Key highlights
- NVIDIA launched Nemotron for Science, combining its Nemotron 3.5 Super VL open model with tools for researchers.
- The release includes LitmusBench v0.2, an open chemistry dataset with nearly 50,000 questions.
- NVIDIA said Nemotron 3.5 Super VL and LitmusBench v0.2 are available now on Hugging Face.
NVIDIA (NASDAQ: NVDA) has introduced NVIDIA Nemotron for Science, a package the company said is designed to help researchers adapt its NVIDIA Nemotron 3.5 Super VL open model for life sciences and chemistry work.
The company described Nemotron for Science as combining the Nemotron 3.5 Super VL model with scientific evaluations, a new open chemistry dataset and post-training recipes intended to help researchers build domain-specific science agents.
This article is a journalistic opinion piece which has been written based on independent research. It is intended to inform investors and should not be taken as a recommendation or financial advice.
As part of the release, NVIDIA introduced LitmusBench v0.2, which it described as an open dataset with nearly 50,000 questions for training and evaluating language models on chemistry reasoning. NVIDIA said the dataset draws on molecular properties and experimental drug discovery data, and is meant to help improve models’ ability to predict how molecules behave, including solubility and how the human body processes them.
NVIDIA also outlined how it expects researchers to use post-training to adapt Nemotron models to specialised tasks, including chemistry experiments and protein design, and to integrate the resulting models into agent workflows that can interpret scientific information, generate code and call computational tools.
The company listed several organisations it said are using or integrating Nemotron 3.5 Super VL for science applications, including life sciences companies Lila Sciences, Terray Therapeutics and Therna, as well as research institutions such as the U.S. Department of Energy’s Argonne National Laboratory and Jfclich Supercomputing Centre.
In examples provided in the release, NVIDIA said Terray Therapeutics post-trained Nemotron to extract reaction information from scientific diagrams and recovered about 76% of annotated reactions on the RxnScribe test under the benchmark’s matching criteria. NVIDIA also said Lila Sciences used Nemotron to generate laboratory protocols for automated labs and reported that post-training with laboratory feedback nearly doubled accuracy for making quantum dots to 61%.
NVIDIA said Nemotron 3.5 Super VL and the LitmusBench v0.2 dataset are available now on Hugging Face. It also said the Nemotron 3.5 Super VL model was optimised to run locally on a 128GB NVIDIA DGX Spark system.
The original press release is available on stockhouse.com
s-13v4_h-14v4_i-15v4
