AstroLLaMA: Towards Specialized Foundation Models in Astronomy

Large language models often excel in many human-language tasks but tend to falter in highly specialized domains like scholarly astronomy. To bridge this gap, we introduce AstroLLaMA, a 7-billion-parameter model fine-tuned from LLaMA-2 using over 300,000 astronomy abstracts from arXiv. Optimized for traditional causal language modeling, AstroLLaMA shows marked domain adaptation by achieving a 30% lower perplexity than LLaMA- 2. Compared to state-of-the-art foundation models, AstroLLaMA generates more insightful and scientifically relevant text completions and embedding extraction despite having significantly fewer parameters. AstroLLaMA serves as a highly domain-specific model with broad fine-tuning potential: Its public release aims to spur astronomy-focused research, including automatic paper summarization, conversational agent development and hypothesis generation.