During November 2025, this developer contributed to the PaddlePaddle/PaddleFormers repository by implementing data-parallel training support for Mixture of Experts within the Zero-Cost Checkpointing framework. Using Python and leveraging deep learning and distributed systems expertise, they enabled scalable large-model training by combining expert-parallelism with memory-efficient checkpointing. Their work involved updating state_dict loading and handling to support expert-parallel weights and global expert ID management, ensuring correct weight assignment during distributed training. This addition allowed for more flexible and efficient experimentation with model parallelism, addressing the challenges of memory usage and scalability in modern machine learning workflows for large models.