Yagna1/functiongemma-270m-mobile-actions
Yagna1/functiongemma-270m-mobile-actions is a 270 million parameter FunctionGemma model, fine-tuned for on-device function calling on Android devices. Optimized with INT8 quantization and an extended 1024-token KV cache, it enables low-latency execution of 6 native Android functions like flashlight control, contact creation, and map display. This model is specifically designed for deployment via Google AI Edge Gallery and LiteRT-LM runtime, facilitating offline mobile actions.
Loading preview...
Overview
Yagna1/functiongemma-270m-mobile-actions is a specialized 270 million parameter model, derived from Google's FunctionGemma, and fine-tuned for on-device function calling within Android environments. It is pre-converted to the .litertlm format, making it ready for immediate deployment using the Google AI Edge Gallery and LiteRT-LM runtime. The model is optimized with INT8 quantization, resulting in a compact size of approximately 271 MB, and features an extended KV cache of 1024 tokens for efficient mobile performance and longer conversations.
Key Capabilities
- On-device Function Calling: Executes functions directly on Android devices without requiring an internet connection.
- Optimized for Mobile: INT8 quantization and
.litertlmformat ensure efficient deployment and low latency on ARM-based Android devices. - Native Android Actions: Supports 6 specific Android functions:
- Flashlight control
- Contact creation
- Email sending
- Map display
- WiFi toggling
- Calendar event creation
- Natural Language Interface: Translates natural language prompts into structured function calls.
Use Cases and Limitations
This model is ideal for developers looking to integrate offline, voice-controlled actions into Android applications, particularly within the Google AI Edge ecosystem. It excels at mapping natural language commands to predefined device functionalities. However, it is currently limited to the 6 specified Android functions, supports only the English language, and may not handle complex multi-step actions or function parameters that deviate from its expected schema. It requires an Android device with ARMv8-A or newer architecture.