Resets rnn _flat_weights on _apply #28562
Closed
Conversation
This file contains hidden or 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
Contributor
facebook-github-bot
left a comment
facebook-github-bot
left a comment
There was a problem hiding this comment.
@mruberry has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
Contributor
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Outdated suggestions cannot be applied.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.
Currently when _apply() is called on RNNBase (or one of its children, like LSTM), the _flat_weights attribute may or may not be updated. In particular, when using .to() and sending a module like LSTM to XLA, a third party device type, the tensors in _flat_weights will not be updated and will remain on CPU. This causes the LSTM forward to fail since the forward call receives a mix of XLA and CPU tensors.
This occurs because third party device types, like XLA, may not be a compatible shallow copy type to native tensors. When this is the case and _apply is called Module parameters are replaced, not updated. RNNBase would not sync _flat_tensors with its params in this case, and that caused the references in _flat_tensors to not reflect the module's current params.
This small change forces a resync of the _flat_tensors and the actual params on each _apply. This lets .to('xla') work for LSTMs, for example. A test will be added to PyTorch/XLA (which runs in our CI) to validate this behavior after the change appears in PyTorch.