single mention · inference & serving
Key-as-value reuse in global attention layers
Reuses keys as values in global attention layers to reduce memory use.
The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See KV cache management for the methods filed beside it.
Also called key-value reuse in global attention layers, reuse of keys as values in global layers.
Filed under inference & serving › KV cache management