ggml-org releases llama.cpp b12413 with image token fix
llama.cpp release b12413 introduces a bug fix for the placement of begin and end of image tokens in chat templates, along with standard cross-platform binary distributions. This update ensures correct multimodal token handling in chat interfaces, which is essential for maintaining accurate context and processing capabilities in vision-language models. The patch specifically addresses inkling 2 image token placement and includes corresponding test cases to prevent future regressions.
## BACKGROUND
llama.cpp is an open-source software library that performs inference on various large language models using the GGML tensor library. It serves as a core infrastructure component for local AI execution across diverse hardware backends like CPU, CUDA, and Vulkan.