---
language:
- en
tags:
- text-generation
- gpt2
- culinary
- recipe-generation
license: mit
---

# chef_gpt2_recipe_generator

## Overview

`chef_gpt2_recipe_generator` is a Causal Language Model designed to generate cooking recipes based on a provided list of ingredients. It is fine-tuned on a large corpus of structured recipes, learning the relationship between ingredients, quantities, and instructions.

## Model Architecture

This model is based on the `gpt2-medium` architecture (355M parameters).

- **Base Model:** GPT-2 Medium.
- **Training Objective:** Causal Language Modeling (CLP) / Next-token prediction.
- **Data Format:** The model was trained on data structured as: `INGREDIENTS: [list of items] \n INSTRUCTIONS: [step-by-step guide] <|endoftext|>`.

## Intended Use

- **Culinary Inspiration:** Generating creative ideas for leftover ingredients in a fridge.
- **Creative Writing:** Assisting in generating content for food blogs or culinary fiction.
- **Data Augmentation:** Creating synthetic recipe datasets for downstream culinary NLP tasks.

### How to use

The model works best when prompted with the specific format it was trained on. You should provide a list of ingredients following the `INGREDIENTS:` tag.

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

model_name = "your_username/chef_gpt2_recipe_generator"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

chef = pipeline('text-generation', model=model, tokenizer=tokenizer, device=-1)

# Define ingredients
prompt = "INGREDIENTS: Chicken breast, garlic, soy sauce, honey, broccoli. \n INSTRUCTIONS:"

output = chef(
    prompt,
    max_length=400,
    num_return_sequences=1,
    do_sample=True,
    top_k=50,
    top_p=0.92,
    temperature=0.8,
    pad_token_id=tokenizer.eos_token_id
)

print(output[0]['generated_text'])