Albert Malewski enhanced the reliability of distributed data parallelism on ROCm within the pytorch/pytorch repository by expanding and stabilizing test coverage for critical tensor operations and backward pass scenarios. He focused on unskipping previously disabled ROCm-specific tests, enabling earlier detection of regressions and improving CI feedback for ROCm-enabled distributed training. Using Python, PyTorch, and distributed systems expertise, Albert addressed test regressions and reduced flaky failures in the continuous integration pipeline. His work deepened validation coverage for ROCm architectures, ensuring safer releases and more robust cross-architecture support, while collaborating closely with peers to maintain high standards in automated testing workflows.