Summary of Harnessing Diversity For Important Data Selection in Pretraining Large Language Models, by Chi Zhang et al.
Harnessing Diversity for Important Data Selection in Pretraining Large Language Models
by Chi Zhang, Huaping Zhong, Kuan Zhang, Chengliang Chai, Rui Wang, Xinlin Zhuang, Tianyi Bai, Jiantao Qiu, Lei Cao, Ju Fan, Ye Yuan, Guoren Wang, Conghui He
First submitted to arxiv on: 25 Sep 2024
Categories
- Main: Artificial Intelligence (cs.AI)
- 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 Medium Difficulty Summary: The paper introduces Quad, a data selection approach for pre-training large language models. Quad considers both the quality and diversity of data instances, unlike existing methods that focus solely on influence scores. To achieve this, the authors utilize accelerated computations for attention layers to evaluate influence, then cluster datasets into similar and diverse instances. The Multi-Armed Bandit method is used to select clusters based on their influential instances or frequency of selection, balancing quality and diversity. Quad achieves state-of-the-art pre-training results, addressing limitations in existing methods such as computation time and non-diverse selected data. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Low Difficulty Summary: This paper helps improve how we train language models by choosing the right data to use. Currently, researchers are trying different ways to measure which data points are most important for training a model. The problem is that these methods can be slow and don’t always choose diverse enough data. The authors introduce Quad, an approach that considers both the quality and diversity of data. Quad uses a special way to compute how important each piece of data is, then groups similar data together and selects from different groups to ensure variety. This leads to better pre-trained models and solves some of the existing problems. |
Keywords
» Artificial intelligence » Attention
