Join us for a demo showcasing the power of Google Cloud TPU by walking through the complete lifecycle of a model, from post-training to inference at massive scale.
We’ll show you how to post-train a model using new RL capabilities in MaxText and Tunix, then take that same model and serve it on TPU seamlessly with the new JAX backend in vLLM.
Learn about how to rightsize a workload for TPU for both training and inference, optimizing for performance-per-dollar from beginning to end.
Sponsor(s):
Google
Speaker(s):
Brittany Rockwell
Product Manager
Google Cloud
Brittany Rockwell is a Product Manager for AI Inference at Google Cloud, where she leads the development of the TPU backend in vLLM, the most popular OSS library for language model inference. She currently lives in Seattle, WA.
Session Type:
General Session (Presentation)