AstroLLaMA is a 7B parameter AI model fine-tuned from LLaMA-2 using over 300K astronomy abstracts, outperforming other models like GPT-4 and PaLM in astronomy-specific tasks.
Despite its impressive performance, AstroLLaMA has limitations such as lack of knowledge in specific areas of astronomy, and researchers are working on enhancing its training dataset.