llama.cpp version b9745 introduces support for Step3.5/3.7 flash MTP3, including new APIs for layer offset and nextn flags. The release provides prebuilt binaries for macOS, Linux, Android, Windows, and openEuler, with options for CPU, Vulkan, CUDA, OpenVINO, and SYCL acceleration.
llama.cpp release b9745 adds MTP3 support and cross-platform binaries
from English