---
base_model: google/functiongemma-270m-it
tags:
- function-calling
- mobile-actions
- gemma
- flashlight
language:
- en
license: gemma
---

# FunctionGemma 270M — Mobile Actions (Flashlight)

Fine-tuned from [`google/functiongemma-270m-it`](https://huggingface.co/google/functiongemma-270m-it) on the
[Google Mobile Actions](https://huggingface.co/datasets/google/mobile-actions) dataset,
filtered to **flashlight** related samples (1,509 train / 175 eval).

## Training Details

| Setting            | Value                     |
|--------------------|---------------------------|
| Base model         | google/functiongemma-270m-it             |
| Dataset            | google/mobile-actions     |
| Filter             | flashlight samples only   |
| Train samples      | 1,509                     |
| Eval samples       | 175                       |
| Epochs             | 2                         |
| Batch size         | 4 (effective 32)          |
| Optimizer          | AdamW (fused)             |
| Precision          | BF16 + TF32               |
| Train loss         | 0.023                     |
| Eval loss          | 0.0084                    |
| Token accuracy     | 99.72%                    |

## Usage

```python
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline

model     = AutoModelForCausalLM.from_pretrained("arunkumar629/functiongemma-270m-it-mobile-actions")
tokenizer = AutoTokenizer.from_pretrained("arunkumar629/functiongemma-270m-it-mobile-actions")
pipe      = pipeline("text-generation", model=model, tokenizer=tokenizer)
```