Verified90% confidenceFactTime unknown
DeepSeek V4采用自研的分层压缩注意力策略，设计了CSA（压缩吸收注意力）和HCA（重度压缩注意力）两种注意力模式交替使用
6
Sources
90%
Confidence
Long-term
Relevance
5/31/2026
First Seen
Sources
DeepSeek V4编码实测：榜单第一Kimi翻车，Claude稳居最强
bilibiliAGI_Ananas
Related Entities
Related Claims
UnverifiedDeepSeek-V3.2-Exp首次引入自研的DeepSeek Sparse Attention（DSA）稀疏注意力技术80% similarUnverifiedDeepSeek-V2提出MLA（Multi-head Latent Attention），通过低秩压缩将KV Cache压缩至MHA的5%左右79% similarUnverifiedDeepSeek在多头潜在注意力（MLA）和FP8混合精度训练方向上的创新将KV Cache内存占用压缩至标准多头注意力的5%-13%77% similarVerifiedMLA（多头潜在注意力）由DeepSeek团队在DeepSeek-V2中首次提出，通过将Key-Value对压缩到低维潜在空间大幅降低KV Cache显存占用76% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/13855API
curl https://kongchang.com/api/v1/knowledge/claims/13855MCP
get_claim(id=13855)