openai-gradio

is a Python package that makes it very easy for developers to create machine learning apps that are powered by OpenAI's API.

Installation

You can install openai-gradio directly using pip:

pip install openai-gradio

That's it!

Basic Usage

Just like if you were to use the openai API, you should first save your OpenAI API key to this environment variable:

export OPENAI_API_KEY=<your token>

Then in a Python file, write:

import gradio as gr
import openai_gradio

gr.load(
    name='gpt-4-turbo',
    src=openai_gradio.registry,
).launch()

Run the Python file, and you should see a Gradio Interface connected to the model on OpenAI!

Voice Chat

OpenAI-Gradio also supports voice chat capabilities. You can enable this in two ways:

1. Using a realtime model:

import gradio as gr
import openai_gradio

gr.load(
    name='gpt-4o-realtime-preview-2024-10-01',
    src=openai_gradio.registry
).launch()

2. Explicitly enabling voice chat with any realtime model:

import gradio as gr
import openai_gradio

gr.load(
    name='gpt-4o-mini-realtime-preview-2024-12-17',
    src=openai_gradio.registry,
    enable_voice=True
).launch()

This will create a WebRTC-based interface that allows for real-time voice conversations with the AI model.

Voice Chat API Keys

For voice chat functionality, you'll need:

1. OpenAI API key (required for all features):

export OPENAI_API_KEY=<your OpenAI token>

2. Twilio credentials (required for WebRTC voice chat):

export TWILIO_ACCOUNT_SID=<your Twilio account SID>
export TWILIO_AUTH_TOKEN=<your Twilio auth token>

You can get Twilio credentials by:

1. Creating a free account at Twilio
2. Finding your Account SID and Auth Token in the Twilio Console

Without Twilio credentials, the voice chat will still work but might have connectivity issues in some network environments.

Customization

Once you can create a Gradio UI from an OpenAI endpoint, you can customize it by setting your own input and output components, or any other arguments to gr.Interface. For example, the screenshot below was generated with:

import gradio as gr
import openai_gradio

gr.load(
    name='gpt-4-turbo',
    src=openai_gradio.registry,
    title='OpenAI-Gradio Integration',
    description="Chat with GPT-4-turbo model.",
    examples=["Explain quantum gravity to a 5-year old.", "How many R are there in the word Strawberry?"]
).launch()

Composition

Or use your loaded Interface within larger Gradio Web UIs, e.g.

import gradio as gr
import openai_gradio

with gr.Blocks() as demo:
    with gr.Tab("GPT-4-turbo"):
        gr.load('gpt-4-turbo', src=openai_gradio.registry)
    with gr.Tab("GPT-3.5-turbo"):
        gr.load('gpt-3.5-turbo', src=openai_gradio.registry)

demo.launch()

Under the Hood

The openai-gradio Python library has two dependencies: openai and gradio. It defines a "registry" function openai_gradio.registry, which takes in a model name and returns a Gradio app.

Supported Models in OpenAI

All chat API models supported by OpenAI are compatible with this integration. For a comprehensive list of available models and their specifications, please refer to the OpenAI Models documentation.

Note: if you are getting a 401 authentication error, then the OpenAI API Client is not able to get the API token from the environment variable. This happened to me as well, in which case save it in your Python session, like this:

import os

os.environ["OPENAI_API_KEY"] = ...