Adding "spectral clipped" weight decay to be used together with matrix-based preconditioning optimizers #34
Conversation
This file contains hidden or 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
mkhona-nvidia
changed the title
mkhona-nvidia
changed the title
skyw
reviewed
Oct 3, 2025
emerging_optimizers/orthogonalized_optimizers/spectral_clipping_utils.py
Outdated
Show resolved
Hide resolved
emerging_optimizers/orthogonalized_optimizers/spectral_clipping_utils.py
Show resolved
Hide resolved
emerging_optimizers/orthogonalized_optimizers/spectral_clipping_utils.py
Outdated
Show resolved
Hide resolved
emerging_optimizers/orthogonalized_optimizers/spectral_clipping_utils.py
Outdated
Show resolved
Hide resolved
skyw
reviewed
Oct 3, 2025
emerging_optimizers/orthogonalized_optimizers/spectral_clipping_utils.py
Outdated
Show resolved
Hide resolved
emerging_optimizers/orthogonalized_optimizers/spectral_clipping_utils.py
Outdated
Show resolved
Hide resolved
skyw
reviewed
Oct 3, 2025
emerging_optimizers/orthogonalized_optimizers/spectral_clipping_utils.py
Outdated
Show resolved
Hide resolved
mkhona-nvidia
force-pushed
the
mkhona/spectral_clipping_weight_decay
branch
from
8e5d24e to
a00c443
Compare
Signed-off-by: mikail <[email protected]>
Signed-off-by: mikail <[email protected]>
Signed-off-by: mikail <[email protected]>
* save unnecessary matmul Signed-off-by: Hao Wu <[email protected]> * simplify criteria logic Signed-off-by: Hao Wu <[email protected]> * remove max precondition dim Signed-off-by: Hao Wu <[email protected]> Signed-off-by: mikail <[email protected]>
Signed-off-by: mikail <[email protected]>
Signed-off-by: mikail <[email protected]>
Signed-off-by: mikail <[email protected]>
Signed-off-by: mikail <[email protected]>
Signed-off-by: mikail <[email protected]>
Signed-off-by: mikail <[email protected]>
Signed-off-by: mikail <[email protected]>
Signed-off-by: mikail <[email protected]>
Signed-off-by: mikail <[email protected]>
Signed-off-by: mikail <[email protected]>
Signed-off-by: mikail <[email protected]>
Signed-off-by: mikail <[email protected]>
Signed-off-by: mikail <[email protected]>
mkhona-nvidia
force-pushed
the
mkhona/spectral_clipping_weight_decay
branch
from
a00c443 to
4afbc39
Compare
Contributor
Author
|
/ok to test 4afbc39 |
Use math equation in some of the docstrings. Get rid off long lines. Modified some unnecessary content. No function change. Signed-off-by: Hao Wu <[email protected]>
* Update eigen value criteria testing to save compute Signed-off-by: Hao Wu <[email protected]> * remove deprecated function in __all__ Signed-off-by: Hao Wu <[email protected]> --------- Signed-off-by: Hao Wu <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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.
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