Worked on cross-repository memory accounting enhancements for TPU v4i devices, focusing on Intel-tensorflow/tensorflow and Intel-tensorflow/xla. Addressed device management challenges by standardizing memory handling, treating TPU v4i as an alias of TPU v4 lite within the GetDeviceMemoryInBytes function. This approach ensured that memory allocation and reporting for TPU v4i matched the established behavior for TPU v4 lite, improving allocation accuracy and device observability. Utilized C++ for backend development, implementing both a new feature and a bug fix to reduce memory-related risks and support more reliable capacity planning for deployments using newer TPU configurations in production environments.