Summary of What Should Baby Models Read? Exploring Sample-efficient Data Composition on Model Performance, by Hong Meng Yam and Nathan J Paek
What Should Baby Models Read? Exploring Sample-Efficient Data Composition on Model Performance
by Hong Meng Yam, Nathan J Paek
First submitted to arxiv on: 11 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 paper investigates how the quality of pre-training data affects the performance of small language models in a scenario where limited data is available. The authors test various datasets, including child-directed speech, classic books, synthetic data, and a mix of these, across different model sizes ranging from 18 million to 705 million parameters. The results show that smaller models perform better when trained on more complex and rich datasets like Gutenberg. In contrast, models trained on CHILDES and TinyStories underperform across all model sizes. The study suggests that the best dataset for training language models depends on their size, and that neither child-directed speech nor simplified stories are ideal for language models of all sizes. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This research looks at how the quality of information used to train small language models affects their performance when they only have a little data. The researchers test different types of information, such as things said to children and books from long ago, against each other. They find that smaller models do better when trained on more detailed information. On the other hand, models trained on simpler information don’t perform well at all. This means that the right kind of information to use for training language models depends on how big they are. |
Keywords
» Artificial intelligence » Synthetic data
