NVIDIA NeMo-DCR cuts 1T RL weight sync time

Researchers at NVIDIA introduced NeMo-DCR, a delta-compressed refit system designed for cross-cluster agentic reinforcement learning at trillion-parameter scale. By sending only changed weights using direct projection, mixed XOR/overwrite encoding, and native loader placement, the system reduces cross-cluster weight synchronization times for a 1T model from 87.5 minutes to 150 seconds. Open article →