Worked on the Lightning-AI/pytorch-lightning repository to enhance distributed training reliability for TPU/XLA-backed environments. Developed compatibility for retrieving distributed training parameters such as global_ordinal, local_ordinal, and world_size under XLA, ensuring correct parameter propagation during distributed runs. The implementation leveraged Python and PyTorch, conditionally utilizing torch_xla.runtime to maintain support for newer torch_xla versions and prevent breakages in updated environments. Expanded the test suite to verify the new retrieval behavior across supported configurations, increasing test coverage and confidence. This work reduced setup friction for users and contributed to more robust, scalable distributed machine learning workflows using XLA hardware.