/
Independent academic archive · Rights remain with the original publishers · Streamed via YouTube
Play15s30s1x0:00/--:--
DoneWatched
Expand
GPU MODE·Dmitry Trifonov
54:01
0:00/54:01
Type below to lock the video timestamp, or press S to capture a slide.
Building a GPU Compiler from Scratch
Demystifying the construction of GPU compilers, this talk illustrates the process of lowering PyTorch graphs to efficient CUDA kernels through multi-level intermediate representations.
Abstract: Dmitry Trifonov demystifies ML compilers by showing how PyTorch layers become efficient GPU kernels through progressive IR lowering. Perfect if you plan to participate in the upcoming riftstack.ai hackathon or want a gentle, hands-on introduction to modern GPU compiler technology.