Unverified50% confidenceFactExact time
Multi-Head Latent Attention（MLA）由 DeepSeek-V2 于2024年提出
1
Sources
50%
Confidence
Long-term
Relevance
7/14/2026
First Seen
Sources
15岁少年从零训练200M MoE语言模型：完整架构与实战拆解
redditr/learnmachinelearning7/12/2026
Related Claims
UnverifiedDeepSeek-R1模型于2025年初发布并引发全球关注67% similarUnverified2024年底至2025年初，DeepSeek-V3和DeepSeek-R1相继发布66% similarUnverifiedDeepSeek-V3于2024年底发布64% similarUnverifiedDeepSeek-V2 was released in May 2024 and uses MoE (Mixture of Experts) architecture.64% similarUnverifiedDeepSeek-V4-Pro API即日起至2026年5月5日享受75折优惠64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/503162API
curl https://kongchang.com/api/v1/knowledge/claims/503162MCP
get_claim(id=503162)