Unverified50% confidenceFactExact time
Naive-N0.5-Flash natively supports a 1 million token context and does not use a full attention mechanism
1
Sources
50%
Confidence
Long-term
Relevance
9/28/2026
First Seen
Sources
Related Entities
Related Claims
VerifiedFlashAttention通过重新设计注意力计算的内存访问模式，将显存占用从与序列长度平方成正比降低为线性关系62% similarUnverifiedSelf-Attention机制计算复杂度与序列长度平方成正比（O(n²)），一个128K token请求的原生注意力计算量约为4K token请求的1024倍62% similarUnverified大模型中存在注意力汇聚（Attention Sink）现象，即序列最开头的几个 token 往往保持较高的注意力权重61% similarUnverifiedMuse Glimmer采用混合注意力机制：大部分层只关注2048 token的局部窗口，每隔四层做一次全局全注意力60% similarUnverified标准Attention计算的显存占用和带宽消耗均为O(N²)，FlashAttention将显存占用降至O(N)60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/957115API
curl https://kongchang.com/api/v1/knowledge/claims/957115MCP
get_claim(id=957115)