zai-org/androidgen-llama-3-70b
AndroidGen-Llama-3-70B is a 70 billion parameter Llama-3 based model developed by Zhipu AI. It is the open-source version of AndroidGen, designed to enable large language model-based agents to autonomously perform tasks across various Android applications. This model excels at interacting with Android apps like messaging, clock, and email without requiring manually labeled interaction data, making it specialized for mobile automation and agentic behavior on the Android platform.
Loading preview...
Overview
AndroidGen-Llama-3-70B is an open-source model from Zhipu AI, built upon the Llama-3-70B architecture. It represents the publicly available version of AndroidGen, a system designed to empower large language model (LLM) agents to operate autonomously within diverse Android applications.
Key Capabilities
- Autonomous Android Task Execution: The model enables LLM-based agents to perform tasks across a wide range of Android applications, including common ones like messaging, clock, email, and settings.
- Data Scarcity Handling: A core feature is its ability to function effectively without the need for extensive, manually labeled interaction data, addressing a significant challenge in developing mobile automation agents.
- Agentic Behavior: It facilitates the creation of intelligent agents capable of understanding and interacting with the Android environment.
Good For
- Developing LLM-based agents for Android automation.
- Research and applications requiring autonomous interaction with mobile applications.
- Scenarios where manual data labeling for Android interactions is limited or impractical.
For more technical details and inference code, refer to the official GitHub repository.