llama.cpp b9387 improves AMD/ROCm prompt processing

View on Reddit

llama.cpp b9387 focuses on prompt processing improvements for AMD GPUs, leveraging MFMA instructions on CDNA architecture (MI100, MI200, MI300 series). Users are encouraged to test and share performance results.