Update README.md
Browse files
README.md
CHANGED
|
@@ -6,12 +6,20 @@ base_model:
|
|
| 6 |
- OpenGVLab/InternVL3-8B
|
| 7 |
---
|
| 8 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 9 |
# Code2Logic: Game-Code-Driven Data Synthesis for Enhancing VLMs General Reasoning
|
| 10 |
|
| 11 |
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 solely with a GRPO strategy on **GameQA** (synthesized via our proposed **Code2Logic** approach), multiple cutting-edge open-source models exhibit significantly enhanced out-of-domain generalization.
|
| 12 |
|
| 13 |
[[📖 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) ]
|
| 14 |
|
|
|
|
|
|
|
| 15 |
## News
|
| 16 |
|
| 17 |
* We've open-sourced the ***three*** models trained with GRPO on GameQA on [Huggingface](https://huggingface.co/Code2Logic).
|
|
|
|
| 6 |
- OpenGVLab/InternVL3-8B
|
| 7 |
---
|
| 8 |
|
| 9 |
+
***This model (GameQA-InternVL3-8B) results from trainning InternVL3-8B with GRPO purely on our [GameQA](https://huggingface.co/datasets/Gabriel166/GameQA-140K) dataset.***
|
| 10 |
+
|
| 11 |
+
# Evaluation Results on General Vision BenchMarks
|
| 12 |
+
|
| 13 |
+
<div align=center><img src="https://raw.githubusercontent.com/tongjingqi/Code2Logic/refs/heads/main/assets/evaluation_results_on_general_vision_benchmarks.png" width="90%"></div>
|
| 14 |
+
|
| 15 |
# Code2Logic: Game-Code-Driven Data Synthesis for Enhancing VLMs General Reasoning
|
| 16 |
|
| 17 |
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 solely with a GRPO strategy on **GameQA** (synthesized via our proposed **Code2Logic** approach), multiple cutting-edge open-source models exhibit significantly enhanced out-of-domain generalization.
|
| 18 |
|
| 19 |
[[📖 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) ]
|
| 20 |
|
| 21 |
+
<div align=center><img src="https://raw.githubusercontent.com/tongjingqi/Code2Logic/refs/heads/main/assets/categorized_30_games_images.png" width="90%"></div>
|
| 22 |
+
|
| 23 |
## News
|
| 24 |
|
| 25 |
* We've open-sourced the ***three*** models trained with GRPO on GameQA on [Huggingface](https://huggingface.co/Code2Logic).
|