Summary of Adazeta: Adaptive Zeroth-order Tensor-train Adaption For Memory-efficient Large Language Models Fine-tuning, by Yifan Yang et al.
AdaZeta: Adaptive Zeroth-Order Tensor-Train Adaption for Memory-Efficient Large Language Models Fine-Tuning
by Yifan Yang, Kai Zhen, Ershad Banijamal, Athanasios Mouchtaris, Zheng Zhang
First submitted to arxiv on: 26 Jun 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
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 proposed Adaptive Zeroth-order Tensor-Train Adaption (AdaZeta) framework aims to improve the performance and convergence of Memory-efficient Zeroth-order (MeZO) methods for fine-tuning large language models (LLMs). The AdaZeta framework introduces a fast-forward, low-parameter tensorized adapter to enhance dimension-dependent ZO estimation accuracy. Additionally, an adaptive query number schedule is proposed to tackle the frequently observed divergence issue in large-scale ZO fine-tuning tasks. Experimental results on Roberta-Large and Llama-2-7B models demonstrate the efficacy of AdaZeta in terms of accuracy, memory efficiency, and convergence speed. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper proposes a new way to train large language models using less memory. Right now, we need more and more computer memory to fine-tune these models, which is a problem. The authors suggest an improved method called Adaptive Zeroth-order Tensor-Train Adaption (AdaZeta) that works better than other methods in this area. They also introduce a new way to avoid getting stuck during training, which is important for large-scale tasks. |
Keywords
» Artificial intelligence » Fine tuning » Llama
