Gradio

Gradio is an open-source Python package that allows you to quickly build a demo or web application for your machine learning model, API, or any arbitrary Python function.

• Gradio documentation
• Gradio examples
• Gradio source

Deploying

Deploy Gradio app with the rsconnect-python package.

When deploying a Gradio app, ensure that you specify the correct entrypoint. for the specific application you are deploying. The example application on this page has its source code in a file named app.py. The application object is named demo. Therefore, the entrypoint for this application is app:demo. See the documentation on entrypoints for more information.

rsconnect deploy gradio -n <saved server name> --entrypoint app:demo

Example app

app.py

import gradio as gr

def greet(name):
    return "Hello " + name + "!"

demo = gr.Interface(fn=greet, inputs="textbox", outputs="textbox")

if __name__ == "__main__":
    demo.launch()

To deploy this example, first save the code to a file named app.py.

Install any required dependencies. Then, test the app locally:

python app.py

Then deploy to Posit Connect:

rsconnect deploy gradio -n <saved server name> app.py

User meta-data

import gradio as gr

def get_user_info(text, request: gr.Request):
    user_info = request.headers["RStudio-Connect-Credentials"]
    return {
        "user_info": user_info
    }

demo = gr.Interface(get_user_info, "text", "json")
demo.launch()

User and group uniqueness

Most environments have unique usernames where each user identifies a single user and groups the name of the groups the user is a member of.

However, in large organizations with hundreds of users and groups, this may not be true. See the Admin Guide Credentials for Content sections for more information.

Models and data

Code that relies on the transformers or datasets library may attempt to fetch data from the Hugging Face Hub as the application starts. Depending on the networking configuration of the Connect server, this may result in failed deployments or application timeouts.

To work around this issue:

• Ask your administrator to configure a directory that Connect can use to cache downloaded data.
  
• Set the environment variable HUGGINGFACE_HUB_CACHE on a per-application basis or in a supervisor script.
• Increase the initial timeout for the application.

For smaller volumes of data, you can also download the required data and include it as part of your deployment. See the Hugging Face Transformers documentation for more information.

Alternatively, if your model framework enables installing trained pipelines as Python pacakages, you can specify the URL to those Python packages in the requirements.txt file for your application, which Connect will use to download the pipeline while reconstructing your app environment. See the Spacy documentation for an example: https://spacy.io/usage/models#download-pip

Limitations

Git LFS

Connect does not currently support Git LFS.

Flagging

Flagging is disabled.