Worked on the vllm-project/vllm-ascend repository to optimize continuous integration performance by introducing model caching within the test framework. This approach enabled models to be loaded once and reused across multiple test cases, reducing redundant loading and improving overall test efficiency. Using Python and focusing on model optimization and testing, the solution decreased CI runtime for five test classes from 33 to 23 minutes, resulting in measurable cost and throughput improvements. The work aligned with the vLLM Developer Test Framework Model Reuse initiative, referencing the v0.23.0 baseline and main repository commits to ensure consistency and maintainability.