Fix nan gradient by using native complex abs op #1013
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
This was referenced Nov 6, 2020
mthrok
force-pushed
the
fix-spectrogram-backward
branch
from
a305fb8 to
05bcf4d
Compare
mthrok
commented
Nov 6, 2020
| n_fft=2048, | ||
| hop_length=None, | ||
| win_length=None, | ||
| power=1, |
There was a problem hiding this comment.
@anjali411 Should I also ensure this when power is other than 1?
gibiansky
reviewed
Nov 6, 2020
mthrok
force-pushed
the
fix-spectrogram-backward
branch
2 times, most recently
from
54df934 to
defc8b7
Compare
vincentqb
reviewed
Nov 10, 2020
torchaudio/functional/functional.py
Outdated
Comment on lines
112
to
95
| if power == 1.0: | ||
| return spec_f.abs() | ||
| if power == 2.0 and spec_f.is_cuda: | ||
| return torch.view_as_real(spec_f).pow(power).sum(-1) | ||
| return spec_f.abs().pow(power) | ||
| return torch.view_as_real(spec_f) |
There was a problem hiding this comment.
This logic could be moved to complex_norm directly, no? This would benefit other locations.
There was a problem hiding this comment.
yeah I agree we should move this logic to complex_norm
torchaudio/functional/functional.py
Outdated
Comment on lines
114
to
115
| if power == 2.0 and spec_f.is_cuda: | ||
| return torch.view_as_real(spec_f).pow(power).sum(-1) |
There was a problem hiding this comment.
Can you explain why power == 2.0 and spec_f.is_cuda must be treated differently from the generic case on line 116 below?
There was a problem hiding this comment.
I'm assuming this is the reason. It's not clear the gain is significant. If this is indeed to go in, I would add a comment here.
anjali411
approved these changes
Nov 11, 2020
vincentqb
approved these changes
Nov 11, 2020
mthrok
force-pushed
the
fix-spectrogram-backward
branch
from
defc8b7 to
ed798ad
Compare
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.
Resolves #993