In this workshop, Jim Burtoft (AWS) walks through an end-to-end Trainium setup for the Global Chess Challenge: starting from a Trainium instance, cloning the AWS Neuron workshops repo, installing dependencies, and running the chess deployment workflow. You can check out the video here: https://youtu.be/3V52L65rV4c
The session also covers:
- How the Neuron stack fits together (Torch NeuronX, NeuronX Distributed, Optimum Neuron, vLLM integration) and what “compilation” means on Trainium.
- Deploying a pre-compiled model with a vLLM server, checking server readiness, and monitoring utilization with
neuron-top. - Running a sample game and a tournament, including why concurrency (batch size) matters for throughput.
Note: the hosted workshop environment shown in the video was only available during the live session, but you can still follow the same steps in your own AWS account using the resources below.
Resources from the workshop
- AWS Neuron workshops repo: GitHub - aws-neuron/neuron-workshops: Notebooks and sample code for Build On Trainium
- Workshop join page: https://join.workshops.aws/
- Alternate join page: Workshop Studio
- Run Trainium in your own AWS account (AWS re:Post): Getting started with Inferentia or Trainium | AWS re:Post
- YouTube reference link shared during the session: https://www.youtube.com/watch?v=CyTCTuq1z0Q