Summary of Zyda-2: a 5 Trillion Token High-quality Dataset, by Yury Tokpanov et al.
Zyda-2: a 5 Trillion Token High-Quality Dataset
by Yury Tokpanov, Paolo Glorioso, Quentin Anthony, Beren Millidge
First submitted to arxiv on: 9 Nov 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI)
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 technical report presents Zyda-2, a massive five trillion token dataset for language model pretraining. The Zamba2 series of models, trained on Zyda-2, achieves state-of-the-art performance in its weight class. To build Zyda-2, the authors collate high-quality open-source tokens from FineWeb and DCLM, then apply cross-deduplication and model-based quality filtering to distill them into a highest-quality subset. The dataset is released under a permissive open license and available at huggingface.co/datasets/Zyphra/Zyda-2. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Zyda-2 is a huge language dataset that helps train AI models to be better. It’s like a super-long book of words, with five trillion tokens! The authors used two other datasets, FineWeb and DCLM, to make Zyda-2 by getting rid of duplicates and keeping only the best parts. This dataset is special because it’s open-source and free for anyone to use. |
Keywords
» Artificial intelligence » Language model » Pretraining » Token
