Tag
Build 9254 of llama.cpp fixes a token generation regression and adds Programmatic Dependent Launch (PDL) support for NVIDIA GPUs, yielding up to 10% speedup in token generation on newer hardware.