Summary of Lorta: Low Rank Tensor Adaptation Of Large Language Models, by Ignacio Hounie et al.
LoRTA: Low Rank Tensor Adaptation of Large Language Models
by Ignacio Hounie, Charilaos Kanatsoulis, Arnuv Tandon, Alejandro Ribeiro
First submitted to arxiv on: 5 Oct 2024
Categories
- Main: Computation and Language (cs.CL)
- 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 Low Rank Adaptation (LoRA) method effectively adapts large pre-trained models for downstream tasks by parameterizing model updates using low-rank matrices at each layer. However, existing LoRA methods still require a significant number of trainable parameters, which can be limiting. Recent works have proposed low rank tensor parameterizations to address this limitation, but they only exploit redundancy across layers or use ad-hoc schemes that introduce additional hyperparameters. This paper proposes a higher-order Candecomp/Parafac (CP) decomposition to enable a more compact and flexible representation for model updates. The authors demonstrate the effectiveness of their method on various benchmarks, including Natural Language Understanding, Instruction Tuning, Preference Optimization, and Protein Folding. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary LoRA is a technique that helps big models learn new skills without needing as many calculations. It does this by using special low-rank matrices to make adjustments at each layer of the model. While LoRA works well, it still requires a lot of resources because it needs to store and update many parameters. Some researchers have tried to fix this by using even lower-rank tensors, but these approaches are limited and require extra setup. In this study, scientists developed a new way to break down model updates into smaller, more manageable pieces using something called Candecomp/Parafac decomposition. This allows for a more efficient use of resources while still achieving good results on various tasks. |
Keywords
» Artificial intelligence » Instruction tuning » Language understanding » Lora » Low rank adaptation » Optimization
