Summary of Adam-mini: Use Fewer Learning Rates to Gain More, by Yushun Zhang et al.
Adam-mini: Use Fewer Learning Rates To Gain More
by Yushun Zhang, Congliang Chen, Ziniu Li, Tian Ding, Chenwei Wu, Diederik P. Kingma, Yinyu Ye, Zhi-Quan Luo, Ruoyu Sun
First submitted to arxiv on: 24 Jun 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 Adam-mini optimizer achieves comparable or better performance to AdamW with a significantly reduced memory footprint. By investigating the Hessian structure of neural networks, researchers discovered that many learning rates in Adam’s v parameter could be safely removed by carefully partitioning parameters into blocks and assigning a single effective learning rate to each block. This led to the development of Adam-mini, which uses a simple method to find good learning rates. Experimental results show that Adam-mini performs on par or better than AdamW on various language models for pre-training, supervised fine-tuning, and reinforcement learning with human feedback. The reduced memory footprint also reduces communication overheads among GPUs, increasing throughput. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Adam-mini is a new way to make neural networks work better. Researchers found that most of the information in Adam’s v parameter wasn’t being used effectively. They came up with a new idea to group parameters into blocks and give each block its own learning rate. This helps Adam-mini use less memory, which makes it faster and more efficient. The results show that Adam-mini works just as well or even better than another popular optimizer, AdamW, on different language models. |
Keywords
» Artificial intelligence » Fine tuning » Reinforcement learning » Supervised
