Add a Python alternative to seq2seq.gather_tree #1925
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
|
Marking this PR as draft. The fixture from #1929 is needed to run tests for both the custom op and the py op. |
|
Tests are now running for this new implementation. |
qlzh727
approved these changes
Jun 16, 2020
ashutosh1919
pushed a commit
to ashutosh1919/addons
that referenced
this pull request
Jul 12, 2020
* Add a Python alternative to seq2seq.gather_tree * Enable tests for the Python op
jrruijli
pushed a commit
to jrruijli/addons
that referenced
this pull request
Dec 23, 2020
* Add a Python alternative to seq2seq.gather_tree * Enable tests for the Python op
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Outdated suggestions cannot be applied.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.
This PR adds a pure Python TensorFlow implementation of
tfa.seq2seq.gather_tree, which can be enabled with the global flagTF_ADDONS_PY_OP. This is useful when Addons custom ops are not readily available (e.g. TensorFlow Serving).I tested the performance on a real world application: beam search decoding of a neural machine translation model. I'm using a custom beam search implementation but it is close to the
BeamSearchDecoderincluded in Addons.I did not find significant performance impact. And I think this makes sense: the function is only used at the very end of the decoding and does not involve very complex ops.