Gemma 4 E2B Mobile Actions, 200-Sample Fine-Tune

Main model card: google/gemma-4-E2B-it

This repository provides an experimental Gemma 4 E2B checkpoint fine-tuned on google/mobile-actions for mobile action function calling.

The model was trained locally on an Apple M1 Pro with LoRA, then merged back into the base model for direct Transformers inference.

What It Does

The model converts natural language mobile assistant requests into tool calls, for example:

<|tool_call>call:show_map{query:<|"|>Patisserie Valerie at 208 Kensington High Street, London, W8 7RG<|"|>}<tool|>

Supported functions come from the google/mobile-actions dataset, including:

• create_calendar_event
• create_contact
• show_map
• open_wifi_settings
• send_email
• turn_on_flashlight
• turn_off_flashlight

Performance

Evaluation was run on 200 held-out examples from google/mobile-actions (metadata == "eval").

Per-Function Results

Training

The adapter was trained with TRL SFTTrainer and PEFT LoRA. The final weights in this repository are the merged model weights.

Usage

import torch
from transformers import AutoModelForCausalLM, AutoProcessor, pipeline

model_id = "ClarkBear/gemma4-e2b-mobile-actions-200"

processor = AutoProcessor.from_pretrained(model_id)
tokenizer = processor.tokenizer
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    dtype=torch.bfloat16,
    device_map="auto",
)

tools = [{
    "type": "function",
    "function": {
        "name": "show_map",
        "description": "Shows a location on the map.",
        "parameters": {
            "type": "object",
            "properties": {"query": {"type": "string"}},
            "required": ["query"],
        },
    },
}]

messages = [
    {"role": "system", "content": "You are a mobile assistant that calls tools."},
    {"role": "user", "content": "Show me Patisserie Valerie on Kensington High Street."},
]

prompt = processor.apply_chat_template(
    messages,
    tools=tools,
    tokenize=False,
    add_generation_prompt=True,
)

generator = pipeline("text-generation", model=model, tokenizer=tokenizer)
print(generator(prompt, max_new_tokens=160, do_sample=False)[0]["generated_text"])

Run with LiteRT-LM

This repository also includes pre-converted LiteRT-LM artifacts for on-device experiments:

Download and run the smaller 4-bit artifact with the LiteRT-LM CLI:

uv tool install litert-lm

hf download ClarkBear/gemma4-e2b-mobile-actions-200 model-wi4.litertlm \
  --local-dir ./gemma4-e2b-mobile-actions-200-litert

litert-lm run \
  ./gemma4-e2b-mobile-actions-200-litert/model-wi4.litertlm \
  --prompt "Turn on the flashlight"

For Android or iOS apps, use the downloaded .litertlm file with the LiteRT-LM runtime APIs. If you have enough device storage and memory, compare the 8-bit and 4-bit artifacts for quality and latency.

Examples

This repository includes runnable examples:

python examples/run_transformers.py \
  --model-id ClarkBear/gemma4-e2b-mobile-actions-200 \
  --prompt "Turn on the flashlight"

Example prompts are available in examples/prompts.jsonl.

Limitations

This is an experimental small-data fine-tune.

• It was trained on only 200 examples.
• It may emit extra tool calls after the first valid tool call.
• show_map exact match is sensitive to address string formatting.
• Calendar events remain harder than simple device actions because they require title and datetime extraction.
• The LiteRT-LM artifact is large (model.litertlm is about 5 GB), so mobile deployment should target devices with enough storage and memory.

Fine-Tune and Compile Notes

This repository contains both the merged Hugging Face Transformers checkpoint and a LiteRT-LM .litertlm artifact exported from it.

The local training scripts and evaluation code are maintained in the companion project used to create this checkpoint.