Joint MoE Scaling Laws: Mixture of Experts Can Be Memory Efficient

Open in new window