Summary of More Fine-tuning with 10x Fewer Parameters, by Wenxuan Tan et al.
MoRe Fine-Tuning with 10x Fewer Parameters
by Wenxuan Tan, Nicholas Roberts, Tzu-Heng Huang, Jitian Zhao, John Cooper, Samuel Guo, Chengyu Duan, Frederic Sala
First submitted to arxiv on: 30 Aug 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The paper presents a novel framework called Monarch Rectangular Fine-tuning (MoRe) that efficiently searches over adapter architectures for parameter-efficient fine-tuning (PEFT). MoRe relies on the Monarch matrix class and is more expressive than low-rank adapters (LoRA), which are commonly used PEFT techniques. The approach outperforms state-of-the-art PEFTs on various tasks and models, requiring as few as 5% of LoRA’s parameters. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary MoRe is a simple framework that helps PEFT work better. It uses something called the Monarch matrix class to find the best adapter architectures. This means it can be used with different models and even new ones that haven’t been tried before. The paper shows that MoRe is more powerful than another popular method, LoRA. It also works well with fewer parameters required. |
Keywords
» Artificial intelligence » Fine tuning » Lora » Parameter efficient
