Verified65% confidenceFactExact time
Google's Switch Transformer, released in 2021, alleviates load imbalance by simplifying routing to a Top-1 selection and introducing an expert capacity mechanism.
3
Sources
65%
Confidence
Long-term
Relevance
7/14/2026
First Seen
Sources
PrismML突破：270亿参数大模型如何跑进iPhone？
Related Claims
Unverified2022年Google的《Switch Transformers》将MoE中选Top-K专家简化为选Top-1专家，并通过辅助负载均衡损失函数解决专家坍塌问题80% similarUnverifiedGoogle于2022年发布的Switch Transformer证明MoE架构在相同计算预算下收敛速度可达稠密模型的4-7倍80% similarVerifiedGoogle validated the Mixture of Experts (MoE) architecture at scale in Switch Transformer.76% similarUnverifiedConformer（Convolution-augmented Transformer）架构由Google在2020年提出，结合了Transformer的全局自注意力机制和卷积神经网络的局部特征提取能力75% similarVerifiedThe Transformer architecture was proposed by Google in 2017.74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/507685API
curl https://kongchang.com/api/v1/knowledge/claims/507685MCP
get_claim(id=507685)