How to use GPUs

[pt247]

Reference Issues or PRs

Closes #119

What does this implement/fix?

Put a x in the boxes that apply

• [ ] Bug fix (non-breaking change which fixes an issue)
• [ ] New feature (non-breaking change which adds a feature)
• [ ] Breaking change (fix or feature that would cause existing features not to work as expected)
• [x] Documentation Update
• [ ] Code style update (formatting, renaming)
• [ ] Refactoring (no functional changes, no API changes)
• [ ] Build related changes
• [ ] Other (please describe):

Testing

• [x] Did you test the pull request locally?
• [ ] Did you add new tests?

Documentation

Access-centered content checklist

Text styling

• [x] The content is written with plain language (where relevant).
• [x] If there are headers, they use the proper header tags (with only one level-one header: H1 or # in markdown).
• [x] All links describe where they link to (for example, check the Nebari website).
• [x] This content adheres to the Nebari style guides.

Non-text content

• [x] All content is represented as text (for example, images need alt text, and videos need captions or descriptive transcripts).
• [x] If there are emojis, there are not more than three in a row.
• [x] Don't use flashing GIFs or videos.
• [x] If the content were to be read as plain text, it still makes sense, and no information is missing.

Any other comments?

[netlify[bot]]

✅ Deploy Preview for nebari-docs ready!

Name	Link
🔨 Latest commit	5c37974f93f2e46c286236561602b437fc180531
🔍 Latest deploy log	https://app.netlify.com/sites/nebari-docs/deploys/66704f1b787a0d0008146422
😎 Deploy Preview	https://deploy-preview-471--nebari-docs.netlify.app
📱 Preview on mobile	Toggle QR Code... Use your smartphone camera to open QR code link.

---

To edit notification comments on pull requests, go to your Netlify site configuration.

[ericdatakelly]

If you add just a little more info, you could remove Best Practices for PyTorch on Nebari. Also, I think that info is slightly out of date. pytorch-gpu is now pytorch-cuda as shown by "Run this command" from pytorch.org matrix. Also for the conda yaml config, I suggest including the channels recommended by the pytorch matrix so the user has a complete example. They need to be above conda-forge.

channels:
  - pytorch
  - nvidia
  - conda-forge
dependencies:
  - pytorch
  - ipykernel
variables:
  CONDA_OVERRIDE_CUDA: "12.1"

Have you tested each method (2.1, 2.2) to be sure they work and you are describing all steps?

[pt247]

If you add just a little more info, you could remove Best Practices for PyTorch on Nebari. Also, I think that info is slightly out of date. pytorch-gpu is now pytorch-cuda as shown by "Run this command" from pytorch.org matrix. Also for the conda yaml config, I suggest including the channels recommended by the pytorch matrix so the user has a complete example. They need to be above conda-forge.

Agreed, I have updated it now.

Have you tested each method (2.1, 2.2) to be sure they work and you are describing all steps?

I have now tested and updated the document.

[ericdatakelly]

I think at this point, we can remove https://deploy-preview-471--nebari-docs.netlify.app/docs/how-tos/pytorch-best-practices from the docs. @ericdatakelly is there anything else we need to include here to ensure coverage of the same material?

I added a suggestion above, so if that's incorporated in this page, PyTorch Best Practices is no longer needed.

[kcpevey]

Thanks @ericdatakelly for your review!

@pt247 can you remove this file? docs/static/img/how-tos/conda-store-yaml-toggle.png

[ericdatakelly]

Why remove pytorch and nvidia channels? The pytorch matrix tells you where the packages are, so you should include them. Conda forge sometimes does not have the latest packages.

[kcpevey]

Why remove pytorch and nvidia channels? The pytorch matrix tells you where the packages are, so you should include them. Conda forge sometimes does not have the latest packages.

I think I am torn here between doing what the pytorch website tells you vs what folks who use this for work tell us (which is, pull everything from conda-forge if you can, sometimes conda-forge and pytorch channels have conflicting builds)

[ericdatakelly]

Why remove pytorch and nvidia channels? The pytorch matrix tells you where the packages are, so you should include them. Conda forge sometimes does not have the latest packages.

I think I am torn here between doing what the pytorch website tells you vs what folks who use this for work tell us (which is, pull everything from conda-forge if you can, sometimes conda-forge and pytorch channels have conflicting builds)

Hmmm... I have the opposite experience in my work. Using pytorch and nvidia channels tends to work best 8/10 times. Maybe a middle ground is to show conda forge only, but include a note that says to add pytorch and nvidia if the conda forge solve fails.

[kcpevey]

We are going to leave pytorch in the channel list since there are still some pytorch related packages that are not yet on conda-forge. We'll leave both pytorch and nvidia in the "follow the pytorch docs" since this is what they suggest.