NeMo RL
AI overview
Sign in and the AI will write an overview from our coverage.
Headlines · 4
- NVIDIA Dynamo targets reinforcement-learning rollout bottlenecks
NVIDIA Technical Blog describes Dynamo as a serving layer between reinforcement-learning orchestration and inference workers, with KV-aware routing, ModelExpress weight updates and worker-failure recovery. It says Microsoft AI, CoreWeave, Cognition, Prime Intellect and H Company use components of the stack, while CoreWeave reports up to 15× faster weight updates and H Company reports up to 20% lower training-step time and 5% lower GPU-hour use; these performance figures are vendor claims without independent benchmarks in the supplied report.
NVIDIA Technical Blog · 🔥 3 - NVIDIA Introduces Nemotron for Science for Scientific Agents
NVIDIA introduced Nemotron for Science, a package that combines the open Nemotron 3.5 Super VL model, scientific evaluations, the LitmusBench v0.2 chemistry dataset and post-training recipes for domain-specific scientific agents. LitmusBench v0.2 contains nearly 50,000 questions for evaluating and training models on chemistry reasoning, including molecular properties and experimental drug-discovery data. A separate post-training guide describes building a reinforcement-learning environment with NVIDIA NeMo Gym, applying reinforcement learning with NeMo RL and measuring improvement through molecular-similarity scoring. Nemotron 3.5 Super VL and LitmusBench v0.2 are available on Hugging Face, and the model was optimized to run locally on a 128GB NVIDIA DGX Spark.
NVIDIA Newsroom · 🔥 0 - NVIDIA Open-Sources Reshard for Faster RL Weight Refit
NVIDIA Technical Blog introduced Reshard, a collective communication operation for transferring tensors between arbitrary DTensor source and destination layouts. An open-source reference implementation, NCCL M2N Reshard, transfers weights from M training GPUs to N inference GPUs and is available through NVIDIA's nccl-extensions repository. NVIDIA reports that its NeMo RL implementation achieved 2.7x to 5.7x speedups over gather-and-broadcast across benchmarks on DGX H100 and DGX GB200 systems.
NVIDIA Technical Blog · 🔥 0 - 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.
Hugging Face · Papers · 🔥 0
Experience and discussion from the community
Share my NeMo RL experienceAsk about NeMo RL
Nobody has shared their experience with NeMo RL yet.