PPL
Chen Liang
Papers (7)
The Carbon Footprint of Machine Learning Training Will Plateau, Then ShrinkSlimMoE: Structured Compression of Large MoE Models via Expert Slimming and DistillationGRIN: GRadient-INformed MoETraining-Free Looped Transformers$φ$-Balancing for Mixture-of-Experts TrainingTurning Stale Gradients into Stable Gradients: Coherent Coordinate Descent with Implicit Landscape Smoothing for Lightweight Zeroth-Order OptimizationThe Efficiency Gap in Byte Modeling