Nemotron-Cascade 2: Post-Training LLMs with Cascade RL and Multi-Domain On-Policy Distillation

Zihan Liu's public homepage describes him as a senior research scientist at NVIDIA ADLR working on LLM reasoning, post-training, and retrieval-augmented generation.

OpenReview lists Grace Lam as a researcher at NVIDIA after prior work at Google, with study at Stanford and MIT. NVIDIA public pages connect her to Cosmos Policy for robot control and Nemotron-Cascade 2 for LLM post-training.

Boxin Wang is a senior research scientist at NVIDIA working on practical and scalable large language models, including trustworthiness, multimodal language modeling, and retrieval-augmented generation. He earned a Ph.D. in Computer Science from the University of Illinois Urbana-Champaign.

Zhuolin Yang's personal page says he is a Research Scientist at NVIDIA ADLR whose interests include trustworthy machine learning, multimodal LLMs, and reasoning.

Yangyi Chen is a research scientist at NVIDIA. His homepage says he works on LLM post-training and LLM-based coding agents, and it lists him as a Nemotron-Cascade 2 coauthor.

Wenliang Dai's homepage identifies him as a research scientist at NVIDIA. OpenReview lists prior study at HKUST, University College London, and the University of Nottingham.

Public NVIDIA and HKUST sources identify Renjie Pi as a researcher in multimodal learning and large language models.

NVIDIA identifies Mohammad Shoeybi as a senior research scientist who manages the NLP team within its Applied Deep Learning Research group; the same profile states he received a PhD from Stanford University in 2010.

Jiafan He's public homepage identifies him as a UCLA PhD student advised by Quanquan Gu, with research focused on machine learning, reinforcement learning, and optimization.

Alexander Bukharin is an applied research scientist at NVIDIA working on large language models, with public NVIDIA materials highlighting reinforcement learning and model robustness.

NVIDIA's public author page identifies Chankyu Lee as a research scientist on the Applied Deep Learning research team, focusing on NeMo LLMs and retrieval-augmented generation. NVIDIA public report pages also list him as a coauthor of Nemotron-Cascade.

Dongfu Jiang is a Ph.D. student at the University of Waterloo affiliated with the Vector Institute and TIGER Lab. His homepage lists research interests in LLM and VLM post-training, multimodal evaluation, and agentic tool-use systems.

Public NVIDIA sources identify Bryan Catanzaro as Vice President of Applied Deep Learning Research and maintain a research profile page for him.

Canonical link

Liu, Zihan

Lam, Grace

Wang, Boxin

Yang, Zhuolin

Chen, Yangyi

Dai, Wenliang

Pi, Renjie

Shoeybi, Mohammad

He, Jiafan

Bukharin, Alexander

Lee, Chankyu

Jiang, Dongfu

Catanzaro, Bryan