Developed a scalable distributed training workflow for Direct Preference Optimization (DPO) using Llama 3.1 within the pytorch/torchtune repository, focusing on enabling efficient multi-node fine-tuning of large language models. The work involved designing and implementing a distributed DPO training recipe that improved throughput and resource utilization, supporting faster experimentation at scale. Leveraging deep learning and distributed systems expertise, the developer integrated pipeline adjustments to facilitate robust multi-node experiments. Utilizing Python and PyTorch, the solution established a foundation for scalable LLM fine-tuning, allowing for higher performance and more efficient resource management in large-scale machine learning workflows without addressing bug fixes this period.