Summary of Mars: Unleashing the Power Of Variance Reduction For Training Large Models, by Huizhuo Yuan et al.
MARS: Unleashing the Power of Variance Reduction for Training Large Models
by Huizhuo Yuan, Yifeng Liu, Shuang Wu, Xun Zhou, Quanquan Gu
First submitted to arxiv on: 15 Nov 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Optimization and Control (math.OC); Machine Learning (stat.ML)
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 This paper proposes a unified optimization framework called MARS (Make vAriance Reduction Shine) to efficiently train large models. The framework combines preconditioned gradient methods with variance reduction using a scaled stochastic recursive momentum technique. Three instances of MARS are introduced, leveraging AdamW, Lion, and Shampoo updates respectively. Experimental results on GPT-2 models show that MARS outperforms AdamW by a significant margin. The proposed algorithms and connections to existing optimizers are also discussed. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps large language models learn faster and better. It creates a new way of training called MARS, which uses a combination of old ideas in a new way. This makes it possible for the model to learn more efficiently. The researchers tested this new method on a popular type of model called GPT-2 and found that it worked much better than an existing method called AdamW. |
Keywords
» Artificial intelligence » Gpt » Optimization
