Summary of Less Is More: Efficient Model Merging with Binary Task Switch, by Biqing Qi et al.
Less is More: Efficient Model Merging with Binary Task Switch
by Biqing Qi, Fangyuan Li, Zhen Wang, Junqi Gao, Dong Li, Peng Ye, Bowen Zhou
First submitted to arxiv on: 24 Nov 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
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 approach to equip models with multi-task capabilities without additional training is model merging. However, existing methods face challenges such as redundant parameter conflicts and excessive storage burden. The authors reveal that for task vectors, only those parameters with magnitudes above a certain threshold contribute positively to the task, exhibiting a pulse-like characteristic. Leveraging this characteristic, they binarize the task vectors and reduce storage overhead. The binarized task vectors incur almost no decrease in fine-tuning and merging performance, and even exhibit stronger performance improvements as the proportion of redundant parameters increases. The authors propose Task Switch (T-Switch), which decomposes task vectors into three components: activation switch, polarity switch, and scaling knob. By storing task vectors in a binarized form, T-Switch alleviates parameter conflicts while ensuring efficient task parameter storage. Additionally, they introduce Auto-Switch, which enables training-free switch combination via retrieval from a small query set. The proposed methods achieve significant performance improvements over existing baselines, requiring only 1-3% of the storage space of full-precision parameters. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper proposes a new way to help models learn multiple tasks without needing more training data. They show that by simplifying and compressing task vectors, they can reduce the amount of memory needed while still getting good results. This is achieved through a method called Task Switch (T-Switch), which breaks down task vectors into three parts: an activation switch, a polarity switch, and a scaling knob. By storing these vectors in a simplified form, T-Switch helps to avoid conflicts between different tasks while also saving memory. |
Keywords
» Artificial intelligence » Fine tuning » Multi task » Precision
