Hands-on Virtual Training – Getting Ready to Use the Neocortex System

Session 1 presented on Tuesday, December 8, 2020, 11:00 – 1:00 pm (ET), by Dr. Natalia Vassilieva from Cerebras.
Session 2 presented on Thursday, December 10, 2020, 11:00 – 1:00 pm (ET), by Dr. Natalia Vassilieva from Cerebras.

This training was available to Neocortex Early User Program participants exclusively.

In this hands-on virtual training, we offer the Neocortex Early User Program (EUP) users access to an environment where they can compile their code with Cerebras specific tools and can also produce metrics that would inform the approach to optimally execute their code on the Cerebras CS-1 servers available in the Neocortex system. Users are also presented with sample models that can inform their own model development and instructed on the resources available to advance EUP preparatory activities.

Abstract

The Cerebras CS-1 is the groundbreaking specialized AI hardware technology to be featured on Neocortex, PSC’s upcoming NSF-funded AI supercomputer. This 2-hour hands-on training session is designed for users selected for the Neocortex EUP. The training focuses on subjects including learning how to prepare existing code for running on the Cerebras CS-1 servers. That training explored how to make the codebase of the EUP groups compatible with the Cerebras compiler, as well as how to calculate key performance aspects such as the theoretical max throughput for optimal code execution.

This training involves hands-on running DL models on the proxy compiling environment. An overview of the subjects covered is presented next:

1. Outline of the Cerebras software stack and Neocortex execution model.
2. Deep dive into a relevant Cerebras specific API with a detailed description of various functions and parameters.
3. Training a simple MNIST model with Cerebras specific API.
4. Evaluating the execution on CPU and GPU sample nodes.
5. Resources for using TF and the Cerebras estimator.

For more information about Neocortex, explore the Neocortex project page. For questions about this webinar, please email neocortex@psc.edu.