Inaugural Santa Cruz PyTorch Meetup Draws 45 Attendees

Organized by Steve Pousty of Red Hat and Stephanie Lieggi of the University of California, Santa Cruz Open Source Program Office, the debut event drew 45 attendees. The crowd, consisting of professional engineers, business leaders, graduate students, and undergraduate interns, gathered at a university engineering lecture hall to discuss GPU programming and real-world machine learning applications.

The technical sessions began with hardware-focused presentations. Faradawn Yang from Nvidia explained GPU architecture, detailing how CUDA cores handle basic matrix math while Tensor cores group higher-level operations to accelerate both training and inference. Anil Vishnoi from Red Hat followed with a presentation on programming PyTorch Kernels, comparing CPU and GPU threading models and sharing optimization strategies for developers working directly with graphics hardware.

Three lightning talks highlighted the diverse applications of PyTorch. Filippo Balzaretti, a chemistry postdoctoral researcher, demonstrated how neural networks can approximate interatomic potentials to simulate large molecular systems. Undergraduate Kameron Benjamin showcased Green Guardian, a computer vision application he built to diagnose plant diseases from photos. Finally, doctoral student Manasi Pawar discussed her research on Visual Language Action models in autonomous driving, warning that reasoning traces cannot always be trusted, citing an instance where a model turned left but claimed in its reasoning trace that it turned right.

The organizers plan to continue the series, with the next meetup featuring Rob Timpe from OpenTeams discussing how to use and debug Torch.compile. Future events aim to feature speakers from local companies like Joby, Looker, and Fullpower-AI, while exploring alternative local venues.