llama.cpp Releases Build b10150 with Backend Offloading Adjustments
llama.cpp has released build b10150, which introduces adjustments to the logic for offloading operations to the weight's backend in the ggml library. Additionally, this release includes fixes for dsv4 graphs and disables KleidiAI support for macOS Apple Silicon builds. While this is a routine incremental update, optimizing backend offloading logic helps improve hardware resource utilization during local LLM inference. Ensuring stability in computation graphs like dsv4 prevents errors when running compatible models on diverse hardware configurations. The release includes pre-built binaries for a wide range of platforms, including Windows (with CUDA 12/13, Vulkan, and SYCL), Linux, Android, and macOS. However, the macOS Apple Silicon build with KleidiAI enabled has been disabled in this release.
## BACKGROUND
llama.cpp is a popular open-source, high-performance inference engine written in C/C++ that allows users to run Large Language Models (LLMs) locally. It relies on the ggml tensor library to manage machine learning operations and optimize execution across various hardware backends like CPUs and GPUs.