---
base_model:
- llava-hf/llava-onevision-qwen2-7b-ov-hf
datasets:
- Code2Logic/GameQA-140K
- Code2Logic/GameQA-5K
license: apache-2.0
pipeline_tag: image-text-to-text
library_name: transformers
---

***This model (GameQA-LLaVA-OV-7B) results from training LLaVA-OV-7B with GRPO solely on our [GameQA-5K](https://huggingface.co/datasets/Code2Logic/GameQA-5K) (sampled from the full [GameQA-140K](https://huggingface.co/datasets/Gabriel166/GameQA-140K) dataset).***

# Evaluation Results on General Vision BenchMarks

<div align=center><img src="https://raw.githubusercontent.com/tongjingqi/Code2Logic/refs/heads/main/assets/evaluation_results_on_general_vision_benchmarks.png"></div>

***(The inference and evaluation configurations were unified across both the original open-source models and our trained models.)***

# Code2Logic: Game-Code-Driven Data Synthesis for Enhancing VLMs General Reasoning

This is the first work, to the best of our knowledge, that leverages ***game code*** to synthesize multimodal reasoning data for ***training*** VLMs. Furthermore, when trained with a GRPO strategy solely on **GameQA** (synthesized via our proposed **Code2Logic** approach), multiple cutting-edge open-source models exhibit significantly enhanced out-of-domain generalization.

[[📖 Paper](https://arxiv.org/abs/2505.13886)] [[\ud83d\udcbb Code](https://github.com/tongjingqi/Code2Logic)] [[🤗 GameQA-140K Dataset](https://huggingface.co/datasets/Gabriel166/GameQA-140K)] [[🤗 GameQA-5K Dataset](https://huggingface.co/datasets/Code2Logic/GameQA-5K)] [[🤗 GameQA-InternVL3-8B](https://huggingface.co/Code2Logic/GameQA-InternVL3-8B) ] [[🤗 GameQA-Qwen2.5-VL-7B](https://huggingface.co/Code2Logic/GameQA-Qwen2.5-VL-7B)] [[\ud83e\udd17 GameQA-LLaVA-OV-7B](https://huggingface.co/Code2Logic/GameQA-llava-onevision-qwen2-7b-ov-hf) ]

## News

* We've open-sourced the ***three*** models trained with GRPO on GameQA on [Huggingface](https://huggingface.co/Code2Logic).

## Usage

This model is compatible with the `transformers` library. Here's how to use it for image-to-text generation:

```python
import torch
from PIL import Image
from transformers import AutoProcessor, AutoModelForCausalLM

model_id = "Code2Logic/GameQA-llava-onevision-qwen2-7b-ov-hf"

# Load processor and model
processor = AutoProcessor.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.bfloat16).to("cuda")

# Load your image (replace with an actual image path or PIL Image object)
# Example: a screenshot of a GUI for a typical use case of this model
image = Image.open("your_gui_screenshot.jpg")

# Prepare your text prompt. The model is designed for multimodal tasks,
# so typical inputs involve both an image and a text query.
prompt = "What is highlighted in the screenshot? Provide a concise description."

# Construct the chat history format required by the model
messages = [
    {"role": "user", "content": [{"type": "image"}, {"type": "text", "text": prompt}]}
]
chat_prompt = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)

# Process inputs for the model
inputs = processor(text=chat_prompt, images=image, return_tensors="pt").to(model.device)

# Generate response
with torch.no_grad():
    generated_ids = model.generate(**inputs, max_new_tokens=100) # Adjust max_new_tokens as needed
    generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]

print(generated_text)
```