Unverified50% confidenceFactExact time
MoE架构在长上下文场景下存在能力衰减问题，因路由机制在长序列下决策不稳定，连DeepSeek也未完全解决
1
Sources
50%
Confidence
Long-term
Relevance
7/10/2026
First Seen
Sources
Related Claims
UnverifiedLow-bit quantization of MoE models can easily cause routing bias, which in turn leads to sudden degradation in certain capability dimensions.72% similarUnverifiedThe application of the Mixture of Experts (MoE) architecture has broken the traditional trade-off between parameter scale and inference speed.70% similarUnverifiedThe MoE architecture faces load balancing challenges, and researchers typically introduce auxiliary loss functions to mitigate the expert collapse problem, while teams such as DeepSeek have explored auxiliary-loss-free balancing strategies.68% similarUnverifiedDuring the Expert routing stage of MoE models, dispatching different threads to different Experts causes severe Warp divergence issues.68% similarUnverified简单堆砌上下文长度并不能线性提升长任务性能，因存在中间丢失问题67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/465618API
curl https://kongchang.com/api/v1/knowledge/claims/465618MCP
get_claim(id=465618)