Skip to main content

CI Pathways: Multi-GPU Training with Accelerate

04/22/25 - 01:00 PM - 03:00 PM EDT

Accelerate, HuggingFace's library for efficient multi-GPU training enables practitioners to scale their AI/ML applications effectively. This course will explore advanced training techniques for deep learning models, specifically targeting the ResNet50 architecture. Participants will gain a comprehensive understanding of the architecture and the significance of residual connections in deep neural networks. Additionally, the course will cover scheduling strategies to optimize learning rates during training, Weights & Biases (W&B) use for tracking metrics and experiments, and methods for saving intermittent checkpoints and resuming training.

Location

Virtual

Speakers

Priyam Mazumdar, NCSA