NOTE:

This repository was mostly made redundant by the functions feature of the GPT models.

Better Structured GPT with Pydantic models

Better Structured GPT is a Python library designed to provide structured outputs from the OpenAI GPT model. It uses Pydantic models to shape the text responses generated by GPT into structured data which can be validated on an attribute-level.

Powerful and Reliable Structured Outputs with Better Structured GPT

Here, we demonstrate a user input, the GPT output, and the final Pydantic model populated with the structured output.

Step 1: User Input

user_prompt = "Generate a brief profile for a fictional character."

Step 2: Define a Pydantic Model

from pydantic import BaseModel
from typing import List

class CharacterProfile(BaseModel):
    name: str
    age: int
    profession: str
    hobbies: List[str]

Step 3: Populate the Model with Structured Output

character_profile = validator.validate_model(user_prompt, CharacterProfile)
print(character_profile)

Result

name='John Doe' age=30 profession='Space Explorer' hobbies=['Astrophotography', 'Alien Languages']

Better Structured GPT enables you to convert unstructured GPT outputs into structured data.

Prerequisites

This project requires Python 3.8 or newer. It makes use of the OpenAI API, so make sure to obtain your API key. This API key needs to be provided through an environment variable named OPENAI_API_KEY.

Installation

You'll need Poetry, a tool for dependency management in Python projects. If you haven't installed Poetry yet, you can do so by following the instructions on the official Poetry website.

Once Poetry is installed, clone the repository and navigate into the project's root directory:

git clone https://github.com/Jakob-98/better-structured-gpt.git
cd better-structured-gpt

Next, use Poetry to install the project's dependencies:

poetry install

Usage

After installing the project, you can use the GptValidator class to validate and structure GPT outputs.

You first need to import the necessary classes and create an instance of GptValidator:

from pydantic import BaseModel
from better_structured_gpt import GptValidator


validator = GptValidator()

Next, you can use the validate_model method to generate a structured output:

# define your Pydantic model
class MyModel(BaseModel):
    ...

# define your user prompt
user_prompt = "Your user prompt here"

# generate and validate the GPT output
model_out = validator.validate_model(user_prompt, MyModel())

The validate_model method will return an instance of your Pydantic model, with the data structured according to the model's definition.

Please read the source code and example files for more detailed usage instructions.

Tests

The tests currently do live calls to the OpenAI API. This is not best practice, instead I should probably add some mocking tests. TODO.

Contributing

Contributions are welcome! Please read our contributing guide.