GW-MoE: Resolving Uncertainty in MoE Router with Global Workspace Theory

Documenti analoghi

• Turn Waste into Worth: Rectifying Top-$k$ Router of MoE
  di: Zeng, Zhiyuan, et al.
  Pubblicazione: (2024)
• Stabilizing MoE Reinforcement Learning by Aligning Training and Inference Routers
  di: Ma, Wenhan, et al.
  Pubblicazione: (2025)
• MoE-PHDS: One MoE checkpoint for flexible runtime sparsity
  di: Hannah, Lauren. A, et al.
  Pubblicazione: (2025)
• Is Retraining-Free Enough? The Necessity of Router Calibration for Efficient MoE Compression
  di: Hyeon, Sieun, et al.
  Pubblicazione: (2026)
• KBVQ-MoE: KLT-guided SVD with Bias-Corrected Vector Quantization for MoE Large Language Models
  di: Xu, Zukang, et al.
  Pubblicazione: (2026)