Replace NumPy with Torch in examples/fabric/ #17279
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
github-actions
bot
added
fabric
ryan597
requested review from
williamFalcon,
awaelchli,
carmocca,
justusschock,
lantiga,
Borda and
tchaton
as code owners
|
lightning-fabric (GPUs) (testing pkg: fabric) is failing at the standalone test for I'm not sure how to go about resolving the failed test as the examples I changed aren't going into that test.. |
awaelchli
reviewed
Apr 5, 2023
examples/fabric/reinforcement_learning/train_fabric_decoupled.py
Outdated
Show resolved
Hide resolved
awaelchli
approved these changes
Apr 6, 2023
They are unrelated, you can ignore them and we will rerun them before merging. |
change numpy functions to the torch counterparts remove seeding numpy rng remove the numpy import
replace np.array in layer_init calls remove numpy import
replace the np.sqrt in layer init default arg replace np.logical_or with built-in or, no need for tensors/array. remove numpy import
subtraction was not valid for tensor of bools. Replace with logical not
same changes for all files remove numpy import
for more information, see https://pre-commit.ci
Subtraction with tensors of bools is not supported.
tolist() is significantly slower than numpy() Part of the torch so still allows for removal of numpy import
also to(device) moved within the tensor creation where possible tensor(..., device=device)
remove "from torch import Tensor" in some files where not needed
for more information, see https://pre-commit.ci
for more information, see https://pre-commit.ci
ryan597
force-pushed
the
refactor/16649_numpy_to_torch
branch
from
9bb27c9 to
aa49d64
Compare
justusschock
approved these changes
Apr 6, 2023
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.
What does this PR do?
Refactoring to remove all numpy function calls in the
examples/fabricfolder.Part of Issue: #16649
Related to Pull Requests: #17264, #17267 and #17278
Removal of numpy package import in the listed files.
Most replacements are simple substitutions of
np.<func>totorch.<func>, but also the inclusion of logical_not() where subtraction with tensors of bools are not allowed.Removal of seeding numpy rng
Where.numpy()is called, it has been changed to.tolist()as input to external functions..tolist()is much slower than.numpy()and since.numpy()is part of the torch package this doesn't require the numpy import.Replace
Tensor(...).to(device)withtorch.tensor(..., device=device)Files:
examples/fabric/meta_learning/train_fabric.pyexamples/fabric/meta_learning/train_torch.pyexamples/fabric/reinforcement_learning/rl/agent.pyexamples/fabric/reinforcement_learning/rl/utils.pyexamples/fabric/reinforcement_learning/train_torch.pyexamples/fabric/reinforcement_learning/train_fabric.pyexamples/fabric/reinforcement_learning/train_fabric_decoupled.pyPart of #16649
Before submitting
PR review
Anyone in the community is welcome to review the PR.
Before you start reviewing, make sure you have read the review guidelines. In short, see the following bullet-list:
Reviewer checklist