Added TF specific documentation to DistributedEmbedding.
#94
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
cantonios
requested changes
May 21, 2025
| `DistributedEmbedding` is a layer optimized for TPU chips with SparseCore | ||
| and can dramatically improve the speed of embedding lookups and embedding | ||
| training, in particular for large embedding tables. It works by combining | ||
| multiple lookups in one invocations, and by sharding the embedding tables |
There was a problem hiding this comment.
multiple lookups into one invocation,
| and can dramatically improve the speed of embedding lookups and embedding | ||
| training, in particular for large embedding tables. It works by combining | ||
| multiple lookups in one invocations, and by sharding the embedding tables | ||
| across the available chips. |
There was a problem hiding this comment.
Sharding doesn't necessarily make it faster on its own. That leads to more communication for the collective, which can actually slow the lookup down if the table would otherwise fit on a single chip. We might want to specify that it can make lookups faster for large tables that already need to be model-sharded.
|
|
||
| ### Using in a Keras model | ||
|
|
||
| The use Keras' `model.fit()`, one must compile the model under the |
| using the strategy and also automatically distributes the dataset. | ||
|
|
||
| ```python | ||
| model = create_model(embedding) |
There was a problem hiding this comment.
Should you create the model under the scope as well, since the layer needs to be?
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Outdated suggestions cannot be applied.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.
No description provided.