Worked on the allenai/OLMo-core repository to enable supervised finetuning (SFT) support, integrating end-to-end tooling and training scripts for Olmo-2 and Olmo-3 models. Enhanced the Hugging Face model conversion process to preserve tokenizer and generation configurations, facilitating smoother deployment of fine-tuned models. Improved training robustness by refining checkpoint loading mechanisms and introducing clearer error handling. Developed robust IO utilities for handling both local and remote paths, reducing friction in data and model loading workflows. Delivered comprehensive launch scripts and updated documentation, leveraging Python and deep learning frameworks to streamline model training and deployment within the OLMo-core ecosystem.