Unverified50% confidenceFactExact time
Split-KV uses a cost-aware longest-processing-time-first (LPT) strategy to allocate context KV blocks, with a load imbalance rate below 1%.
1
Sources
50%
Confidence
Long-term
Relevance
11/5/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedKV cache runtime provides a low-cost, fine-grained, real-time intervention capability that fills the gap between model fine-tuning and external scaffolding.70% similarUnverifiedIn everyday use of DeepSeek Harness, the KV cache hit rate for consecutive tasks is basically above 90%, and a second question on the same 60,000-plus-token document waits only 0.4 seconds.69% similarUnverifiedSplit-KV is an exact context-parallel (CP) method that maintains 85–91% weak scaling efficiency at a CP size of 64.69% similarUnverifiedCompressing the number of layers that bear KV cache overhead to one quarter of the total layer count under ultra-long contexts can significantly reduce VRAM costs for long-context inference, and is an important direction in the architectural evolution of long-context models69% similarUnverifiedKV-shareability means that in some trained recurrent models, the KV cache from the last loop can be directly reused by the earlier loops with almost no impact on quality.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1138025API
curl https://kongchang.com/api/v1/knowledge/claims/1138025MCP
get_claim(id=1138025)