This paper introduces aLLoyM, a large-scale language model (LLM) applicable to materials science. aLLoyM is an LLM specifically tuned for alloy composition, temperature, and corresponding phase information. It was developed by curating question-and-answer (Q&A) pairs for binary and ternary phase diagrams based on the open-source Computational Phase Diagram Database (CPDDB) and the Calculation of PHAse Diagrams (CALPHAD). We fine-tuned Mistral, an open-source pre-trained LLM, in two Q&A formats: multiple-choice and short-answer. Benchmark evaluation results demonstrate that fine-tuning significantly improves performance on multiple-choice phase diagram questions. Furthermore, aLLoyM's short-answer model demonstrates its ability to generate novel phase diagrams based solely on constituent elements, highlighting its potential to accelerate the discovery of previously unexplored material systems. To encourage further research and adoption, we have released the short-answer fine-tuned version of aLLoyM and the complete benchmarking Q&A dataset on Hugging Face.
