YAML Metadata Warning: empty or missing yaml metadata in repo card (https://huggingface.co/docs/hub/model-cards#model-card-metadata)

functiongemma-synthetic-270m-int8

FunctionGemma 270M fine-tuned on synthetic Mobile Actions dataset with INT4 quantization.

Features

  • โœ… Trained on 138 synthetic examples for LiteRT-LM compatibility
  • โœ… INT8 quantized for optimal on-device performance
  • โœ… Compatible with Google AI Edge Gallery app
  • โœ… Supports: flashlight, contacts, email, maps, WiFi, calendar

Files

  • mobile/functiongemma_synthetic_int4.task - Ready-to-use .task bundle for Android
  • mobile/functiongemma_int8.tflite - INT8 quantized TFLite model
  • mobile/tokenizer.model - SentencePiece tokenizer

Usage

  1. Download functiongemma_synthetic_int8.task
  2. Import into Google AI Edge Gallery app
  3. Enable "Mobile Actions" support
  4. Test with prompts like "Turn on flashlight" or "Create contact John Doe"

Training Details

  • Base: unsloth/functiongemma-270m-it
  • Dataset: Custom synthetic Mobile Actions (138 examples)
  • Format: <start_function_call>call:function{params}<end_function_call>
  • Quantization: INT4 for edge deployment
Downloads last month
20
Safetensors
Model size
0.3B params
Tensor type
BF16
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support