Unverified60% confidenceFactExact time
Switch Transformer, proposed in 2021, simplified the number of experts routed to per token to 1.
2
Sources
60%
Confidence
Long-term
Relevance
7/24/2026
First Seen
Sources
Related Claims
UnverifiedSwitch Transformer将专家数量扩展至2048个，2022年谷歌GLaM模型进一步验证了MoE在多任务泛化上的优势80% similarUnverifiedSwitch Transformer was proposed by William Fedus et al. in 2021, and was the first to scale MoE to the trillion-parameter range71% similarUnverifiedSwitch Transformer was released in 2021, and Mixtral 8×7B was released in 2023.71% similarUnverifiedGshard（2020）首次实现了万亿参数MoE模型的分布式训练，Switch Transformer（2022）将每个token的路由选择收敛为单一专家70% similarUnverifiedTransformer于2017年提出后被快速引入时序预测领域70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/604999API
curl https://kongchang.com/api/v1/knowledge/claims/604999MCP
get_claim(id=604999)