Summary of Scaling Parameter-constrained Language Models with Quality Data, by Ernie Chang et al.
Scaling Parameter-Constrained Language Models with Quality Data
by Ernie Chang, Matteo Paltenghi, Yang Li, Pin-Jie Lin, Changsheng Zhao, Patrick Huber, Zechun Liu, Rastislav Rabatin, Yangyang Shi, Vikas Chandra
First submitted to arxiv on: 4 Oct 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 The proposed paper extends the conventional understanding of scaling laws in language modeling by considering the impact of data quality on model generalization. The authors formulate a new term, effective training tokens, which combines measures of text diversity and syntheticity. They estimate constants that relate this term to model size, training tokens, and accuracy scores for eight reasoning tasks. The results show high correlation with true accuracies and highlight the importance of considering data quality in optimizing language models. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper studies how good a language model is when it’s trained on different types of text data. The researchers want to know what makes some texts better than others for training these models. They came up with a new way to measure the quality of this text, combining two factors: how unique each piece of text is and how much it looks like human-written text. By studying over 200 different language models trained on synthetic data, they found that their new metric is closely related to how well the models can answer questions. |
Keywords
» Artificial intelligence » Generalization » Language model » Scaling laws » Synthetic data
