Develop for the Cerebras System

With the Cerebras Software Platform, CSoft, you’ll spend more time pushing the frontiers of AI instead of optimizing distributed implementations.

Easily continuously pre-train multi-billion GPT-family models with up to an 20 billion parameters on a single device.

Get up and running quickly without changing your workflows. Utilize existing frameworks for your work or dive deep developing custom kernels – the Cerebras Software Platform is designed for flexibility and optimization.

Training Multi-Billion-Parameter GPT Models Is Easy

Multi-Billion Parameter GPT Video Walkthrough

Watch our PyTorch Overview

Community Meeting July 12th

Read SDK Technical Overview


Our PyTorch interface library is a simple wrapper for PyTorch program exposed through API calls that is easy to add as few extra lines of code for an existing Pytorch implementation.

Get started


Integration of TensorFlow is via Cerebras Estimator, which is a wrapper class we developed based on standard TensorFlow Estimator and standard TensorFlow semantics.

Get started


The Cerebras SDK allows researchers to extend the platform and develop custom kernels – empowering them to push the limits of AI and HPC innovation.

Request access

Example Reference Implementations

This repository contains examples of common deep learning models demonstrating best practices for coding for the Cerebras hardware.


Developer blogs

Training Multi-Billion-Parameter Models on a Single Cerebras System is Easy

Changing model size is trivial on Cerebras, rather than a major science project using conventional graphics accelerators.

Multi-Billion-Parameter Model Training Made Easy with CSoft R1.3

CSoft R1.3 delivers GPT-J continuous pre-training, more reference model implementations in PyTorch and even faster training with Variable Tensor Shape…

Increasing Model Throughput with Variable Tensor Shape Computations

We write machine learning algorithms to fit the data, not pad the data to suit hardware limitations.

Getting Started with PyTorch BERT Models on the Cerebras CS-2 System

This walk through shows how easy it is to adapt and run PyTorch models on a Cerebras system using the many convenient wrappers exposed in our API.


Don’t see your question?

Send us an email at

Please find our example reference model implementation here: To get access to our full list, please contact us at 

Please sign up for our newsletter!