SYMI: Efficient Mixture-of-Experts Training via Model and Optimizer State Decoupling

Open in new window