#askleo How can I use Vision LLMs while blocking their Vision capabilities, and should I expect the model to run faster that way?

For context I'm using Koboldcpp which runs models through Llamacpp.