Summary of Rule-based Data Selection For Large Language Models, by Xiaomin Li et al.
Rule-based Data Selection for Large Language Models
by Xiaomin Li, Mingye Gao, Zhiwei Zhang, Chang Yue, Hong Hu
First submitted to arxiv on: 7 Oct 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
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 explores the impact of training data quality on large language models (LLMs) and proposes a novel framework for evaluating and selecting high-quality data. The authors critique conventional rule-based approaches, which rely heavily on human heuristics and lack effective metrics for assessing rules. Instead, they introduce an automated pipeline that generates diverse sets of rules using LLMs, rates data based on these rules, and selects the most orthogonal score vectors using determinantal point processes (DPPs). The proposed method outperforms other approaches in terms of rating precision and model performance, demonstrating its effectiveness in various scenarios, including pre-training and fine-tuning for specific domains such as IMDB, Medical, Math, and Code. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about how to get better results from large language models by using good training data. The authors are unhappy with the way people usually choose which data to use because it’s based on things humans think are important, but doesn’t really work well. Instead, they created a new way of choosing data that uses special computer algorithms to pick the best pieces of data. They tested this method and found that it worked better than other ways of choosing data, even when using big language models for tasks like movie reviews or medical research. |
Keywords
» Artificial intelligence » Fine tuning » Precision
