How KDA Changes Caching for Kimi K3
A Juejin AI source-code analysis examines how Kimi K3 combines token-addressable MLA history with KDA convolution and recurrent state. It argues that prefix caching must restore a coherent set of MLA, convolution, and recurrent objects at the same boundary rather than merely reuse KV blocks. The article compares SGLang’s KDA-aware checkpointing and unified-memory approach with vLLM’s state blocks, metadata, fine-grained partial hits, and retention strategies.
The analysis identifies a concrete serving-stack requirement for Kimi K3: prefix reuse becomes a state-restoration problem rather than a KV-only operation, requiring frameworks to preserve state boundaries, ownership, and lifecycle consistency.
Some source content is omitted. Continue reading on the original site.