Skip to content

Conversation

@kausv
Copy link
Contributor

@kausv kausv commented Nov 18, 2025

Summary:
X-link: https://github.com/facebookresearch/FBGEMM/pull/2144

Support 2D weights in 1D length permute tensor kernel.
This kernel is invoked on variable strides per rank

https://github.com/meta-pytorch/torchrec/blob/main/torchrec/sparse/jagged_tensor.py#L3111-L3134

2D weights are needed for write dist where the weights are actually embedding values.

Differential Revision: D87261479

@meta-codesync
Copy link
Contributor

meta-codesync bot commented Nov 18, 2025

@kausv has exported this pull request. If you are a Meta employee, you can view the originating Diff in D87261479.

@meta-cla meta-cla bot added the cla signed label Nov 18, 2025
kausv added a commit to kausv/FBGEMM that referenced this pull request Nov 18, 2025
Summary:

X-link: facebookresearch/FBGEMM#2144

Support 2D weights in 1D length permute tensor kernel.
This kernel is invoked on variable strides per rank. Similar to D80466458

https://www.internalfb.com/code/fbsource/[5e1d1c3734c75d0664ba817ee05ce9a91a1f02e4]/fbcode/torchrec/sparse/jagged_tensor.py?lines=3111-3134

2D weights are needed for write dist where the weights are actually embedding values.

Differential Revision: D87261479
kausv added a commit to kausv/FBGEMM that referenced this pull request Nov 18, 2025
Summary:

X-link: facebookresearch/FBGEMM#2144

Support 2D weights in 1D length permute tensor kernel.
This kernel is invoked on variable strides per rank. Similar to D80466458

https://github.com/meta-pytorch/torchrec/blob/main/torchrec/sparse/jagged_tensor.py#L3111-L3134

2D weights are needed for write dist where the weights are actually embedding values.

Differential Revision: D87261479
kausv added a commit to kausv/FBGEMM that referenced this pull request Nov 18, 2025
Summary:

X-link: facebookresearch/FBGEMM#2144

Support 2D weights in 1D length permute tensor kernel.
This kernel is invoked on variable strides per rank. Similar to D80466458

https://github.com/meta-pytorch/torchrec/blob/main/torchrec/sparse/jagged_tensor.py#L3111-L3134

2D weights are needed for write dist where the weights are actually embedding values.

Reviewed By: spcyppt, q10

Differential Revision: D87261479
kausv added a commit to kausv/FBGEMM that referenced this pull request Nov 18, 2025
Summary:

X-link: facebookresearch/FBGEMM#2144

Support 2D weights in 1D length permute tensor kernel.
This kernel is invoked on variable strides per rank. Similar to D80466458

https://github.com/meta-pytorch/torchrec/blob/main/torchrec/sparse/jagged_tensor.py#L3111-L3134

2D weights are needed for write dist where the weights are actually embedding values.

Reviewed By: spcyppt, q10

Differential Revision: D87261479
Summary:

X-link: facebookresearch/FBGEMM#2144

Support 2D weights in 1D length permute tensor kernel.
This kernel is invoked on variable strides per rank. Similar to D80466458

https://github.com/meta-pytorch/torchrec/blob/main/torchrec/sparse/jagged_tensor.py#L3111-L3134

2D weights are needed for write dist where the weights are actually embedding values.

Reviewed By: spcyppt, q10

Differential Revision: D87261479
@meta-codesync
Copy link
Contributor

meta-codesync bot commented Nov 18, 2025

This pull request has been merged in e1fc43e.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants