This repository has been archived by the owner on Jul 7, 2023. It is now read-only.
Enable depth_k != depth_v in local_attention_2d and masked_local_attention_2d #1899
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Outdated suggestions cannot be applied.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 resolves a TODO in tensor2tensor/layers/common_attention_test.py : enable
depth_v != depth_k
forcommon_attention.local_attention_2d
andcommon_attention.masked_local_attention_2d
Modification is simple: one just needs to alter the shape passed to
scatter_blocks_2d
when generating the output, so that its last dimension is taken fromv
, and notq
Tests:
pytest tensor2tensor/layers/common_attention_test.py tensor2tensor/layers/common_image_attention_test.py
passesv
is split along the depth dimension, then applyinglocal_attention_2d
on each of the two parts separately and concatenating the results should be equivalent to applyinglocal_attention_2d
with originalv
- see the code belowExpand sanity check code and output