Eval bug: Error on multimodal processing of Gemma 3 270M

[matbgn]

Name and Version

$ llama-server --version
version: 6139 (f4586ee)
built with cc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0 for x86_64-linux-gnu

Operating systems

Linux

GGML backends

Vulkan

Hardware

NVIDIA RTX 3060

Models

ggml-org/gemma-3-270m-it-GGUF

Problem description & steps to reproduce

Run server:

llama-server -hf ggml-org/gemma-3-270m-it-GGUF --api-key dummy

Query it:

llm -m llama-server-vision 'describe this image' -a https://static.simonwillison.net/static/2025/pelican-group.jpg

First Bad Commit

Potentially #15319

Relevant log output

main: server is listening on http://127.0.0.1:8080 - starting the main loop
srv  update_slots: all slots are idle
got exception: {"code":500,"message":"image input is not supported - hint: if this is unexpected, you may need to provide the mmproj","type":"server_error"}
srv  log_server_r: request: POST /v1/chat/completions 127.0.0.1 500
got exception: {"code":500,"message":"image input is not supported - hint: if this is unexpected, you may need to provide the mmproj","type":"server_error"}
srv  log_server_r: request: POST /v1/chat/completions 127.0.0.1 500
got exception: {"code":500,"message":"image input is not supported - hint: if this is unexpected, you may need to provide the mmproj","type":"server_error"}
srv  log_server_r: request: POST /v1/chat/completions 127.0.0.1 500

[matbgn]

Side note: Of course no error with llama-server -hf ggml-org/gemma-3-4b-it-GGUF --api-key dummy

[CISC]

The error tells you what you need to do.

Since you seem to be using some kind of middleware I can't tell you how to achieve it. Oh, nvm, I see. :)

gemma-3-270m-it is text-only, the model card is slightly confusing I guess. In the case of a multimodal model you would have to download the mmproj-* file separately and load it with --mmproj parameter.

[matbgn]

My bad, I'm very confused and sorry to have bother your time. 😥

Is the gemma-3-270m model multimodal capable?

No. gemma-3-270m is text-only—it doesn’t accept image inputs. Multimodal (image+text) support starts at the 4B, 12B, and 27B sizes; the 270M (and 1B) variants have a 32K context and handle text only.