The central idea
AlloyGPT represents alloy information in a language-model framework for forward prediction and inverse design. It connects composition with microstructure and properties relevant to additive manufacturing.
What the work shows
The model predicts alloy structures and properties and generates multiple compositions for a given target. CALPHAD-based calculations assess the generated candidates; accuracy decreases with distance beyond the training domain.
Scope of the result
A generated composition is a candidate. Its reliability depends on the training domain and subsequent physical and experimental validation.