JackJ1/functiongemma-270m-it-mobile-actions-litertlm

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:0.3BQuant:BF16Context Size:32kPublished:Dec 19, 2025License:gemmaArchitecture:Transformer0.0K Featherless Exclusive Cold

JackJ1/functiongemma-270m-it-mobile-actions-litertlm is a specialized 270 million parameter model, fine-tuned from Google's FunctionGemma-270M-IT. Optimized for mobile actions, it translates natural language commands into structured function calls for Android OS tools. Packaged in the .litertlm format, this ultra-lightweight model is designed for low-latency, on-device inference, ensuring privacy and efficiency without cloud connectivity.

Loading preview...

Overview

This model, JackJ1/functiongemma-270m-it-mobile-actions-litertlm, is a specialized fine-tune of Google's FunctionGemma-270M-IT (270M parameters, Gemma 3 architecture). It is specifically optimized for Mobile Actions, enabling on-device translation of natural language commands into structured function calls for Android OS tools. The model is provided in a quantized q8 .litertlm format, making it suitable for immediate deployment on mobile devices using LiteRT (formerly TensorFlow Lite).

Key Capabilities

  • Ultra-Lightweight: Designed for low-latency, on-device inference without requiring cloud connectivity.
  • Expert at Mobile Tasks: Translates commands like "Turn on the flashlight" or "Set an alarm for 7 AM" into actionable function calls.
  • Privacy-Centric: Operates entirely offline, ensuring user data and queries remain private and secure on the device.
  • Optimized Format: Quantized q8 .litertlm for efficient performance on edge hardware.
  • Function Calling Logic: Requires an essential system prompt to activate its function-calling capabilities.

Good for

  • Developers building Android applications requiring on-device natural language understanding for mobile actions.
  • Use cases where privacy and offline functionality are paramount.
  • Integrating AI capabilities directly into mobile devices with minimal resource overhead.
  • Demonstrating specialized agentic workflows on small, efficient models.