---

license: gemma

library_name: llama.cpp

tags:

- gguf

- medgemma

- gemma

- gemma3

- multimodal

- vision

base_model: google/medgemma-1.5-4b-it

---

MedGemma 1.5 4B IT GGUF

GGUF conversion and quantization of Google's

MedGemma 1.5 4B IT model.

Original Model

• Model: google/medgemma-1.5-4b-it
• Architecture: Gemma3ForConditionalGeneration
• Parameters: ~4B
• Original format: BF16 SafeTensors
• Modality: Text + Vision

Original model:

https://huggingface.co/google/medgemma-1.5-4b-it

Files

| File | Format | Approx. Size | Purpose |

|---|---|---:|---|

| medgemma-1.5-4b-it-f16.gguf | F16 | ~8.5 GB | Main model |

| medgemma-1.5-4b-it-Q4_K_M.gguf | Q4_K_M | ~2.4 GB | 4-bit quantized model |

| mmproj-medgemma-1.5-4b-it-f16.gguf | F16 | ~850 MB | Vision projector |

Quantization

The original model was downloaded from Hugging Face and

converted to GGUF using llama.cpp.

Conversion pipeline:

MedGemma 1.5 4B IT

|

v

BF16 SafeTensors

|

v

F16 GGUF

|

v

Q4_K_M GGUF

The original model weights are split into two SafeTensors

shards:

model-00001-of-00002.safetensors

model-00002-of-00002.safetensors

These are two parts of the same model, not two separate models.

Q4_K_M

The Q4_K_M file is a 4-bit quantized version of the F16

GGUF model.

Approximate sizes:

• F16: ~8.5 GB
• Q4_K_M: ~2.4 GB

Quantization reduces storage and memory requirements while

keeping the same underlying model architecture and parameter

count.

Multimodal / Vision

MedGemma 1.5 4B IT supports multimodal input.

For vision/image use, the main model should be used together

with:

mmproj-medgemma-1.5-4b-it-f16.gguf

The mmproj file contains the multimodal vision projector and

is separate from the main language model.

llama.cpp

These files were created using:

https://github.com/ggml-org/llama.cpp

Example Q4 quantization command:

llama-quantize \

medgemma-1.5-4b-it-f16.gguf \

medgemma-1.5-4b-it-Q4_K_M.gguf \

Q4_K_M

Model Usage

Use the appropriate llama.cpp multimodal interface for

image input and the main GGUF model for text generation.

The exact command-line options may depend on the version of

llama.cpp being used.

Disclaimer

This repository contains a GGUF conversion and quantization

of the original MedGemma model. It is not a newly trained or

fine-tuned model.

Please read the original model documentation, license, terms

of use, intended use, limitations, and safety information:

https://huggingface.co/google/medgemma-1.5-4b-it

This model should not be considered a medical device and

should not replace qualified medical professionals.

Credits

Original model:

Google — MedGemma 1.5 4B IT

GGUF conversion and quantization:

llama.cpp