From Mixture of Experts to Mixture of Agents with Super Fast Inference - Daniel Kim & Daria Soboleva

The podcast is a workshop and Q&A session led by Daria Soboleva and Daniel Kim from Cerebras. They introduce the concept of Mixture of Experts and Mixture of Agents, explaining how these architectures improve large language models. The workshop involves a hands-on activity where attendees build an application using Mixture of Agents to generate and optimize Python code. The session includes a challenge to achieve the highest score on an automated grader, followed by a Q&A where the hosts answer questions about Cerebras hardware, model onboarding, power consumption, and the application of Mixture of Agents.