Cohere Officially Adds llama.cpp Support for Latest Model Architectures
Cohere has officially integrated llama.cpp support for their latest model architectures, including vision encoder compatibility for North Micro Vision Instruct. This update ensures that all of Cohere's recent model lineups, alongside prior support for architectures like Command A+ and North Mini Code, can now run locally. This development directly fulfills strong demands from the local LLM community by making advanced models accessible for local deployment and hardware-efficient inference. Additionally, making Cohere Transcribe available in Unsloth and Unsloth Studio broadens fine-tuning options for state-of-the-art speech-to-text models. The newly merged vision encoder support for North Micro Vision Instruct allows multimodal capabilities to function smoothly within the llama.cpp ecosystem. Furthermore, Cohere Transcribe is now accessible on Unsloth and Unsloth Studio for fine-tuning speech-to-text tasks.
## BACKGROUND
llama.cpp is a widely adopted open-source software library that performs inference on large language models using minimal hardware resources. A vision encoder is a specialized neural network component that processes visual information from images and converts it into numerical representations that vision-language models can understand.