You can deploy this model and start earning money today!

billylo/
medgemma-1.5-4b-it-mmproj-gguf

Name: medgemma-1.5-4b-it-mmproj-gguf
Availability: InStock

google/medgemma-1.5-4b-it

quantized

apache-2.0

Views

No views yet

MedGemma 1.5 4B – Vision encoder (mmproj) GGUF

Vision encoder (mmproj) for MedGemma 1.5 4B, converted to GGUF for use with llama.cpp and llama-mtmd-cli. Use this file together with a text GGUF of MedGemma 1.5 4B to run multimodal (image + text) inference.

Source model: google/medgemma-1.5-4b-it (Health AI Developer Foundation terms apply for the base model).
Conversion: Extracted with convert_hf_to_gguf.py --mmproj from the Hugging Face model; no training or fine-tuning.

Requirements

A text GGUF of MedGemma 1.5 4B, e.g. mradermacher/medgemma-1.5-4b-it-GGUF, or your own conversion.
llama.cpp built with multimodal support, or the llama-mtmd-cli binary (e.g. brew install llama.cpp on macOS).

Download

huggingface-cli download billylo/medgemma-1.5-4b-it-mmproj-gguf mmproj-medgemma-1.5-4b-it-BF16.gguf --local-dir ./mmproj

Usage with llama-mtmd-cli

bash

llama-mtmd-cli -m /path/to/medgemma-1.5-4b-it-text.gguf \
  --mmproj mmproj-medgemma-1.5-4b-it-BF16.gguf \
  --image page.png \
  -p "Describe the findings in this medical report."

For apps that expect a single mmproj file next to the text model (e.g. named mmproj.gguf), copy or symlink this file accordingly.

Intended use

Same as the base MedGemma model: research and education in medical/health AI. Not for direct clinical decision-making. See the MedGemma model card and Google Health AI documentation for limitations and terms.

Citation

If you use MedGemma, cite the MedGemma paper and the llama.cpp project as appropriate.