Explore the world of efficient AI in this 45-minute lecture from the Scalable Parallel Computing Lab at ETH Zurich. Delve into the realm of billion-parameter artificial intelligence models and their exceptional performance across various tasks. Discover the techniques used to improve the efficiency of AI supercomputers during model training, including data ingestion, parallelization, and accelerator optimization. Learn about the challenges of efficient inference and explore model compression and optimization techniques such as fine-grained sparsity and quantization. Understand how these methods can reduce model size and significantly improve efficiency, potentially enabling powerful AI models to run on handheld devices.