Uncensored MultiModal (ShareGPT4V-13B_GGUF)

When dealing with multimodal models, the term uncensored means something a little different than it does with your typical LLM.

With something like llama 3, “uncensored” means that you can have conversations that would typically be considered inappropriate for public consumption. However when dealing with an MMPROJ, What it means is that it will discuss images or scenes that other models will not. This does not mean it will give “pervy” replies but rather it will describe things in a professional manner.

In other words, an Uncensored Multimodal simply means it’s willing to have the conversation

Introducing ShareGPT4v – https://huggingface.co/nakodanei/ShareGPT4V-13B_GGU

This is not the greatest model in the world (check out my video on BUNNY) however it is the most consistent.

Here’s how to run it:

llama-llava-cli -m ".\path\to\ShareGPT4V-13B-Q4_K_M.gguf" -t 6 -c 4096 -ngl 26 --mmproj ".\path\to\mmproj-model-f16.gguf"  --image "D:\path\to\image.jpg"