llama.cpp's b10354 release fixes Android CPU affinity mask bug
This update corrects ignored CPU affinity masks on Android—boosting on-device LLM performance.
Deep Dive
llama.cpp release b10354 fixes the ggml-cpu issue where the CPU affinity mask was being ignored on Android. The release includes prebuilt binaries for macOS, Linux, Android, Windows, openEuler, and UI assets, and was published by github-actions with GitHub’s verified signature.
Key Points
- Fixes CPU affinity mask being ignored on Android (issue #26838)
- Includes prebuilt binaries for Windows CUDA 12/13, Linux Vulkan, and macOS Apple Silicon
- llama.cpp holds 123k stars and 21.5k forks on GitHub
Why It Matters
Android developers running local LLMs can now trust CPU core assignments, delivering smoother, more predictable inference.