 This paper proposes a deep learning-based methodology for the design of high-performance solar metamaterial absorbers. It utilizes a metamaterial spectrum transformer, MST, which divides the optical spectrum into multiple patches, allowing for more accurate optimization of the metamaterials properties. Additionally, a flexible design tool was created to allow for customization of the metamaterials properties. The proposed methodology was tested using graded refractive index nanostructures, demonstrating high average absorptance of 94%, as well as exceptional efficiency compared to existing counterparts. This article was authored by Wei Chen, Yuan Gao, Yu Yangli, and others.