Model techniques map
Techniquesmodel architecturetoken mixersparse attentionsparse attention indexer

specific method · filed under model architecture

Compressed Sparse Attention 2

A CSA variant that shares global key-value state across layers and reuses sparse selections to reduce cache and indexing costs.

Also called Compressed Sparse Attention 2 (CSA2), cross-layer KV-cache and index reuse, cross-layer KV and index reuse, CSA2.

sources
2
model
1
lab adopt it
1
strongest
core

How sources treat it

One count per evidence span, weakest treatment to strongest.

used 1core 5

Documented in

Evidence

6 spans quoted from the sources, strongest treatment first.

CSA2 shares global KV across layers to reduce cache storage and reuses sparse selections to reduce indexing work.

coremodel architecturein DeepSeek-V4.1-FlashDeepSeek

cross-layer KV cache reuse in Compressed Sparse Attention 2 (CSA2) and FP4 KV caching reduce its global KV cache footprint

coremodel architecturein DeepSeek-V4.1-FlashDeepSeek

we design Compressed Sparse Attention 2 (CSA2), which applies cross-layer reuse to global KV (including main KV and indexer K) and Top-K indices to substantially reduce KV cache storage.

coremodel architecturein DeepSeek-V4.1-FlashDeepSeek

CSA2 shares main KV and indexer K across layers and allows layers to reuse Top-K indices

coreinference servingin DeepSeek-V4.1-FlashDeepSeek

DeepSeek-V4.1-Flash uses CSA2, which assigns each attention layer one of three static modes — Full, Reindex, or Reuse — to share main KV and indexer K across layers and reuse Top-K sparse-attention indices.

coremodel architecturein DeepSeek-V4.1-FlashDeepSeek

The remaining 18 encoder layers use CSA2 with a compression rate of 𝑚= 2.

usedunclearin DeepSeek-V4.1-FlashDeepSeek

Filed alongside

Other methods under model architecture :: token mixer :: sparse attention :: sparse attention indexer.