Summary of Diversity Measurement and Subset Selection For Instruction Tuning Datasets, by Peiqi Wang et al.
Diversity Measurement and Subset Selection for Instruction Tuning Datasets
by Peiqi Wang, Yikang Shen, Zhen Guo, Matthew Stallone, Yoon Kim, Polina Golland, Rameswar Panda
First submitted to arxiv on: 4 Feb 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 This paper aims to improve the fine-tuning of large language models by selecting relevant data subsets for instruction following. Building upon prior work that emphasized dataset diversity, this research uses determinantal point processes to capture both diversity and quality in dataset curation. The proposed log determinant distance measure is shown to be correlated with downstream performance, enabling informed data selection and analysis of curation strategies. Experimental results demonstrate the effectiveness of this approach on various instruction tuning datasets. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps us choose the right parts of big language model training data sets to make it better at following instructions. Right now, people use rules of thumb like “use more tasks” but that’s not perfect. This research uses a special way to measure how different each part of the data set is from others and from an ideal set. It shows that this measurement is related to how well the language model does on instruction-following tasks. So, it can be used to decide when selecting certain parts of the data set will help the most and to test different ways of preparing the data. |
Keywords
* Artificial intelligence * Fine tuning * Instruction tuning * Language model
