cdameron1024/gemma-food-drink-extractor

Gemma Food & Drink Extractor

A small language model project based on Gemma, designed for structured food and drink extraction from text descriptions.

Task

Given a text description, the model is intended to return structured output in the following format:

food_or_drink: 0 or 1
tags:
foods:
drinks:

Base Model

Gemma 3 270M

Dataset

FoodExtract-1k

Intended Use

This project demonstrates:

• dataset preparation
• prompt/response formatting
• train/test splitting
• Hugging Face pipelines
• supervised fine-tuning workflow with TRL
• evaluation of predictions against test labels
• model packaging and Hugging Face Hub deployment

Example

Input:

A plate contains grilled chicken, rice, and broccoli.

Expected structured output:

food_or_drink: 1
tags: fi
foods: grilled chicken, rice, broccoli
drinks:

Limitations

This is an educational model project. Model quality depends on training completion, dataset quality, and hardware used for fine-tuning.

Author

Christopher Dameron