Draft
Conversation
Implements NOBLE from arXiv:2603.06492 which augments transformer linear layers with nonlinear low-rank branches that can accelerate training by up to 1.47x. Key changes: - Add noble.py with CosNet activation, NobleBranch, NobleLinear, and NobleConfig - Update attention.py to support optional NOBLE branches on Q/K/V/proj and MLP layers - Update flexi_vit.py to pass noble_config through FlexiVitBase, Encoder, Predictor - Add EncoderConfig.noble_config and PredictorConfig.noble_config - Add scripts/official/noble.py experiment script for NOBLE training - Add comprehensive unit tests in tests/unit/nn/test_noble.py NOBLE computes: output = xW + σ(xW_down)W_up where σ is CosNet (two-layer cosine nonlinearity with learnable frequency/phase). Paper reports: - Up to 1.47x step speedup to reach baseline eval loss - ~4% additional parameters (at scale) with 7% step time overhead - Up to 1.22x net wallclock speedup Note: Paper found Mixup/CutMix interferes with NOBLE benefits. Co-authored-by: henryh <henryh@allenai.org>
Co-authored-by: henryh <henryh@allenai.org>
|
Cursor Agent can help with this pull request. Just |
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Implement NOBLE (Nonlinear Low-Rank Branches) to accelerate transformer pretraining, based on arXiv:2603.06492.