devcontainers / devcontainers/images

[Notice] Removing "torch" package from the "universal" image

Open
#856 3 comments 3 reactions 0 assignees View on GitHub
notice
Dominant language
Shell
Stars
2.1k
Forks
963
Avg merge
7h 48m
Merged PRs (30d)
11

Description

Hello 👋

We are thinking of removing the [torch](https://pypi.org/project/torch/) package from the [universal](https://github.com/devcontainers/images/tree/main/src/universal) image. Feel free to share your concerns or thoughts, thank you! ✨

### Background

Currently, the size of universal image is `12.7 GB` and we have been looking at ways to shrink it. It’s interesting to see that the [machine learning packages](https://github.com/devcontainers/images/blob/main/src/universal/.devcontainer/local-features/machine-learning-packages/install.sh#L41-L49) consume the most space which is `5.24 GB`. We understand, that we want to support ML relevant scenarios, however, 41% kinda seems like a big chunk of the image.

### Alternative

As `torch` package is installed with `pip`, it is very simple to add to your dev container (either in a Dockerfile or with a lifecycle command).

Also, we are thinking of adding a new Feature option to the [Python Feature](https://github.com/devcontainers/features/tree/main/src/python) (eg. `packages`) which accepts a list of packages which can be added with `pip`. The user can install `torch` or any other packages they would like to with the update Python Feature. Would appreciate any feedbacks on this as well.

### Conclusion

The `torch` package is `1.4 GB` and it’s add a dependency of `nvidia` which is another `3.0 GB` . Hence, removing the `torch` package saves us `~ 4.4 GB`.

Contributor guide

No contributing guide indexed for this repository

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.