Worked on the pytorch/pytorch repository to deliver complex data type support for tensor communication in the ProcessGroupGloo module, enabling distributed training with complex-valued tensors on both CPU and GPU. The implementation involved developing a new shared utilities module in C++ and Python to handle complex data types, as well as updating and expanding the test suite to ensure robust validation of the new feature. By aligning Gloo’s capabilities with those of ProcessGroupNCCL, this work improved interoperability and deployment flexibility for teams using PyTorch distributed systems, reducing integration friction for projects adopting complex-valued models in their machine learning workflows.