Skip to content

Conversation

@runame
Copy link
Contributor

@runame runame commented Dec 15, 2025

Summary: Sync KL-Shampoo implementation to OSS. Also, add KL-Shampoo reference to README.

Differential Revision: D89193149

@meta-cla meta-cla bot added the CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. label Dec 15, 2025
@meta-codesync
Copy link

meta-codesync bot commented Dec 15, 2025

@runame has exported this pull request. If you are a Meta employee, you can view the originating Diff in D89193149.

Summary:

Sync KL-Shampoo implementation to OSS. Also, add KL-Shampoo reference to README.

**Original summary:**

Adding a simplified version of [KL-Shampoo](https://arxiv.org/abs/2509.03378). The main difference to the algorithm presented in the paper is that here only the update of the factor matrices is adjusted, but the eigenvalue correction to the individual factor matrices is not implemented. This correction can also be used for regular Shampoo and will be added in the future.

KL-Shampoo is implemented by storing either the inverse root matrices directly, or storing the eigendecompositions of the factor matrices, analogous to regular Shampoo.

Reviewed By: hjmshi

Differential Revision: D89193149
@meta-codesync
Copy link

meta-codesync bot commented Dec 16, 2025

This pull request has been merged in 89f00f7.

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

Labels

CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. fb-exported Merged meta-exported

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants