During January 2025, S. Bose developed multi-GPU training support for the APPFL/APPFL repository, focusing on scalable deep learning workflows. By refactoring device handling and introducing utilities to parse device strings, S. Bose enabled models to be flexibly assigned to specific GPUs, including multi-GPU configurations using PyTorch’s nn.DataParallel. This work laid the foundation for distributed training across multiple GPUs, addressing the need for scalable, GPU-accelerated model training in distributed systems. The implementation, written in Python and leveraging PyTorch, improved maintainability and flexibility of the codebase. The depth of the changes reflects a strong understanding of GPU computing and distributed architectures.