NVIDIA & Marlowe: Post-training language agents with NeMo RL and NeMo Gym

Abstract

This talk covers modern post-training and reinforcement learning techniques used to train effective language models and agents. An overview of NVIDIA NeMo and Nemotron will be covered with a focus on NeMo-RL, NeMo-Gym and how these frameworks are used to train the latest Nemotron models.

Special Guest Speakers | Christian Munley & Shibani Likhite

Christian Munley is an applied research scientist at NVIDIA focused on post-training and alignment of language models including Nemotron. Previously, he was a solutions architect at NVIDIA, supporting partners to leverage and accelerate computing in their research and products. His education is in physics and computer science, where his diverse research topics include biophysics and LLMs for HPC.

Shibani Likhite is a Solutions Architect at NVIDIA, working closely with cloud service providers on large-scale post-training, alignment, and high-throughput inference using NVIDIA technologies. She has worked across multiple industries to enable partners in building and deploying efficient post-training and inference workflows. Prior to joining NVIDIA, she earned her Master’s degree in Computer Science from UC San Diego, with a focus on Artificial Intelligence.