How should I use this as a visual model with Ollama?
When I download one of the GGUF files and create a Modelfile with a FROM line pointing to it, I can import this model into Ollama with the create command, and it works with text prompts. However, when I try to pass an image, by typing something like "Describe this image. ./image.png", it doesn't send the image to the model. I tested the standard Gemma 3 12b model, and it accepts and describes images. I'm running Ollama 0.6.2, compiled from git using the ollama-git AUR package. I also tried showing the gemma3:12b Modelfile and then copying the template from it but changing the FROM line to the downloaded GGUF, but that didn't work either.
When I tried to send an image using the API with curl, it says "this model is missing data required for image input".
I don't know if you've solved your problem. Let me tell you about the situation I encountered. I'm also trying to run it in ollama, but ollama doesn't seem to support the import of external visual models and can only use the ones that come with it.So I tried llama.cpp. When you deploy it on the llama.cpp, you will be prompted that you need the mmproj file (full name: mmproj-model-f16-12B.gguf). The file needs to be downloaded from Google's official GGUF file publishing address on Huggingface, and preferably 12b parameters correspond to 12b parameters.After that, you just need to properly configure and run the main Gemma3 and mmproj files on the llama.cpp. Hope it helps those who see it. My English is very poor, please forgive me for using machine translation.