t5-recipe-generation

Maintained By
flax-community

t5-recipe-generation

PropertyValue
Parameter Count223M parameters
Model TypeT5-based Seq2Seq
Training Data2.2M+ cooking recipes (2.14GB)
Developed ByFlax Community

What is t5-recipe-generation?

The t5-recipe-generation model is a specialized T5-based transformer designed to generate detailed cooking recipes from lists of ingredients. Developed as part of the Flax/Jax Community Week and sponsored by Google, this model demonstrates impressive capabilities in understanding ingredient combinations and generating coherent, structured cooking instructions.

Implementation Details

Built on the T5 architecture, this model processes input ingredients using a text-to-text generation approach. It significantly outperforms the baseline RecipeNLG model across all metrics, including COSIM (0.7282), WER (0.7613), and BLEU (0.3245). The model uses specialized tokens for formatting, including separators and section markers, to maintain a clear recipe structure.

  • Trained on RecipeNLG dataset with over 2.2 million recipes
  • Implements beam search and sampling strategies for generation
  • Supports structured output with title, ingredients, and directions sections
  • Uses F32 tensor type for computations

Core Capabilities

  • Generates complete recipes from simple ingredient lists
  • Produces properly formatted, step-by-step cooking instructions
  • Maintains coherent recipe structure with titles, ingredients, and directions
  • Handles various ingredient combinations and cooking styles

Frequently Asked Questions

Q: What makes this model unique?

This model stands out for its superior performance compared to previous recipe generation models, achieving significantly better scores across all evaluation metrics. It's particularly notable for maintaining coherent recipe structure while generating creative yet practical cooking instructions.

Q: What are the recommended use cases?

The model is ideal for recipe suggestion systems, cooking applications, and meal planning tools. It can help users generate recipes from available ingredients, assist in meal preparation, and provide creative cooking suggestions.

The first platform built for prompt engineering