Key concepts: inferencing and training models with Caikit

To use Caikit to manage your AI model, you must first define the following two key components:

• Module: The module defines the entry points for Caikit to manage your model. In other words, it tells Caikit how to load, infer, and train your model. An example is the text sentiment module.

• Data model: The data model defines the input and outputs of the model task. An example is the text sentiment data model.

The model is served by a gRPC server that can run as is or in any container runtime, including Knative and KServe. Here is an example of the text sentiment server code for gRPC. This example references the module configuration in this config.yaml file. This configuration specifies the module(s) (which wraps the model(s)) to serve.

Examples

This example of a client is a simple Python CLI that calls the model and queries it for sentiment analysis on two different pieces of text. The client also references the module configuration.

Check out the full Text Sentiment example or the model user tutorial to understand how to load and infer a model using Caikit.