This repository was archived by the owner on Jul 7, 2023. It is now read-only.
Fix transformer_moe model has wrong logic in pre/postprocessing #1233
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
|
Thanks a lot @twilightdema 👍 |
tensorflow-copybara
pushed a commit
that referenced
this pull request
Nov 21, 2018
PiperOrigin-RevId: 222429349
kpe
pushed a commit
to kpe/tensor2tensor
that referenced
this pull request
Mar 2, 2019
kpe
pushed a commit
to kpe/tensor2tensor
that referenced
this pull request
Mar 2, 2019
PiperOrigin-RevId: 222429349
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
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Outdated suggestions cannot be applied.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.
There is a wrong logic in transformer_moe model that make training loss not decrease and make decoding always generate empty output.
By comparing logic with transformer.py and common_attention.py, I found that 'dp_postprocess' should receive input 'x' before passing 'dp_preprocess' in order to have it run with the same logic with transformer model. I changed the logic as in this commit and ran test data to confirm that training loss is decrease and decoding generate correct result.
Unit Testing Result:
Before Fix:
After Fix:
