Skip to content

Commit c2101e4

Browse files
committed
docs(kvcache): note extra_keys (pos 8) deferred to follow-up PR
Signed-off-by: pjdurden <prajjwalchittori1@gmail.com>
1 parent 1430d3e commit c2101e4

1 file changed

Lines changed: 4 additions & 1 deletion

File tree

pkg/cache/kvcache/event_types.go

Lines changed: 4 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -107,7 +107,10 @@ type BlockStoredEvent struct {
107107
LoraID *int64 // position 5; deprecated upstream in favor of LoraName
108108
Medium *string // position 6; storage tier, e.g. "GPU"/"cpu" (nil = unspecified)
109109
LoraName *string // position 7; canonical adapter id
110-
GroupIdx *int64 // position 9; KV-cache group for hybrid-attention models
110+
// position 8 (extra_keys) is intentionally not decoded here; it is consumed
111+
// in the follow-up PR for block-hash reconstruction. GroupIdx is read at its
112+
// fixed position 9 regardless.
113+
GroupIdx *int64 // position 9; KV-cache group for hybrid-attention models
111114

112115
// NOTE: These are NOT part of msgpack
113116
Timestamp time.Time `msgpack:"-"`

0 commit comments

Comments
 (0)