Event ·Official first-hand

MetaNet：任务条件专家动态分配

MetaNet提出了一种支持集控制器，可为每层预测专家保留阈值和有界路由偏差。在DeepSeek-MoE-16B-Chat模型上，保守设置激活专家数减少40%，MMLU准确率0.489；激进设置激活专家减少62%，准确率下降3.7个百分点。MMLU训练的控制器无需重训即可迁移到C-Eval，激活专家减少52%，准确率达0.386。

Current read

MetaNet让MoE模型根据任务难度动态分配专家，大幅减少计算量同时保持准确率。

2 sources73° AI heatLast updated 2026/8/27 06:01:46

Evidence

2 sources

Primary sourcearXiv: DeepSeek

2026/8/27 06:01:46View source

Related source 1elvis

2026/8/25 23:13:33View source

Conflict check

Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.

Open the story