Unverified50% confidenceFactExact time
KV-cache reuse presupposes a stable prefix, and GraphRAG flattens the graph structure into linear text, so the context prefix differs almost every query and the cache hit rate is extremely low.
1
Sources
50%
Confidence
Long-term
Relevance
10/19/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedTraditional inference frameworks pre-allocate contiguous KV Cache memory blocks for each concurrent request, resulting in severe memory fragmentation and low utilization.72% similarUnverifiedEvicting the KV entries of low-value tokens has become a key means of reducing the cost of long-context inference.67% similarUnverifiedKV cache reuses the Key and Value matrices of historical tokens during the inference phase, avoiding repeated attention computation on already-generated sequences.66% similarUnverifiedLinear attention-style architectures use a fixed-dimensional recurrent state to accumulate historical information; this state is a lossy compression of the context read so far, with a dimension far smaller than the original KV cache.65% similarUnverifiedKV感知路由通过维护“哪个节点缓存了哪段上下文”的索引，把新请求优先导向已持有相关KV的节点，从而复用缓存、跳过重复prefill；但KV缓存需要在节点间迁移或复制，路由决策失准会使命中率下降、收益迅速消失。65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1067955API
curl https://kongchang.com/api/v1/knowledge/claims/1067955MCP
get_claim(id=1067955)