Accelerating NLP Model Training and Deployment with PyTorch
Accelerating NLP Model Training and Deployment with PyTorch


NLP models are big – they take a lot of time and money to train and can be hard to deploy. You’re going to want all the help you can get to reduce costs, improve agility, and enable production deployments. In this session you’ll see how massive scale products like Microsoft Office are using PyTorch, ONNX Runtime, and Azure Machine Learning to significantly accelerate their training and inferencing to increase agility and reduce costs. You’ll learn how you can do the same for your own NLP scenarios


Prasanth Pulavarthi is Principal Program Manager for the AI Frameworks team at Microsoft. His team works on making ML practitioners and engineers more efficient through optimized libraries, tools, and communities. ONNX Runtime ( is an open-source engine from his team that integrates with TensorFlow, PyTorch, and other frameworks to accelerate inferencing and training on a variety of cloud and edge hardware.
Prasanth is also the Co-Founder of ONNX (, the open standard for machine learning interoperability. ONNX is now a graduate projected in Linux Foundation Artificial Intelligence. He serves on the ONNX Steering Committee and is actively involved in the ONNX community.