Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Explore a search-based deep learning compiler designed for CPUs, GPUs, and ASICs in this 25-minute conference talk. Learn how Luminal takes an innovative search-first approach to automatically discover efficient kernels, including advanced techniques like flash attention, revolutionizing the way deep learning models are optimized across different hardware platforms. Discover the technical foundations behind this compiler's ability to automatically identify and implement high-performance computational patterns, understand its cross-platform compatibility spanning traditional CPUs to specialized ASICs, and gain insights into how search-based optimization can significantly improve deep learning model performance without manual kernel development.
