UnityPredict Model Inference Page

The Basics

The Model Inference page is a dynamic and user-friendly interface designed to help users explore, interact with, and use various AI models in real time. It also serves as an easy to use, yet simple interface for developers to experiment with the model and analyze its behavior and predictions.

All components of this page are generated from the information that the model creator has provided in the Model Editor

Below are a functionalities specific to this page.

• Total Performance Score: This field is automatically populated from the average of all other numeric metrics for the model.

• Claim This Model: Allows a request to be submitted to the UnityPredict team to claim ownership to for a model. You must provide proof that you are the rightful owner of this model.

• Reviews: As UnityPredict is an open marketplace for AI solutions, users can rate and review any public model.

Inference

This section allows users to try out an model by providing the required inputs and clicking the Predict button.

Once Inference is complete, the outcomes are displayed based on the configuration that the model developer provided in the Model Editor.

The Copy icon next to the calculation cost allows users to easily copy the Model Id for use through UnityPredict APIs. The Warning icon (only displayed to the model developer) show the logs of the engine as the model is running.

Cost of Inference

Running a model requires compute on the cloud. UnityPredict engines are completely serverless. This means that charges occur only when inference is being executed. The cost of running a model consists of two parts:

1) UnityPredict Compute Cost: These are costs incurred by running the model/engine code on the cloud.

2) Developer Usage Fees: This is a cost that the model developer specifies for using the model.

The Pricing Info for the model are provided by model developers as a way to better explain how much a model costs to run. In addition to this information, UnityPredict also displays an Estimated Calculation Cost which is a value calculated from the average cost of prior runs of th model by users on the platform.