Developed and integrated Torch export functionality into the training pipeline for the pytorch/ao repository, enabling compatibility with Torch 2.5 and above while supporting conditional usage based on the installed Torch version. This work involved migrating the training intermediate representation to leverage Torch’s export API, streamlining model export and deployment processes. By reducing manual conversion steps and future-proofing the workflow against version changes, the update improved reliability and efficiency for quantization and model training. The implementation utilized Python and PyTorch, with a focus on machine learning workflows and close collaboration with CI/CD systems to ensure a stable and maintainable rollout.