfile using --exclude or --no-install

FROM nvcr.io/nvidia/pytorch:<xx.xx>-py3

# more preparation ...

# Sync environment, but exclude torch-related deps already present in the base image
RUN uv sync --locked --exclude torch --exclude torchvision --exclude torchaudio

# some more preparation ...

CMD ["python", "-m", "src.main:train_model"]

answered Oct 23 at 18:25

colonel

8885 silver badges8 bronze badges

Sign up to request clarification or add additional context in comments.

1 Comment

helt Oct 24 at 9:40

cool. And do you know how to stick to the provided python, too?

Collectives™ on Stack Overflow

Can uv integrate with e.g. pytorch prebuilt docker env?

1 Answer 1

1 Comment

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

1 Comment

Your Answer

Sign up or log in

Post as a guest

Related