The per-forward masks of the model this layout serves.
The window read of the storage it was made from (KVCache.windowRead).
ReadonlymasksThe per-forward masks of the model this layout serves.
ReadonlywindowThe window read of the storage it was made from (KVCache.windowRead).
Every row unpadded and at one position: a windowless read needs no per-row mask. Layouts that keep their positions in this one read it.
Compiled-decode trace adapters expose the RoPE offset as an int32 array input here; real caches leave it unset (static int path).
Fill spans run on this layout (the target layout of a full-attention layer). A bf16 committed span reads exactly as a window does.
Decode: one new position per row (L = 1). Each row's query sees that
row's valid stored positions up to and including the one just appended,
limited to the cache's sliding window.
Window: L new positions per row, causal within the window, for prefill
chunks, prefill tails and speculative verify windows. Query i of a row
sees the row's earlier valid positions and window positions 0..i, limited
to the cache's sliding window. Left padding and prepared prefill padding are
never visible. Outputs at padded query positions are unspecified and the
caller ignores them. To append context without attending, make this call
and dispose the read unused.
Physical cache storage consumed by one additional token for one row. Recurrent caches return 0 because their state does not grow with the sequence. Optional for stateless/training adapters that are never admitted or persisted.
Can trim(n) drop the last n tokens? (Ring caches lose trimability
once wrapped.)
Stable storage identity for compatibility guards and persistence. REQUIRED: every capability predicate (isPlainKvCache, isRotating*, packages/inference/src/state/persistence.ts codecs) is a string compare on this value, so a missing override used to fail OPEN into wrong routing (PRs #42/#43: BatchedRotatingCache shipped without one and batched joins dropped running rows). Wrappers report their own kind plus the inner kind.
Speculative verify-round support for NON-trimmable recurrent caches
(SSMCache — gated-DeltaNet conv + recurrent state). A cache exposing
these is spec-eligible without isTrimmable(): the serve loop arms a
round before the verify forward (the layer snapshots pre-round state —
free, MLX arrays are immutable — and records its position-local kernel
inputs), then resolves it: commit on full accept, or rollback(keep)
which restores the snapshot and bit-exactly REPLAYS the first keep
window tokens through the recurrence (identical arithmetic prefix ⇒
identical state). Trimmable caches leave these unset and keep trim().
Drop the last n tokens (future writes overwrite them). bypass (default
false → unchanged behavior) skips the isTrimmable guard for the
speculative-decode rollback case (trim the last n ≤ γ tips right after a
1 concat write); rotating caches then physically shrink the buffer tail.
Full-attention storage with independently advancing row positions. Speculative rollback changes coverage, without moving the retained KV. Subsequent appends overwrite each row's rejected suffix in place when MLX can donate the buffer. Queue membership is outside this layout.
Reads: append through
updateAndFetch, then the stock fused SDPA, a window throughwindowRead(its source storage's). While every row is unpadded and at one position, decode passes no mask and a window the fused causal mask; otherwise both pass the per-row mask (causal, past each row's left padding) for their query count, built once per forward throughmasks.