Unverified50% confidenceFactExact time
MLP层通常由两个线性变换和一个激活函数（如GELU）组成，中间隐藏维度往往是输入维度的4倍
1
Sources
50%
Confidence
Long-term
Relevance
8/26/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedFlattening a matrix and feeding it to an MLP results in a non-fixed input dimension due to variable row counts, and breaks permutation invariance.66% similarUnverifiedResearch (Dettmers et al., 2022) has shown that specific hidden dimensions in Transformer models consistently exhibit abnormally large activation values across all tokens and layers, with the phenomenon becoming more pronounced in models exceeding 6.7B parameters.59% similarUnverified正交变换保持向量的L2范数和向量间夹角不变，正交矩阵的条件数恒为158% similarUnverified该研究生的核心成果是建立了函数在分形集合上的集中程度与其傅里叶变换之间的定量约束，证明两者不可能同时被限制在过于稀疏的分形集合上57% similarUnverifiedGPTQ（2022）引入逐层Hessian矩阵近似，通过最小化权重量化前后的输出重建误差优化量化参数56% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/804468API
curl https://kongchang.com/api/v1/knowledge/claims/804468MCP
get_claim(id=804468)