T5X Introductory Tutorial Series

Overview

This series of guides is a self-contained introduction to T5X, a modular, composable, research-friendly framework for high-performance, configurable, self-service training, evaluation, and inference of sequence models (starting with language) at many scales.

How to Use These Guides

Most entries in this series are colab notebooks (click the blue banners to the right of each heading below), allowing you to run our tutorial code interactively. We encourage you to do that! Play around, change things, see what happens!

T5X Guides

Codelab 1: An Introduction to T5X

In this colab, you will learn about some of the basic T5X components and put them to use to run training, inference, and evaluation on natural text inputs.

Codelab 2: Training Deep Dive

In this colab, you will dive into how to restore T5X models from checkpoints and run training, while also getting an introduction to the T5X trainer.

Codelab 3: Inference Deep Dive

In this colab, you will dive into how the Interactive Model does decoding to generate predictions and scores for a given input.

Codelab 4: Evaluation Deep Dive

In this colab, you will dive into how the InteractiveModel takes a batch of inputs and targets and runs evaluation to produce various metrics.

More Colabs coming soon!