training optimization

3 papers

• Advancing Expert Specialization for Better MoE
• Advancing Expert Specialization for Better MoE
• Momentum-SAM: Sharpness Aware Minimization without Computational Overhead