Dataset · nvidia
Dataset for RL training of scientific reasoning
nvidia/Nemotron-RL-Science-v1
Physics, biology, and chemistry problems from NVIDIA with verifier settings for reinforcement learning.
What's inside
An English text dataset of 100K to 1M records from NVIDIA, created for training natural science reasoning models using RL. Each record contains an open-question physics, biology, or chemistry problem, reference answer, and verifiable reward environment configuration (agent prompt, verifier data, and answer extraction template).
What you can do with it
- Training reasoning language models (RL with verifiable rewards) on scientific tasks
- Fine-tuning models to solve open-ended physics, chemistry, and biology questions
- Setting up automated answer verification pipelines in scientific domains
- Tasks
- Text generation
- Languages
- English
- Size
- 100K–1M rows
- Downloads this month
- 1.1K
- Likes
- 18
How to load it
from datasets import load_dataset
ds = load_dataset("nvidia/Nemotron-RL-Science-v1")