InfiX-ai/InfiGUI-R1-3B
InfiX-ai/InfiGUI-R1-3B is a 3 billion parameter multimodal model developed by Reallm-Labs, based on Qwen2.5-VL-3B-Instruct, with a 32768 token context length. It is specifically trained using the Actor2Reasoner framework and reinforcement learning to enhance its planning and reflection capabilities for Graphical User Interface (GUI) tasks. This model excels at GUI grounding and trajectory prediction, enabling automated interaction with mobile devices.
Loading preview...
Overview
InfiX-ai/InfiGUI-R1-3B is a 3 billion parameter multimodal model developed by Reallm-Labs, building upon the Qwen2.5-VL-3B-Instruct architecture. It is specifically designed for Graphical User Interface (GUI) interaction tasks, leveraging a 32768 token context length. The model's core innovation lies in its training with the Actor2Reasoner framework, which incorporates reinforcement learning to significantly improve its planning and reflection abilities for complex GUI operations.
Key Capabilities
- GUI Grounding: Accurately identifies and locates UI elements relevant to a given instruction on a screen.
- Trajectory Prediction: Generates a sequence of actions (e.g., clicks, swipes, typing) to achieve a specific GUI task.
- Reinforcement Learning Enhanced: Utilizes an Actor2Reasoner framework to refine planning and reflective reasoning for robust GUI automation.
- Multimodal Input: Processes both visual (screenshots) and textual (instructions) information to understand and interact with GUIs.
Good for
- Automated Mobile Device Interaction: Ideal for developing agents that can interact with mobile applications by performing actions like clicking, typing, and swiping.
- GUI Task Automation: Suitable for automating repetitive or complex tasks within graphical user interfaces.
- Research in GUI Agents: Provides a strong foundation for further research into multimodal GUI agents with advanced reasoning capabilities.