TF: T5 can now handle a padded past (i.e. XLA generation) #17969
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
|
related issue: #17935 |
|
The documentation is not available anymore as the PR was closed or merged. |
|
Great job on finding and fixing the bug here @gante - cool that T5 works now :-) |
viclzhu
pushed a commit
to viclzhu/transformers
that referenced
this pull request
Jul 18, 2022
…e#17969) * get the right slicing index for position_bias
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Outdated suggestions cannot be applied.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?
In TF T5, we now fetch the correct slice of
position_bias-- the same way we do it in FLAX. The key difference is that FLAX relies on an external variable for the generated length that gets incremented every time past gets updated, and here the same value is obtained dynamically from the past array (latest filled past index = generated length - 1, where latest filled past index corresponds to the maximum index with non-0 values).All slow tests are passing and we no longer have length restrictions on the XLA beam search test, which means that: