llama.cpp Release b10885 Fixes Parameter Handling for IBM Granite Models
llama.cpp release b10885 integrates pull request #28643 to correct parameter count handling across the IBM Granite model family. The patch also adds missing active expert prefixes and cleans up expert block code alignment. This fix ensures accurate parameter accounting and proper execution for users running IBM Granite foundation models locally. It highlights direct collaboration between enterprise AI developers at IBM and open-source inference engine maintainers. Submitted by IBM engineer Aaron Teo, the update adds a missing 'A' prefix for active experts, corrects include directives, and removes an unused 40-block case. Pre-built binaries are provided for Windows, Linux, macOS, iOS, and Android across CPU, CUDA, Vulkan, and SYCL targets.
## BACKGROUND
llama.cpp is a high-performance open-source C/C++ library designed for running large language models locally on consumer hardware. IBM Granite is a family of open AI foundation models developed by IBM for enterprise applications including language and coding tasks.