Modify relevant PyTorch modules to internally use and support the GroupedTensor type, ensuring transparent integration for grouped MoE operations.