Unverified50% confidenceFactExact time
128K上下文窗口的注意力矩阵规模是32K窗口的16倍
1
Sources
50%
Confidence
Long-term
Relevance
7/8/2026
First Seen
Sources
Context Warp Drive：AI智能体确定性上下文折叠详解
hackernewshackernews7/7/2026
Related Claims
Unverified128K长度的序列所需的注意力计算量是4K长度序列的1024倍73% similarUnverified170k tokens场景下，注意力得分矩阵在BF16精度下约需54GB显存（170000×170000×2字节）73% similarUnverifiedGPT-4的上下文窗口从8K扩展到128K乃至百万级别，上下文窗口越大每次调用消耗的令牌越多70% similarVerified当前主流模型的上下文窗口约为128K至100万Token70% similarUnverified序列长度为4096时，N×N注意力矩阵在FP16精度下仅此一项就消耗约64MB显存68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/294258API
curl https://kongchang.com/api/v1/knowledge/claims/294258MCP
get_claim(id=294258)