llama.cpp Releases Build b10873 with Multimodal Video ID Fix
llama.cpp released build b10873, which includes a fix in the multimodal (mtmd) toolset to propagate video IDs to bitmap representations (PR #28601). Although this is a routine automated release rather than a major feature launch, it ensures proper handling and identification of video inputs within llama.cpp's growing multimodal capabilities. The change directly updates the mtmd library component used for processing vision and video models. Pre-compiled binaries for Windows, Linux, macOS/iOS, and Android across CPU, CUDA, Vulkan, and OpenVINO backends are included in the release.
## BACKGROUND
llama.cpp is a popular open-source C/C++ framework for running LLMs locally across diverse hardware backends. Its mtmd (multimodal) module provides support for vision-language models, enabling inference on image and video inputs alongside text.