Support PyTorch 2.0.0 #559
Merged
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
borzunov
added a commit
that referenced
this pull request
Mar 28, 2023
PyTorch 2.0 is not supported yet, it's support will be added in #559 (there are multiple issues to resolve). Until then, we need to require `torch<2.0.0` (otherwise 2.0.0 is installed, so CI doesn't work right now). This PR also adds Python 3.10 to the CI.
|
I've rebased this PR to the master branch after #558 was merged. |
borzunov
reviewed
Mar 28, 2023
| @@ -1,5 +1,5 @@ | |||
| PyYAML | |||
| torch>=1.9.0,<2.0.0 | |||
| torch>=1.9.0 | |||
There was a problem hiding this comment.
We'll have torch 1.13.0 on Python 3.7, torch 2.0.0 on Python 3.8-3.10. This allows us to test both torch 1.13.0 and 2.0.0.
We can drop Python 3.7 and PyTorch < 2.0 support when necessary.
borzunov
reviewed
Mar 28, 2023
hivemind/optim/optimizer.py
Outdated
| def zero_grad(self, set_to_none: bool = False): | ||
| _SET_TO_NONE_DEFAULT = Version(torch.__version__).major >= 2 | ||
|
|
||
| def zero_grad(self, set_to_none: bool = _SET_TO_NONE_DEFAULT): |
There was a problem hiding this comment.
So it's consistent with the torch-wide default.
mryab
reviewed
Mar 28, 2023
tests/test_optimizer.py
Outdated
Comment on lines
154
to
156
| assert (model1.weight.grad is None or torch.all(model1.weight.grad == 0)) and ( | ||
| model2.weight.grad is None or torch.all(model2.weight.grad == 0) | ||
| ), "zero grad did not trigger" |
There was a problem hiding this comment.
Can we also use Version(torch.__version__).major >= 2 here? (or at least import _SET_TO_NONE_DEFAULT)
This way, the test won't have a bug where we always set the grad to None regardless of the PyTorch version
mryab
approved these changes
Mar 28, 2023
Codecov Report
@@ Coverage Diff @@
## master #559 +/- ##
==========================================
- Coverage 76.01% 75.92% -0.09%
==========================================
Files 81 81
Lines 7995 8008 +13
==========================================
+ Hits 6077 6080 +3
- Misses 1918 1928 +10
|
mryab
pushed a commit
that referenced
this pull request
Mar 31, 2023
- Fix LRSchedulerBase - Handle None after .zero_grad() in torch 2.0.0 - Use set_to_none=True by default in torch>=2.0 - Add set_to_none param to TrainingStateAverager.step() Co-authored-by: Aleksandr Borzunov <[email protected]> (cherry picked from commit 98531ce)
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.
@borzunov: