OpenDFM/CharTool-7B
OpenDFM/CharTool-7B is a 7 billion parameter, tool-integrated multimodal agent developed by OpenDFM, built upon the Qwen2.5-VL-7B-Instruct base model. It is specifically designed for fine-grained chart perception and accurate numerical reasoning, leveraging image cropping for localized visual analysis and Python code execution for computation. This model excels at complex chart understanding tasks by integrating external tools for enhanced analytical capabilities.
Loading preview...
CharTool-7B: Tool-Integrated Multimodal Agent for Chart Understanding
CharTool-7B is a 7 billion parameter multimodal agent developed by OpenDFM, specifically engineered for advanced chart perception and precise numerical reasoning. Built on the Qwen2.5-VL-7B-Instruct base model, it integrates external tools to enhance its analytical capabilities, making it distinct from general-purpose multimodal models.
Key Capabilities
- Fine-grained Chart Perception: Excels at detailed visual analysis of charts, understanding intricate elements.
- Accurate Numerical Reasoning: Performs precise calculations and data interpretation using integrated tools.
- Tool-Integrated Architecture: Utilizes image cropping for localized visual perception and Python code execution for complex numerical computation.
- Agentic Reinforcement Learning: Trained with a cold-start supervised fine-tuning followed by agentic reinforcement learning on the DuoChart dataset, optimizing its performance in tool-integrated environments.
Good for
- Applications requiring deep understanding and analysis of various chart types.
- Tasks involving numerical extraction and computation from visual data.
- Developing intelligent agents that interact with and interpret graphical information.
- Research and development in multimodal AI, particularly for tool-augmented reasoning.