KV Chunk (Key-Value Chunk)

Appears in 1 paper

A segment of the key and value matrices corresponding to a subset of the sequence.

As used in Paper 19 — Ring Attention with Blockwise Transformers for Near-Infinite Context →

A segment of the key and value matrices corresponding to a subset of the sequence. In Ring Attention with P GPUs, each GPU initially holds one KV chunk. As computation proceeds, chunks circulate around the ring.