Unverified50% confidenceFactExact time
Longformer and BigBird use fixed local windows combined with global token sparse patterns, reducing complexity to O(N) but sacrificing direct connections between arbitrary positions.
1
Sources
50%
Confidence
Long-term
Relevance
10/10/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedLongformer、BigBird通过局部滑动窗口注意力加少量全局token的混合注意力策略，将计算复杂度从O(n²)降至O(n)80% similarUnverifiedLongformer由Allen AI提出，采用滑动窗口局部注意力加全局注意力token的组合策略，将复杂度从O(n²)降至O(n)64% similarUnverified滑动窗口注意力（如Mistral采用的方案）通过限制每个token只关注固定窗口内的历史，将复杂度从O(n²)降至O(n·w)64% similarUnverifiedSliding Window Attention reduces the standard Transformer's O(n²) complexity to O(n·w), but the model cannot directly see information outside the window.62% similarUnverifiedA larger context window is not equivalent to stronger reasoning ability; the two are capability metrics on different dimensions62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/995073API
curl https://kongchang.com/api/v1/knowledge/claims/995073MCP
get_claim(id=995073)