Skip to content

Conversation

@mkhona-nvidia
Copy link
Contributor

Based on @leloykun's spectral clipping algorithm

Also updated Polar Express Newton-schulz coefficients to their "stable" version as described by the paper, excluding the last entry

@copy-pr-bot
Copy link

copy-pr-bot bot commented Oct 3, 2025

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

@mkhona-nvidia mkhona-nvidia requested a review from skyw October 3, 2025 22:36
@mkhona-nvidia
Copy link
Contributor Author

\ok to test 64c9aa9

@mkhona-nvidia mkhona-nvidia force-pushed the mkhona/spectral_weight_decay branch from fe42f31 to a170f44 Compare October 3, 2025 22:41
@mkhona-nvidia mkhona-nvidia force-pushed the mkhona/spectral_weight_decay branch from 4f7d2c2 to e46e361 Compare October 3, 2025 22:44
@mkhona-nvidia
Copy link
Contributor Author

\ok to test e46e361

@skyw
Copy link
Contributor

skyw commented Oct 3, 2025

/ok to test b6cc6b6

@skyw skyw enabled auto-merge (squash) October 3, 2025 23:06
@skyw skyw merged commit beacaab into NVIDIA-NeMo:main Oct 3, 2025
12 checks passed
mkhona-nvidia added a commit to mkhona-nvidia/Emerging-Optimizers that referenced this pull request Oct 7, 2025
…ased preconditioning optimizers (NVIDIA-NeMo#38)

* added spectral clipping from previous PR

Signed-off-by: mikail <[email protected]>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants