Expand description
Radix tree prefix cache for KV block reuse.
Token sequences are chunked at block_size granularity. Each node in
the tree corresponds to one KV cache block. Lookup walks the tree
matching block-aligned chunks, returning cached physical block indices.
Thread-safe via Mutex<RadixTreeInner>.
Structsยง
- Radix
Tree - Thread-safe radix tree prefix cache.