---
pipeline_tag: image-text-to-text
license: apache-2.0
base_model: Qwen/Qwen2-VL-2B
library_name: zeromodels
language:
- en
tags:
- keras
- zeromodels
- qwen2_vl
- qwen2-vl
- multimodal
- vision
- image-text-to-text
- pytorch
- jax
- tf
---

# Run Qwen2-VL with Keras 3: JAX, PyTorch, or TensorFlow

[![GitHub](https://img.shields.io/badge/GitHub-ZeroModels-181717?logo=github)](https://github.com/IMvision12/ZeroModels) [![Docs](https://img.shields.io/badge/Docs-Qwen2--VL-1f6feb)](https://imvision12.github.io/ZeroModels/qwen2_vl/) [![HuggingFace](https://img.shields.io/badge/HuggingFace-Qwen2--VL-ffd21e?logo=huggingface&logoColor=black)](https://huggingface.co/collections/zeromodels/qwen2-vl-6a8eae2d09e489fe58f3910c)

# zeromodels/qwen2-vl-2b

Pure-**Keras 3** conversion of [`Qwen/Qwen2-VL-2B`](https://huggingface.co/Qwen/Qwen2-VL-2B) for [zeromodels](https://github.com/IMvision12/ZeroModels). One implementation runs unmodified on **TensorFlow / Torch / JAX**. This is the **2B** variant, served here as **image + text -> text** via `Qwen2VLProcessor`; weights are stored in **bfloat16**.

For model details, license, and usage terms, see the upstream [model card](https://huggingface.co/Qwen/Qwen2-VL-2B).

Paper: [Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution (arXiv:2409.12191)](https://arxiv.org/abs/2409.12191) · [HF Papers](https://huggingface.co/papers/2409.12191)

Paper: [Qwen-VL: A Frontier Large Vision-Language Model with Versatile Abilities (arXiv:2308.12966)](https://arxiv.org/abs/2308.12966) · [HF Papers](https://huggingface.co/papers/2308.12966)

## ✨ Quick start

### Text-only

```python
import os
os.environ["KERAS_BACKEND"] = "torch"  # or "jax" / "tensorflow"

from zeromodels.models.qwen2_vl import Qwen2VLTextGenerate, Qwen2VLProcessor

model = Qwen2VLTextGenerate.from_weights("zeromodels/qwen2-vl-2b")
processor = Qwen2VLProcessor.from_weights("zeromodels/qwen2-vl-2b")

inputs = processor(conversation=[
    {"role": "user", "content": [{"type": "text", "text": "Hello, who are you?"}]}
])
outputs = model.generate(**inputs, max_new_tokens=64)
print(processor.decode(outputs[0]))
```

### Image + text

```python
import os
os.environ["KERAS_BACKEND"] = "torch"  # or "jax" / "tensorflow"

from PIL import Image
from zeromodels.models.qwen2_vl import Qwen2VLConditionalGenerate, Qwen2VLProcessor

model = Qwen2VLConditionalGenerate.from_weights("zeromodels/qwen2-vl-2b")
processor = Qwen2VLProcessor.from_weights("zeromodels/qwen2-vl-2b")

inputs = processor(conversation=[
    {"role": "user", "content": [
        {"type": "image", "image": Image.open("photo.jpg")},
        {"type": "text", "text": "Describe this image in one sentence."},
    ]}
])
outputs = model.generate(**inputs, max_new_tokens=64)
print(processor.decode(outputs[0]))
```

Load any Qwen2-VL variant the same way with `from_weights("zeromodels/<variant>")`:

| Variant | Hub |
| --- | --- |
| `qwen2-vl-2b` | [zeromodels/qwen2-vl-2b](https://huggingface.co/zeromodels/qwen2-vl-2b) |
| `qwen2-vl-2b-instruct` | [zeromodels/qwen2-vl-2b-instruct](https://huggingface.co/zeromodels/qwen2-vl-2b-instruct) |
| `qwen2-vl-7b` | [zeromodels/qwen2-vl-7b](https://huggingface.co/zeromodels/qwen2-vl-7b) |
| `qwen2-vl-7b-instruct` | [zeromodels/qwen2-vl-7b-instruct](https://huggingface.co/zeromodels/qwen2-vl-7b-instruct) |
| `qwen2-vl-72b` | [zeromodels/qwen2-vl-72b](https://huggingface.co/zeromodels/qwen2-vl-72b) |
| `qwen2-vl-72b-instruct` | [zeromodels/qwen2-vl-72b-instruct](https://huggingface.co/zeromodels/qwen2-vl-72b-instruct) |

## Special Thanks

A huge thank you to the Qwen team at Alibaba for creating and releasing these models.

License: Apache 2.0.