Expand description
QKVP: Q42 Runtime Page Profile for Semantic KV and Cache Reuse
This module defines the page-aware structures for the KV cache, allowing the engine to intelligently reuse, quantize, or evict KV memory based on cognitive and thermodynamic context rather than simple chronological LRU.
Structs§
- Q42Chunk
Policy - Controls semantic chunking, boundary weights, and thermal biases. Padded to exactly 64 bytes.
- Q42Kv
Page Header - The root header for a KV cache page.
- Q42Query
Sketch - Drives query-aware min/max selection for the KV cache pages. Padded to exactly 128 bytes.