[KongchangAI]
Unverified50% confidenceSolutionExact time

分散损失是一种在训练目标中显式加入约束项、鼓励嵌入向量在空间中尽可能散开的正则化方法,用于对抗小语言模型中的嵌入坍缩

1
Sources
50%
Confidence
Long-term
Relevance
7/7/2026
First Seen

Sources

Related Claims

Cite This Claim

Stable URI
https://kongchang.com/claim/189245
API
curl https://kongchang.com/api/v1/knowledge/claims/189245
MCP
get_claim(id=189245)