isahsn/FoodExtract-gemma-3-270m-fine-tune-v1
FoodExtract-gemma-3-270m-fine-tune-v1 by isahsn is a 0.3 billion parameter language model built on Gemma 3 270M, specifically fine-tuned for extracting food and drink related information from raw text. It classifies text as food or drink, assigns relevant tags, and extracts lists of edible food and drink items. This model is optimized for structured data extraction from text, making it suitable for filtering large datasets like image captions for food-related content.
Loading preview...
Model Overview
isahsn/FoodExtract-gemma-3-270m-fine-tune-v1 is a specialized language model, fine-tuned from Google's Gemma 3 270M, designed for the precise extraction of food and drink related entities from unstructured text. It processes raw text inputs to perform three key functions: classifying whether the text pertains to food or drink, tagging the content with predefined categories, and extracting specific food and drink items into separate lists.
Key Capabilities
- Food/Drink Classification: Determines if a given text describes food or drink items.
- Content Tagging: Assigns one or more relevant tags (e.g., 'ingredient list', 'menu', 'recipe', 'food items', 'drink items') to the text for fast filtering.
- Item Extraction: Identifies and lists all edible food-related items and drink-related items present in the text.
- Efficient Output: Generates condensed, structured outputs rather than verbose JSON, reducing token generation.
Training and Dataset
This model was trained using Supervised Fine-Tuning (SFT) via Hugging Face's TRL library on the FoodExtract-1k dataset. This dataset comprises 1400 samples of raw text paired with structured food extraction outputs, originally generated by gpt-oss-120b.
Use Cases
This model is particularly well-suited for:
- Data Filtering: Efficiently filtering large text datasets, such as image captions (e.g., DataComp-1B), to identify and categorize food and drink related content.
- Structured Data Generation: Converting unstructured text descriptions into structured data for analysis or database population.
Differentiation
Unlike general-purpose LLMs, this model is specifically optimized for the domain of food and drink extraction, offering a highly focused and efficient solution for this particular task. Its small parameter count (0.3B) makes it suitable for deployment in resource-constrained environments while maintaining high accuracy for its specialized function.