Worked on the apple/axlearn repository to improve the reliability of first-time training runs by addressing initialization issues in the SpmdTrainer component. Used Python and machine learning expertise to implement a fix that initializes a new trainer state with a provided prng_key when no checkpoint is available, ensuring smoother startup behavior. Enhanced project documentation to clarify the _prepare_training process, reducing ambiguity for new users and supporting reproducibility. This work focused on stabilizing the onboarding experience and laying groundwork for future enhancements, demonstrating attention to both code quality and user experience through targeted bug fixes and clear, accessible technical documentation.