Worked on upgrading the continuous integration infrastructure for the pytorch/pytorch repository by migrating CI workloads from an MI300 to an MI325 GPU cluster. This migration focused on optimizing GPU resource allocation for testing, resulting in improved CI throughput and faster feedback cycles. The approach preserved all existing testing plans and workflows, ensuring no test regressions during the transition. Leveraged skills in CI/CD, GPU computing, and configuration management using Python and YAML to implement the changes. The work addressed the need for more efficient resource utilization in large-scale machine learning projects, enhancing the reliability and speed of the PyTorch CI process.