YFarmX

PagedAttention

AI

PagedAttention: A memory management technique that stores the KV cache in small non-contiguous blocks, like operating-system paging, reducing waste and letting more requests run at once.

Related terms

Browse the full glossary →