Worked on the PrimeIntellect-ai/prime-rl repository to deliver SFT LoRA fine-tuning and training workflow enhancements focused on efficient model adaptation. Developed support for low-rank adaptation (LoRA) in supervised fine-tuning, introducing new configuration files for both training and resuming processes. Integrated a MultiRunManager to streamline adapter export and manage multiple experimental runs, improving resource utilization during deep learning experiments. Refactored the training workflow for better structure and readability, including renaming and clarifying helper functions. Applied code quality improvements such as ruff formatting and added documentation to enhance maintainability. Utilized Python and machine learning frameworks throughout the development process.