[rafram]

Ollama is not a wrapper around llama.cpp anymore, at least for multimodal models (not sure about others). They have their own engine: https://ollama.com/blog/multimodal-models

[iphone_elegance]

looks like the backend is ggml, am I missing something? same diff