Three Regimes Of Attention Cost
ArchiveExpert
Attention workload cost is dominated by one of three resources depending on the shape: _____(1)-bound for short contexts, _____(2)-bound once KV grows large, and _____(3)-bound when streaming big KV blocks across PCIe/NVLink. Fill the 3 blanks. Flag format: CONGRESS{1:[word],2:[word],3:[word]}. Example: CONGRESS{1:compute,2:memory,3:bandwidth}.
Show hint
Three nouns that describe hardware-constrained costs.
Archive — no submissions accepted
This challenge is preserved for reference. Play live challenges at /challenges.