Update Dockerfile, fix compatibility with PyTorch>=2.3.0 #614
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Co-authored-by: Luciferian Ink <[email protected]>
mryab
changed the title
mryab
added a commit
that referenced
this pull request
Oct 13, 2024
* Update Docker base image and Python version * Update optim.grad_scaler to use torch.amp --------- Co-authored-by: Luciferian Ink <[email protected]> (cherry picked from commit 96f9ba3)
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The current Dockerfile is based on an outdated version of CUDA, and the Python dependency specification allows Python 3.12 to be installed (which is not supported by Hivemind at the moment). Lastly, this PR fixes an error in
hivemind.optim.grad_scalerthat emerged to a change in the API of PyTorch by using eithertorch.amportorch.cuda.ampdepending on the version of PyTorch.This PR fixes all three issues: to make all of CI workflows pass, both the Python version fix and the
torch.ampfix are necessary.