Fix sequence expand op #11618
Merged
Fix sequence expand op #11618
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
added 12 commits
June 21, 2018 02:42
kuke
reviewed
Jun 26, 2018
| LoDTensor* dx) { | ||
| math::SetConstant<platform::CPUDeviceContext, T> set_zero; | ||
| set_zero(context, dx, static_cast<T>(0)); | ||
| // math::SetConstant<platform::CPUDeviceContext, T> set_zero; |
Contributor
There was a problem hiding this comment.
Please remove these two lines
wanghaoshuang
approved these changes
Jun 27, 2018
Contributor
wanghaoshuang
left a comment
wanghaoshuang
left a comment
There was a problem hiding this comment.
Great job! Actually, the sequence expand op may also give wrong gradient value even memory optimizer is off.
Contributor
|
@wanghaoshuang Please have a test on the attention-based OCR model to make sure that this change solves the problem. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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Outdated suggestions cannot be applied.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.
Sequence expand op's GPU grad kernel implementation is not robust enough if memory optimizer is on.
The GPU kernel directly computed the sum of gradient without checking the initial value in d_x tensor.
In this PR, I moved the "set zero" function outside the functor to guarantee d_x is set to zero both on CPU and GPU.