| language: | |
| - en | |
| license: mit | |
| task_categories: | |
| - question-answering | |
| - image-text-to-text | |
| library_name: datasets | |
| tags: | |
| - game | |
| - reasoning | |
| - vlm | |
| - multimodal | |
| - synthetic-data | |
| - chain-of-thought | |
| ***In this repository, we specifically provide the 5k training samples from the complete [GameQA-140K](https://huggingface.co/datasets/Gabriel166/GameQA-140K) dataset used in our work for GRPO training of the models.*** | |
| Refer to our [paper](https://arxiv.org/abs/2505.13886) for details. And our code for training and evaluation is at https://github.com/tongjingqi/Code2Logic. | |
| # 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)] [🤗 [GameQA-140K Dataset](https://huggingface.co/datasets/Gabriel166/GameQA-140K)] [🤗 [GameQA-InternVL3-8B](https://huggingface.co/Code2Logic/GameQA-InternVL3-8B) ] [🤗 [GameQA-Qwen2.5-VL-7B](https://huggingface.co/Code2Logic/GameQA-Qwen2.5-VL-7B)] [🤗 [GameQA-LLaVA-OV-7B](https://huggingface.co/Code2Logic/GameQA-llava-onevision-qwen2-7b-ov-hf) ] | |
| <div align=center><img src="https://raw.githubusercontent.com/tongjingqi/Code2Logic/refs/heads/main/assets/categorized_30_games_images.png"></div> | |
| ## 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.)*** | |
| It's also found that getting trained on the 5k samples from our GameQA dataset can lead to better results than on [multimodal-open-r1-8k-verified](https://huggingface.co/datasets/lmms-lab/multimodal-open-r1-8k-verified). | |
| <div align=center><img src="https://raw.githubusercontent.com/tongjingqi/Code2Logic/refs/heads/main/assets/GameQA_generalizes_better.png"></div> | |
| ## News | |
| * We've open-sourced the ***three*** models trained with GRPO on GameQA on [Huggingface](https://huggingface.co/Code2Logic). |