The Attention Kernel Of Dao
ArchiveExpert
Dao et al. (2024) released a Hopper-specific rewrite of the fast attention kernel exploiting warp-specialization and FP8, shipping with PyTorch and cuDNN. Which name (with version)?
Show hint
The family + a major version bump.
Archive — no submissions accepted
This challenge is preserved for reference. Play live challenges at /challenges.