Developed Decode Context Parallel (DCP) support for distributed training in the ROCm/aiter repository, focusing on backend development and distributed systems using Python and PyTorch. The work involved implementing group initialization, lifecycle management, and communication primitives, all integrated into the existing model-parallel initialization flow. This approach reduced setup complexity for large-scale training runs and enabled new distributed training configurations. Additionally, code cleanup was performed to remove obsolete workarounds as part of enabling DCP. The changes established a foundation for more efficient experimentation and automation in distributed environments, supporting scalable model training without introducing new bugs during the development period.