TL;DR: llama.cpp's latest release, b10015, includes a fix for OpenCL 2.x compatibility, improving performance and stability on certain hardware.
Summary: The ggerganov/llama.cpp project released version b10015, which specifically addresses an issue where clCreateBufferWithProperties was incorrectly used when targeting OpenCL 2.x. This fix, detailed in pull request #25673, prevents potential compatibility problems and optimizes OpenCL operations.
Why it matters: This update is crucial for developers leveraging llama.cpp on systems with OpenCL 2.x compatible GPUs, potentially leading to more stable and efficient local LLM inference. Builders should update to b10015 to benefit from improved hardware utilization and reduced errors on OpenCL-enabled devices.
Source: github_releases