Worked on the OpenBMB/VoxCPM repository to enhance the stability and reliability of the inference workflow, focusing on Lora and FT testing. Addressed a parameter naming inconsistency in Python inference scripts, ensuring that the correct arguments are passed during function calls and aligning the test scripts with the updated API. This debugging effort reduced false negatives in automated tests and improved the overall robustness of experiment results. Emphasized Python scripting and thorough testing practices to eliminate sources of test drift, resulting in a more dependable workflow for model evaluation. No new features were introduced during this period, prioritizing reliability improvements.