Scaling AI workloads with Kubernetes

Tsavo Knott avatar Shivay Lamba avatar Kunal Kushwaha avatar

w/ Tsavo Knott, Shivay Lamba & Kunal Kushwaha

Recorded: 3rd July 2024

As AI models, particularly Large Language Models (LLMs), grow in size and complexity, their deployment becomes increasingly challenging. We explored the complexities involved and effective strategies for managing LLM/AI deployments on Kubernetes, focusing on cost-efficiency and scalability.

Given the intricate overlap between AI and DevOps, involving multiple aspects such as ML models, deployment scripts, and configurations, we also explored how to streamline these processes effectively. By leveraging live context from various documents and terminal screens via Pieces Copilot, we aim to provide real-time assistance and troubleshooting tips for deployment challenges.

What we discussed:

  • Best practices for configuring Kubernetes for AI workloads.
  • Solutions to facilitate smoother AI deployments.
  • Handling common issues and pitfalls in AI model deployment on Kubernetes.
  • Live demo showing how to deploy an AI model on Civo using GPU nodes.