Reconfigurable Computing Challenge: Transformer for Jet Tagging on Versal AI Engines
Researchers have developed a novel framework for deploying transformer models on the AMD Versal AI Engine, specifically for jet tagging applications at the CERN LHC. This framework quantifies transformer layers into an integer-only format, making them suitable for low-latency, resource-constrained trigger systems. The reusable software allows for the automatic generation of Vitis graph code from high-level Python descriptions, offering a foundation for future research in this area. AI