FlyDSL Accelerates PyTorch on AMD MI350 GPUs

Published 22 October 2026

PyTorch users on AMD MI350-series GPUs can access optimized FlyDSL kernels through familiar APIs for GEMM, attention, RMSNorm, and TopK. Reported gains include up to 3.01× whole-request speedup in vLLM and substantial operator-level improvements for MXFP workloads.