Support S32/U32 indices for BWD embedding & Neuron implicit downcast #8462
Conversation
This file contains 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
rpsilva-aws
changed the title
|
FYI, I split the previous PR: @miladm @ManfeiBai @tengyifei, this one is needed for 2.6. Unfortunately #8463 has a dependency on PT. |
tengyifei
requested changes
Dec 6, 2024
rpsilva-aws
force-pushed
the
rpsilva_downcast_v2
branch
2 times, most recently
from
c2fb7ef to
95d0f0c
Compare
tengyifei
approved these changes
Dec 6, 2024
rpsilva-aws
force-pushed
the
rpsilva_downcast_v2
branch
from
95d0f0c to
75f37b0
Compare
|
@tengyifei Ran yapf over the test file. PTAL, thanks! |
rpsilva-aws
force-pushed
the
rpsilva_downcast_v2
branch
from
75f37b0 to
a210920
Compare
tengyifei
approved these changes
Dec 6, 2024
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Outdated suggestions cannot be applied.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.
In this PR, we extend embedding tensor operations to allow S32 indices. This follows suits with other operations, in order to add flexibility and potentially performance benefits for accelerator backends. Reference for embedding dense bwd: https://github.com/pytorch/pytorch/blob/main/aten/src/ATen/native/Embedding.cpp#L117
In addition, we also re-introduce the implicit downcasting for Neuron S64/U64 types, since the Neuron compiler does not support 64 bits.
There is an ongoing effort to further extend this requirement to other tensor operations involving indices: pytorch/pytorch#142160. Once this is resolved, we adapt it on XLA as well.