Artificial Intelligence1 min read

Post-Training NVIDIA Nemotron to Increase Accuracy in Your Scientific Domain

By · Published by Everything Blog

In short

The open-source scientific suite integrates Nemotron 3.5 Super VL with specialized benchmarks to power autonomous hypothesis testing and simulation workflows. Researchers adapt the base model using the NVIDIA NeMo Gym infrastructure alongside the NeMo RL framework for verifiable reinforcement learning. In a drug-metabolism demonstration, the model reasons step-by-step to predict metabolites based on parent compounds and specific enzymes. The pipeline then evaluates these outputs automatically by measuring Tanimoto similarity against verified molecular fingerprints.

Key points

  • Researchers can refine domain-specific models by combining NVIDIA NeMo Gym with the NeMo…: Researchers can refine domain-specific models by combining NVIDIA NeMo Gym with the NeMo RL post-training framework.
  • Reinforcement learning with verifiable rewards eliminates the necessity for manual human…: Reinforcement learning with verifiable rewards eliminates the necessity for manual human preference labeling in specialized domains.
  • Group relative policy optimization combined with low rank adapters prompts models to reas…: Group relative policy optimization combined with low rank adapters prompts models to reason step-by-step through complex chemical tasks.

Sources checked by our analysis pipeline: 1

Agentic AI is reshaping scientific research and innovation. Researchers communicate complex scientific hypotheses in natural language and let autonomous agents independently plan tasks, execute simulations, evaluate results, and self-correct across multiple iterations. Scientific agents are taking on longer, more complex, and increasingly visual workflows.

Accuracy and inference efficiency matter at every step. Without verification, errors compound across the workflow, while latency and compute costs multiply with continued operation.

NVIDIA Nemotron for Science, a collection of models, datasets, tools, and recipes for accelerating agentic scientific workloads, addresses these hurdles. It includes Nemotron 3.5 Super VL—an open, customizable, multimodal model for these specialized workflows—and the NVIDIA Nemotron Litmus-Bench V2.0, an open source chemistry-reasoning dataset and benchmark that was designed to train and evaluate LLMs and agentic AI models on molecular properties. Out of the box, Nemotron delivers frontier level accuracy for scientific benchmarks (see Figure 1 below), along with the fastest time to solution for scientific coding.

An important benefit of Nemotron models is NVIDIA post-training support. You can customize models for your specific research domain to increase accuracy.

This blog provides…

Artificial Intelligence