Gpu, Cuda, and PyTorch Performance Optimizations

Chris Fregly leads two back-to-back talks on GPU and CUDA optimizations for PyTorch, drawing from his systems performance engineering expertise. Expect concrete techniques for profiling bottlenecks, leveraging CUDA kernels, and squeezing throughput out of your models in production. This is hands-on content from someone who's written the book on it literally, so come ready to take notes if you're running models at scale or hitting performance walls. The meetup includes an intro segment with updates from organisers Fregly and Antje Barth, and a GitHub repo accompanies the session if you want to dig deeper afterward.