Summary of Orbit: Cost-effective Dataset Curation For Large Language Model Domain Adaptation with An Astronomy Case Study, by Eric Modesitt et al.
ORBIT: Cost-Effective Dataset Curation for Large Language Model Domain Adaptation with an Astronomy Case Study
by Eric Modesitt, Ke Yang, Spencer Hulsey, Chengxiang Zhai, Volodymyr Kindratenko
First submitted to arxiv on: 19 Dec 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 methodology, ORBIT, is a cost-efficient approach to curating high-quality domain-specific datasets from noisy web sources for training specialist large language models. This paper presents a case study in astronomy, refining the 1.3T-token FineWeb-Edu dataset into a high-quality, 10B-token subset focused on astronomy. The fine-tuning of LLaMA-3-8B on an astronomy subset improves performance on the MMLU astronomy benchmark from 69% to 76%, and achieves top results on AstroBench. Additionally, ORBIT outperforms LLaMA-3-8B-base in GPT-4o evaluations across 1000 astronomy-specific questions. The paper also validates ORBIT’s generalizability by applying it to law and medicine. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary ORBIT is a new way to make language models better for specific tasks. Right now, most language models are good at general tasks, but they can struggle with very specialized topics. To fix this, the authors created ORBIT, which helps curate large amounts of high-quality data from the internet. They tested it in astronomy and saw big improvements. The model even outperformed other similar models on some tests! This is important because language models need to be good at many different things. |
Keywords
» Artificial intelligence » Fine tuning » Gpt » Llama » Token
