Explore the Documentation#

This documentation will help you program for the CS system. It covers both basic and advanced topics. Use these docs to accelerate your machine learning training and inference applications on the CS system. Here you will find getting started guides, quickstarts, tutorials, code examples, release notes, and more.

Learn Cerebras basics

Big picture view of a CS system How Cerebras works

Start with this big picture before you dive into your ML development with Cerebras system.

Programming model and the compiler

Get to know how Cerebras separates compile vs execution, and the compiler flow from framework to the executable.

The Cerebras server cluster

How a Cerebras multi-worker configuration differs from a GPU multi-worker configuration.

Weight streaming vs pipelined execution

How to run an extremely large model such as a GPT-3 model (with ~1 billion parameters) or a smaller sized model.

ML workflow on Cerebras Cerebras ML workflow

Whether your framework is TensorFlow or PyTorch, get to know the general ML workflow on Cerebras first.

Quickstarts

But first, a checklist Checklist

Go over this quick checklist to ensure you have access credentials, hostnames, IP address of the CS system and more.

Checkout the reference models

Reference code Cerebras Model Zoo Reference Models

GitHub repo with a partial list of various neural network models currently supported.

Start developing

Dive into developing with TensorFlow and PyTorch

PyTorch or TensorFlow, we got you covered. Start your framework-to-CS journey here.

Optimize your ML code Performance optimization practices

Determine optimal Slurm resources, use the compiler report to enhance input function.

Extensions

Contact

For questions and feedback on documentation Contact docs@cerebras.net via email.

For support Contact support@cerebras.net via email.

Cerebras company website

Visit cerebras.net.