KV Cache ·
Linear-time decoding
un
seel
.com · Cache K & V · O(n) per token · Transformer inference
Tokens
0
Attn ops/step
0
State
—
New query (Q)
Cached keys (K)
Cached values (V)
Attention output
▶ Play
←
→
🔇 Unmute
↻ Reset
Un
seel
.com · KV Cache