Summary of Sparse High Rank Adapters, by Kartikeya Bhardwaj et al.
Sparse High Rank Adapters
by Kartikeya Bhardwaj, Nilesh Prasad Pandey, Sweta Priyadarshi, Viswanath Ganapathy, Shreya Kadambi, Rafael Esteves, Shubhankar Borse, Paul Whatmough, Risheek Garrepalli, Mart Van Baalen, Harris Teague, Markus Nagel
First submitted to arxiv on: 19 Jun 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 proposes Sparse High Rank Adapters (SHiRA), a novel approach to generative AI research that addresses the limitations of Low Rank Adaptation (LoRA). SHiRA enables rapid switching, incurs no inference overhead, and reduces concept-loss when multiple adapters are used concurrently. It achieves this by directly tuning only 1-2% of the base model weights while leaving others unchanged, resulting in a highly sparse adapter that can be switched directly in the fused mode. The paper presents theoretical and empirical insights on how high sparsity aids multi-adapter fusion and provides extensive experiments on LVMs and LLMs demonstrating its effectiveness. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about a new way to make AI models work better together. Currently, when you combine multiple AI models, it can be slow and use too much computer memory. The authors of this paper have come up with a solution called SHiRA that makes combining models faster and more efficient. They tested their idea on big language models and found that it worked really well. This means that we can now use AI to do things like understand and generate text, images, or videos much better than before. |
Keywords
* Artificial intelligence * Inference * Lora * Low rank adaptation
