Summary of Quantized Side Tuning: Fast and Memory-efficient Tuning Of Quantized Large Language Models, by Zhengxin Zhang et al.
Quantized Side Tuning: Fast and Memory-Efficient Tuning of Quantized Large Language Models
by Zhengxin Zhang, Dan Zhao, Xupeng Miao, Gabriele Oliaro, Qing Li, Yong Jiang, Zhihao Jia
First submitted to arxiv on: 13 Jan 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 proposed Quantized Side Tuing (QST) approach enables memory-efficient and fast finetuning of large language models (LLMs). By quantizing model weights into 4-bit representations, reducing intermediate activations through a separate side network, and utilizing low-rank adaptors and gradient-free downsample modules to decrease optimizer states, QST achieves significant memory savings (up to 2.3x) and speed improvements (up to 3x) while maintaining competitive performance. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Quantized Side Tuing is a new way to make language models work better and use less computer memory. It does this by changing how the model stores its weights, reducing the amount of information it needs to remember, and using a separate part of the model for making predictions. This makes the process faster and uses less memory (up to 2.3 times less). The results show that QST works well and is better than other methods. |
