Summary of Deltadq: Ultra-high Delta Compression For Fine-tuned Llms Via Group-wise Dropout and Separate Quantization, by Yanfeng Jiang et al.
DeltaDQ: Ultra-High Delta Compression for Fine-Tuned LLMs via Group-wise Dropout and Separate Quantization
by Yanfeng Jiang, Zelan Yang, Bohua Chen, Shen Li, Yong Li, Tao Li
First submitted to arxiv on: 11 Oct 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 In this paper, researchers propose a novel framework called DeltaDQ to compress the delta weight of large language models, addressing the challenge of deploying multiple fine-tuned models. The proposed method, which combines Group-wise Dropout and Separate Quantization, achieves ultra-high compression while maintaining accuracy. Experimental results show that DeltaDQ outperforms baselines for WizardMath and WizardCoder models across different parameter scales, with compression ratios reaching up to 128x for the WizardMath-7B model and 512x for the WizardMath-70B model. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large language models can perform well on many tasks when fine-tuned. However, it’s hard to use multiple full models because they’re too big. Some methods try to make them smaller by only saving the changes made from one task to another. But this doesn’t help much and still makes deployment tricky. The researchers in this paper found a way to compress these “delta weights” really well using two new techniques: Group-wise Dropout and Separate Quantization. They tested their method on some big language models and showed it works better than others. |
Keywords
» Artificial intelligence » Dropout » Quantization
