Summary of Perplexed by Perplexity: Perplexity-based Data Pruning with Small Reference Models, By Zachary Ankner et al.
Perplexed by Perplexity: Perplexity-Based Data Pruning With Small Reference Models
by Zachary Ankner, Cody Blakeney, Kartik Sreenivasan, Max Marion, Matthew L. Leavitt, Mansheej Paul
First submitted to arxiv on: 30 May 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 paper investigates whether small language models can identify high-quality subsets of large text datasets that enhance the performance of larger language models. It explores whether smaller models can be used for perplexity-based pruning, which has previously been shown to yield high-quality data when using a larger model. The study demonstrates that perplexity-based pruning of pretraining data can significantly improve downstream task performance by up to 2.04 and reduce pretraining steps by up to 1.45. This improvement is achieved across multiple dataset compositions, including the over-trained and data-constrained regimes. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper looks at how small language models can help make larger language models better. They want to know if these smaller models can pick out important parts of big text datasets that will improve performance. The researchers show that using a smaller model to pick out good parts of the dataset can really help, making bigger models work up to 2% better and needing fewer training steps. |
Keywords
» Artificial intelligence » Perplexity » Pretraining » Pruning
