Unverified50% confidenceFactExact time
MiniMax在长上下文领域据公开信息采用了类似「闪电注意力」（Lightning Attention）的线性复杂度机制
1
Sources
50%
Confidence
Long-term
Relevance
7/17/2026
First Seen
Sources
Related Claims
UnverifiedThe root cause of attention dilution lies in Softmax normalization: the more tokens in a sequence, the effective attention weight allocated to each position decreases approximately inversely proportional to the sequence length.67% similarUnverifiedSSMs have a linear time complexity advantage in processing long sequences, whereas the attention computation complexity of standard Transformers is quadratic with respect to sequence length.62% similarUnverified长上下文场景的计算复杂度以序列长度的平方级增长，使成本改善速度远慢于短上下文场景62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/538981API
curl https://kongchang.com/api/v1/knowledge/claims/538981MCP
get_claim(id=538981)