Nemotron-Science-v1 scientific reasoning dataset is a synthetic scientific reasoning dataset released by NVIDIA in 2025, designed to enhance the reasoning and problem-solving capabilities of large language models in the scientific domain.

The dataset contains 226,334 samples, with a total disk size of approximately 2.5 GB, in JSONL format, covering two subsets: MCQA and RQA. The data uses GPT-OSS-120B to generate reasoning traces and synthetic questions, suitable for supervised training and data augmentation.

Dataset Composition

• MCQA: A total of 174,155 samples. This is an improved version of the STEM portion of Nemotron-Post-Training-Dataset-v1, using GPT-OSS-120B to generate GPQA-style questions and reasoning traces, aimed at enhancing the reasoning capabilities of large language models in the scientific domain.
• RQA: A total of 52,179 samples. Composed of synthetic chemistry questions, also used to enhance the reasoning capabilities of large language models in the scientific domain.