Unverified50% confidenceBenchmarkExact time
Split-KV is an exact context-parallel (CP) method that maintains 85–91% weak scaling efficiency at a CP size of 64.
1
Sources
50%
Confidence
Long-term
Relevance
11/5/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedSplit-KV uses a cost-aware longest-processing-time-first (LPT) strategy to allocate context KV blocks, with a load imbalance rate below 1%.69% similarUnverifiedPrompt缓存（KV Cache）在相同系统提示多次使用时可复用缓存的KV矩阵，通常将延迟降低40-60%、成本降低50-90%61% similarUnverifiedMagnitude fuses KV-cache updates with attention computation through kernel fusion, cutting cudaMemcpy overhead from 22% to 4%.60% similarUnverifiedOperator fusion merges multiple consecutive operators into a single composite operator, allowing intermediate results to complete computation in registers or cache, reducing memory bandwidth consumption by 60%–80%.60% similarUnverifiedThe KV cache introduced by extremely long contexts grows linearly with context length, and Unsloth's optimized offload planner prioritizes keeping the KV cache on the GPU.60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1138022API
curl https://kongchang.com/api/v1/knowledge/claims/1138022MCP
get_claim(id=1138022)