App Launch Icon Prompting for New Feature Discovery

The patent divides the large language model into multiple smaller language models with different scales and capabilities. These segmented models can be selectively activated based on task requirements, reducing overall memory occupation and power consumption while maintaining the ability to handle complex natural language interactions when needed.

The system dynamically selects and switches between different language models based on the specific interaction scenario, user needs, and resource availability. This dynamic adaptation allows the terminal to optimize power consumption by using smaller models for simple tasks and larger models only when complex language understanding is required.

The large language model is segmented into multiple smaller models with different parameter scales. These segmented models can be loaded into the limited terminal memory based on task requirements, avoiding the need to load the entire large model while maintaining sufficient accuracy for specific tasks.

The system employs a multi-functional language model architecture where smaller models can handle common tasks and larger models are invoked only when needed. This universal design allows the terminal to achieve high response accuracy for various tasks without permanently occupying large memory resources.

The patent segments the language model processing into local and cloud components. Frequently used smaller models are deployed locally on the terminal for immediate processing, while more complex tasks are offloaded to cloud services. This segmentation reduces network dependency for routine operations and minimizes processing latency.

The system pre-loads and caches commonly used language models and their responses locally on the terminal. This preliminary action allows the terminal to process common queries without network access, reducing latency and network dependency while maintaining high processing capability for urgent tasks.