---
license: llama3.2
base_model: TinyLlama/TinyLlama-1.1B-Chat-v1.0
tags:
- food
- nutrition
- parsing
- health
- on-device
- ios
- fine-tuned
- mlx
- apple-silicon
language:
- en
datasets:
- custom
---

# GutSignal Food Parsing Model (MLX Format)

Fine-tuned language model for parsing food descriptions into structured data. **Optimized for Apple Silicon in MLX format.**

## Model Description

This model is fine-tuned to parse natural language food descriptions and extract:
- Individual ingredients
- Food categories (dairy, grains, protein, vegetables, etc.)
- Beverage classification
- Dairy content detection
- Estimated nutritional information

## Intended Use

- **Primary Use**: On-device food parsing for health tracking applications
- **Target Platform**: iOS devices with Apple Silicon (iPhone 15+)
- **Format**: MLX (Apple's machine learning framework)
- **Privacy**: Designed for offline, on-device inference

## Training Data

The model was fine-tuned on:
- Food journal entries, USDA FoodData Central, Open Food Facts

## Usage

### iOS/macOS (MLX)

This model is in MLX format and ready for Apple Silicon devices:

```python
import mlx.core as mx
from mlx_lm import load, generate

# Load model
model, tokenizer = load("YOUR_USERNAME/gutsync_food_analysis_tinyllama-1.1b")

# Generate
prompt = '''### Instruction:
Parse the following food description and extract structured data.

### Input:
chicken salad with ranch dressing

### Output:
'''

response = generate(model, tokenizer, prompt=prompt, max_tokens=256)
print(response)
```

### iOS App Integration

This model can be downloaded and used directly in iOS apps using the MLXLMCommon framework. See the [GutSignal app](https://github.com/ndlanier/poop.io-ios) for a complete example.

## Output Format

```json
{
  "ingredients": ["chicken", "lettuce", "ranch dressing"],
  "categories": ["protein", "vegetables", "fats"],
  "is_beverage": false,
  "contains_dairy": true,
  "estimated_calories": 450
}
```

## Categories

The model classifies food into these categories:
- dairy, grains, protein, vegetables, fruits
- fats, sugars, beverages, spicy, fiber
- processed, caffeine, alcohol, fermented
- nuts, legumes, unknown

## Training Details

- **Base Model**: TinyLlama/TinyLlama-1.1B-Chat-v1.0
- **Fine-tuning Method**: LoRA (Low-Rank Adaptation)
- **Training Framework**: PyTorch + Transformers
- **Conversion**: Converted to MLX format for Apple Silicon optimization
- **Format**: MLX safetensors (float16)
- **Optimization**: Efficient for on-device inference on Apple devices

## Limitations

- Estimates only - not for medical decisions
- Best performance on common foods
- May struggle with very complex dishes or regional cuisines
- Calorie estimates are approximate

## Ethical Considerations

- This model is for informational purposes only
- Not a substitute for professional nutritional advice
- Should not be used for medical diagnosis or treatment decisions
- Designed with privacy-first principles (on-device processing)

## License

This model inherits the license from its base model: TinyLlama/TinyLlama-1.1B-Chat-v1.0

## Citation

If you use this model, please cite:

```bibtex
@misc{gutsignal-food-parser,
  author = {Your Name},
  title = {GutSignal Food Parsing Model},
  year = {2026},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/YOUR_USERNAME/gutsync_food_analysis_tinyllama-1.1b}}
}
```