Exploring Pytorch + NPU: Lessons from the Lab
This event shares key takeaways and insights inspired by the collaborative environment of the Pytorch+NPU Labs, focusing on Phase 1 objectives: understanding PyTorch internals and basic NPU integration. We'll touch upon the challenges and opportunities in leveraging PyTorch as a development environment for NPUs, drawing inspiration from the work and expertise of industry leaders like Kim Hongseok (Chief Software Architect, Rebellions.ai).
What We'll Cover:
PyTorch Internals: A closer look at the inner workings of the popular deep learning framework.
PyTorch & NPU Integration: Fundamentals of connecting PyTorch with NPUs.
torch.compilefor NPUs: Discussion around the design considerations and implementation requirements for building custom NPU backends fortorch.compile.
Who Should Attend?
This event is ideal for:
Developers with a strong interest in PyTorch's internal mechanisms.
Engineers and researchers focused on NPUs, particularly from a development environment and software perspective.
Anyone involved in optimizing ML models (especially LLMs) for specialized hardware.
Prerequisites for Attendees:
To get the most out of this session, you should have:
Strong Python proficiency.
A basic understanding of Machine Learning (ML), particularly Large Language Models (LLMs).
Fundamental knowledge of Computer Architecture, Operating Systems (OS), and Compilers.
Experience with open-source development environments (like GitHub).
Context & Inspiration:
Our discussion is informed by cutting-edge work happening in the field, exemplified by experts like Kim Hongseok. As Chief Software Architect at Rebellions.ai, Kim focuses on LLM optimization and utilizing PyTorch for NPU development environments. His extensive background includes roles as Head of R&D at Google Korea, software development positions at Microsoft and Facebook, and deep expertise in compiler research and development.
Join us to explore the convergence of deep learning frameworks and next-generation hardware!