Worked on enhancing the Lightning-AI/litgpt repository by improving the stability of Gemma 3 1B model loading and enabling checkpoint conversion. Addressed deployment reliability by adjusting the intermediate size configuration, which reduced manual intervention and potential errors during model initialization. Developed a new copy function to streamline the conversion of Gemma 3 checkpoints, facilitating a more efficient workflow for model updates. All changes were implemented in Python, leveraging skills in checkpoint conversion and model configuration. The work focused on targeted configuration adjustments and code enhancements, resulting in a more robust and maintainable process for deploying and updating large language models.