llama.cpp b10195 release streamlines tests and adds build fixes
The popular LLM inference engine's latest release focuses on test speed and quant type selection.
ggml-org has released llama.cpp b10195, the latest version of the popular open-source C/C++ inference engine for large language models. This update focuses on internal quality and developer experience. Key changes include refactoring tests to avoid building get-model.cpp multiple times, removing the get-model.cpp test entirely, and fixing automatic quant type selection. These improvements reduce CI build times and make the testing pipeline cleaner, which is crucial as the project evolves rapidly with 122k GitHub stars and 21.2k forks.
For end users, b10195 delivers the usual comprehensive set of prebuilt binaries across platforms. Available builds span macOS (Apple Silicon and Intel), iOS as an XCFramework, Linux (x64, arm64, s390x) with support for Vulkan, ROCm 7.2, OpenVINO, and SYCL, plus Android arm64 CPU builds. Windows users get CPU-only, CUDA 12, CUDA 13, Vulkan, OpenVINO, SYCL, and HIP variants. The release also highlights llama.app, the companion Mac and iOS application, tying into the project's growing ecosystem. While this is an incremental release, it underscores llama.cpp's commitment to maintaining a lean, efficient codebase while broadening hardware support.
- Release b10195 from ggml-org optimizes test builds to avoid recompiling get-model.cpp multiple times
- Fixes quant type selection in tests, improving inference correctness across backends
- Ships prebuilt binaries for macOS, Linux, Android, and Windows, including CUDA, Vulkan, ROCm, and OpenVINO variants
Why It Matters
llama.cpp remains the go-to lightweight LLM runtime; this update keeps builds reliable across increasingly diverse hardware.