Lightning-AI / Lightning-AI/litgpt
LoRA matrices dropout
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 13.7k
- Forks
- 1.5k
- Avg merge
- 15h 37m
- Merged PRs (30d)
- 1
Description
Hi everyone,
[recently has been proposed](https://arxiv.org/abs/2404.09610v1) to apply the dropout directly on the LoRA weight matrices A and B: this favors sparsity which improve generalization and reduce overfitting. The dropout is only applied on input/output dimension to avoid reducing the matrices rank.
If you guys think that this could be helpful I can submit a PR with the feature.
Thanks
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Read the linked paper and locate the existing LoRA implementation in litgpt. Compare the proposed input/output-only dropout with current behavior and clarify the API, defaults, and tests with maintainers. Done requires a settled feature specification or accepted implementation plan; this issue currently does not define an implementation target.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- ai, machine-learning
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100