litert-community/FunctionGemma_270M_Mobile_Actions
FunctionGemma_270M_Mobile_Actions is a 270 million parameter instruction-tuned model from Google's FunctionGemma family, specifically fine-tuned for mobile assistant actions. It excels at generating structured function calls for common mobile device tasks like calendar management, email, and settings control. This model is optimized for on-device deployment and achieves 99.67% token-level accuracy on mobile action function calling.
Loading preview...
Overview
This model, litert-community/FunctionGemma_270M_Mobile_Actions, is a specialized fine-tune of Google's google/functiongemma-270m-it. It is designed to translate natural language user requests into structured function calls for mobile device actions. Trained on the google/mobile-actions dataset, it learns to emit outputs in the format call:<function_name>{arg1:value1,arg2:value2,...}.
Key Capabilities
- Structured Function Calling: Generates precise function calls for mobile tasks instead of natural language responses.
- Mobile Action Specialization: Optimized for a specific set of 7 mobile actions, including:
turn_on_flashlight()/turn_off_flashlight()create_contact(first_name, last_name, phone_number?, email?)send_email(to, subject, body?)show_map(query)open_wifi_settings()create_calendar_event(title, datetime)
- High Accuracy: Achieved 99.67% token-level accuracy on the validation set for mobile action function calls.
- On-Device Deployment: Can be converted to
.litertlmformat for efficient deployment on Android devices via Google AI Edge.
When to Use This Model
This model is ideal for:
- Mobile AI assistants requiring execution of device actions based on user input.
- Voice-controlled mobile applications that interact with device features.
- Conversational agents needing to trigger specific mobile functions.
- On-device AI applications where a small, specialized model for function calling is beneficial.