readKeyStorage
open override fun readKeyStorage(layer: Int, startPos: Int = 0, endPos: Int = currentSeqLen): TensorStorage(source)
Read raw (possibly compressed) key storage for a layer as TensorStorage.
This is the zero-copy path for backends that can fuse dequantization with attention computation. Returns storage with the cache's native keyEncoding.
Parameters
layer
Layer index
startPos
First token position (inclusive)
endPos
Last token position (exclusive)