gemma-4-26b-a4b-int4-ov

Gemma 4 26B A4B OpenVINO INT4

This repository contains an unofficial OpenVINO™ IR conversion of google/gemma-4-26B-A4B with INT4 weight compression.

The model is intended for local OpenVINO inference with Optimum Intel. It supports vision-language generation, where the input can include both an image and a text prompt.

Original model

• Original model: google/gemma-4-26B-A4B
• Original creator: Google
• Converted format: OpenVINO IR
• Weight format: INT4
• Task: image-to-text / vision-language generation
• This repository is a converted model repository and is not an official Google or OpenVINO release.

Please refer to the original model card for full training details, intended use, safety notes, license terms, and limitations.

Usage

If you are running from inside the model repository:

python examples/run_gemma4_vlm_openvino.py \
  --model-dir . \
  --device GPU \
  --max-new-tokens 128 \
  --prompt "Describe this image in one paragraph."

If you are running from the parent directory:


```bash
cd ~/ov_models

python ov_gemma_4_26B_A4B_INT4/examples/run_gemma4_vlm_openvino.py \
  --model-dir ov_gemma_4_26B_A4B_INT4 \
  --device GPU \
  --max-new-tokens 128 \
  --prompt "Describe this image in one paragraph."

## Conversion

This model was converted with Optimum Intel using OpenVINO export.

```bash
optimum-cli export openvino \
  --model google/gemma-4-26B-A4B \
  --weight-format int4 \
  --trust-remote-code \
  ov_gemma_4_26B_A4B_INT4