Summary of Hift: a Hierarchical Full Parameter Fine-tuning Strategy, by Yongkang Liu et al.
HiFT: A Hierarchical Full Parameter Fine-Tuning Strategy
by Yongkang Liu, Yiqun Zhang, Qian Li, Tong Liu, Shi Feng, Daling Wang, Yifei Zhang, Hinrich Schütze
First submitted to arxiv on: 26 Jan 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Computation and Language (cs.CL)
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 HiFT (Hierarchical Fine-Tuning) strategy presents a novel approach to adapt language models to downstream tasks while conserving GPU memory. By only updating a subset of parameters at each training step, HiFT reduces the amount of gradients and optimizer state parameters residing in GPU memory, making it an attractive solution for large-scale language models. The results demonstrate that HiFT achieves comparable performance to parameter-efficient fine-tuning and standard full-parameter fine-tuning, while supporting various optimizers such as AdamW, AdaGrad, and SGD. Moreover, HiFT can save more than 60% GPU memory compared to standard full-parameter fine-tuning for a 7B model. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary HiFT is a new way to make language models work better on smaller computers. Right now, we need a lot of computer power to teach these models how to do new things. HiFT helps by only changing some parts of the model at a time, which takes less computer power. This means we can use bigger models that are more powerful and accurate, without needing as many powerful computers. |
Keywords
* Artificial intelligence * Fine tuning * Parameter efficient
