Skip to main content

Module pool

Module pool 

Source

Structs§

CompressedAttentionHit
One selected compressed-attention hit from the shared cold pool.
CompressedAttentionSelection
Output of compressed-domain top-k attention selection over the cold pool.
DecompressedLayer
Reconstructed K/V tensors for one layer of the shared pool.
FullyPreparedCompressedIndex
Fully prepared compressed index: pre-unpacks all indices and norms so the scoring loop is just Gram table lookups with no per-call unpacking.
PoolLayer
One layer’s worth of compressed KV blocks in the shared pool.
PrefetchedGramRows
Pre-fetched Gram rows for a specific query, enabling cache-friendly scoring.
PreparedCompressedIndex
Pre-built compressed index for one layer/head, avoiding per-call reconstruction.
SharedKVPool
A shared, compressed KV cache pool.

Traits§

CacheTarget
Trait for KV cache targets that can receive injected blocks.