The complexity of deep learning models is surging, which warrants enhanced training and inference in distributed compute environments. This session focuses on the essential techniques to use with Intel® Data Center GPUs to balance distributed AI workloads and meet data center challenges to improve advances in efficiency and performance.

Explore Intel® Extension for PyTorch*, which optimizes neural network operations on Intel® hardware, and learn how DeepSpeed* can be integrated to perform training operations at scale.

Included topics:

• Tackle model scalability in a distributed environment skillfully, handling workloads efficiently across Intel Data Center GPUs.
• Gain familiarity with essential Intel tools to simplify operations, including PyTorch Distributed Data Parallel (DDP), Intel® oneAPI Collective Communications Library (oneCCL), and the DeepSpeed library that streamlines network training at scale.
• Deploy practical solutions that maximize hardware efficiency and perfect strategies that ensure top performance for AI development.
• Sample code and see benchmarking milestones, using tools such as Intel Extension for PyTorch large language models (LLM) to illustrate performance achievements.

Skill level: Any

• oneCCL
• Intel Extension for PyTorch (GitHub*) or from the AI Frameworks and Tools Selector
• Intel® Extension for DeepSpeed*