Add Experimental limited sparse embedding bag #8905
+939
−0
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
amjames
force-pushed
the
amjames/sparse_embedding_bag
branch
from
aacfd1b to
c67588f
Compare
|
Hi @amjames if this passes test and is finished feel free to publish as PR and merge. |
amjames
force-pushed
the
amjames/sparse_embedding_bag
branch
from
c67588f to
c58b539
Compare
amjames
commented
May 7, 2025
amjames
commented
May 7, 2025
amjames
force-pushed
the
amjames/sparse_embedding_bag
branch
from
c58b539 to
f394790
Compare
amjames
changed the title
|
A note for reviewers: the failure in xla_op1 shard appears to be unrelated, the tests for the new feature are in xla_op3 |
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Outdated suggestions cannot be applied.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.
Users of
torch_xlaencounter an issue when using thesparse=Trueoption with theEmbeddingorEmbeddingBagmodules.The gradient for weight is created as a sparse tensor and there is no dispatch registered for the combination of sparse creation APIs w/ the
XLAkey, or theSparsefunctionality key and theXLAbacked key used in conjunction.This is a workaround that can be removed, ported to C++, or extended later:
F.embeddingF.embedding_bag,nn.Embedding, andnn.EmbeddingBagwhich forward to a custom implementation of the backward and produce the above tensor subclass rather than a native torch sparse tensor.The tensor subclass may have component tensors
indicesandvalueswhich have xla device without issue.fixes #8719