Xinrui Mao enhanced the XPU continuous integration (CI) test coverage for the PaddlePaddle/ERNIE repository, focusing on improving reliability and reducing maintenance costs. By developing new Python-based tests for dense and Mixture-of-Experts (MoE) XPU models, Xinrui addressed existing coverage gaps and aligned the CI process with ongoing ERNIE model development. The work included removing outdated LoRA training test artifacts, which streamlined the test suite and lowered future maintenance overhead. Leveraging skills in CI/CD, machine learning, and testing, Xinrui’s contributions improved the stability and maintainability of ERNIE’s CI pipeline, supporting scalable validation for future XPU model expansions.