GPU Scheduling Optimization for Distributed Deep Learning: Addressing Blocking and Fragmentation in English Teaching AI Scenarios | Synapse