During October 2025, Can Lin focused on enhancing the vllm-project/vllm-ascend repository by developing end-to-end tests for the InternVL model. Leveraging Python and YAML, Can implemented tests that validate single-image inference and parameterize across multiple InternVL versions, ensuring consistent regression detection as the model evolves. The work included updating the CI/CD workflow so these tests run automatically during pull requests and builds, improving validation coverage and accelerating feedback cycles. By establishing a robust model testing approach, Can addressed the need for reliable release validation, contributing depth in both CI/CD integration and automated end-to-end testing within the project’s infrastructure.