llama.cpp Release b10287 Fixes Unlimited-OCR Converter Bug
The llama.cpp project has released version b10287, which introduces a bug fix for the max_tiles setting in the Unlimited-OCR converter. This release addresses issue #25614 to ensure correct configuration during model conversion. Although this is a minor release, it ensures the stability and correctness of OCR-related model conversions within the llama.cpp ecosystem. Developers utilizing Unlimited-OCR models will benefit from more reliable conversion outputs. The update specifically targets the max_tiles parameter in the converter script, preventing potential errors when processing tiled image inputs. Pre-built binaries remain available for various platforms including macOS, Windows, Linux, and Android.
## BACKGROUND
llama.cpp is a popular open-source LLM inference engine written in C/C++ designed for high performance on consumer hardware. It frequently converts external models into its native GGUF format using specialized converter scripts. OCR (Optical Character Recognition) models often process images by dividing them into smaller tiles, where the max_tiles parameter controls the maximum number of these tiles.