llama.cpp b9658 adds debug fix for parsing errors across platforms
New release improves error messages when parsing prompts on local LLMs
The open-source llama.cpp project has tagged version b9658, a maintenance release that focuses on improving developer experience when running local LLMs. The primary change is in the chat module: when a prompt fails to parse, the full unparsed prompt is now included in the debug output. This small but significant improvement helps users and contributors quickly spot syntax errors or formatting issues in their input text, reducing time spent debugging inference pipelines.
Alongside this fix, the release updates precompiled binaries across all major platforms, including macOS (Apple Silicon with KleidiAI disabled, Intel x64), Linux (CPU, Vulkan, ROCm, OpenVINO, SYCL), Windows (CPU, CUDA, Vulkan, HIP), and Android arm64. The release notes also show ongoing support for s390x and openEuler builds, though some are disabled for now. For developers relying on local inference, b9658 brings a modest but welcome quality-of-life enhancement without breaking existing functionality.
- Chat module now includes full unparsed prompt in debug messages on parse errors
- Prebuilt binaries available for macOS Apple Silicon, Intel, Linux, Windows, Android, and more
- KleidiAI acceleration remains disabled on Apple Silicon in this release
Why It Matters
Simplifies debugging for developers running local LLMs via llama.cpp, improving reliability of custom prompt workflows.