Add scale option to spectral normalization. #3904
Open
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Outdated suggestions cannot be applied.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.
What does this PR do?
Adds a scale parameter to spectral normalization. One can increase the expressivity while retaining some benefits of spectral norm by parameterizing the spectral norm with a single scale parameter. This forces the spectral norm to be independent of the dimensionality of the weight matrix. This was originally proposed in Appendix E of [1] and recently applied to transformer models in [2].
[1] Takeru Miyato, Toshiki Kataoka, Masanori Koyama, and Yuichi Yoshida, 'Spectral Normalization for Generative Adversarial Networks', International Conference on Learning Representations (ICLR), 2018.
[2] Shuangfei Zhai, Tatiana Likhomanenko, Etai Littwin, Dan Busbridge, Jason Ramapuram, Yizhe Zhang, Jiatao Gu, and Joshua M. Susskind, 'Stabilizing Transformer Training by Preventing Attention Entropy Collapse'. International Conference on Machine Learning (ICML), 2023.
Checklist
checks if that's the case).
discussion (please add a
link).
documentation guidelines.
(No quality testing = no merge!)