mradermacher

/

AutoGLM-Phone-9B-Multilingual-GGUF

like 3

I’m a bit confused about the multimodal setup.

The model’s chat template includes image tokens
(<|begin_of_image|>, <|image|>, <|end_of_image|>),
but there is no mmproj / vision projector in the repo,
and llama.cpp throws “image input is not supported”.

After inspecting the GGUF metadata, I don’t see any vision encoder
or mmproj tensors—only text model weights.

Can you confirm:
• Is this model intended to be text-only despite image tokens?
• Should image tokens simply be ignored?
• Or is there a separate vision/projector repo that’s required?

Thanks for clarification.