fix time_distributed layer with mask and partial_batch_size #20765
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Outdated suggestions cannot be applied.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.
If the model includes an Embedding layer with
mask_zero = True
parameter and sub sequent model has time distributed layer it is observed that training fails in graph mode if there ispartial_batch_size
. This happens due to concatenation of partial batch dataset which makesbatch_size
to None and hence shape to (None,...).Hence the model fails with graph execution error if we try to compare batch_size with the respective value from mask.
Hence I am proposing to omit the
batch_size
comparison for TF backend with graph mode. It would have been better if this check is for when there is actually apartial_batch_size
but not sure how to propogate this to time distributed layer .Fixes #20754
Code to replicate the issue: