https://hgpu.org/?p=26834
SOL: Reducing the Maintenance Overhead for Integrating Hardware Support into AI Frameworks