Archive
LLM Infrastructure

The Cache Of Attention

Archive
Very Easy
50pts0 solves
Between tokens of a decode step, most inference engines store the per-layer attention keys and values of every prior token to avoid recomputation. Flag format: CONGRESS{two-words}. Example: CONGRESS{attn cache}.
Show hint
A pair of letters + the memory noun.

Archive — no submissions accepted

This challenge is preserved for reference. Play live challenges at /challenges.