Fully revamp aten embedding bag for JAX backend #8535
Merged
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Outdated suggestions cannot be applied.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.
Replaces non-jittable code with a new, JAX-compatible approach for computing embedding bags.
Handles offsets=None by performing a straightforward reduction (sum, mean, or max) across each row of the embedded tensor, aligning with PyTorch’s behavior for multi-dimensional indices.
Computes and returns offset2bag, bag_size, and max_indices when offsets is given, matching each mode (sum, mean, or max).
Converts offset2bag and bag_size to JAX arrays before returning, maintaining consistent data types.
Ensures the function’s return signature fully matches PyTorch expectations