MTraining: Distributed Dynamic Sparse Attention for Efficient Ultra-Long Context Training

Open in new window