NVIDIA's NeMo-DCR: bit-exact delta refit for trillion-parameter agentic RL
NVIDIA researchers posted NeMo-DCR (Delta-Compressed Refit), a method that synchronizes policy updates between training and rollout clusters by sending only weight changes while remaining bit-exact against a dense refit. The paper reports that about 1% of BF16 training weights change stored values per step, and that at 3% and 5% change rates refits of 30B–1T models run 12–40x faster than a transport-only full-checkpoint reference; a 1T relay-tree refit at 3% takes 150 seconds versus 87.5 minutes to move a full checkpoint between two AWS regions. The code is open-sourced in NVIDIA NeMo RL PR #2444.
Cutting cross-cluster weight sync from roughly 90 minutes to minutes could make continuous training-and-rollout loops practical at trillion-parameter scale.
Computer Science > Distributed, Parallel, and Cluster Computing
arXiv:2610.08430 (cs)
[Submitted on 6 Oct 2026]
Title:NeMo-DCR: Bit-Exact Delta-Compressed Refit for Scalable Agentic RL at Trillion-Parameter Scale
Authors:Songlin Jiang, Zhiyu Li, Terry Kong, Yu Yao, Youngeun Kwon, Bernard Nguyen, Ashwath Aithal, Mario Di Francesco
View a PDF of the paper titled NeMo-DCR: Bit-Exact Delta-Compressed Refit for Scalable Agentic RL at Trillion-Parameter Scale, by Songlin Jiang and 7 other authors