Worked on enhancing distributed checkpointing reliability for the NVIDIA/Megatron-LM repository, focusing on improving fault tolerance and error reporting in asynchronous checkpointing workflows. Addressed error propagation by catching and wrapping exceptions on worker nodes, refactoring the retrieve_write_results function to return wrapped exceptions, and broadcasting failure statuses across distributed processes. These changes reduced silent failures and improved observability, enabling faster recovery during distributed training. Utilized Python and applied expertise in asynchronous programming, distributed systems, and error handling to deliver a robust feature that surfaces write-time failures and ensures more reliable checkpointing in large-scale machine learning environments over the course of the month.