Sharing

GradIO comes with built in support for sharing models. When GradIO launches an interface for a model, it also creates a shareable link that can be sent out by the creator to collaborators so they can access the interface from their browsers. The link does not require these remote users to have GradIO, Python, or any environment set up.

This link, which is printed to the console, is generated by adding the parameter share=True when launching the interface: e.g. io.launch(share=True). The link stays active for 8 hours. Other users can access the GradIO interface from the link remotely, but the model stays on the original host computer. When remote users submit inputs to the GradIO interface, the browser makes a call to the host computer with the input provided. On the host, the model will generate the output based on the provided input, and return the output, which will be rendered in the remote browser’s output pane.

Because these remote interfaces will run models on the host machine and draw from its compute power, we do not recommend publicly share the generated links. Distribute them among trusted colleagues only.

When the remote user submits an input from their interface, the model executes on the host computer.

Public Hosting

The option for permanent public hosting on GradIO servers is coming soon! You will be able share your model publicly and collect user input data in this upcoming feature.