Skip to content

Dataset · nvidia

SFT data for Nemotron-Cascade-2

nvidia/Nemotron-Cascade-2-SFT-Data

Text SFT dataset for training Nemotron-Cascade-2, including mathematics and science.

What's inside

A set of text SFT examples used in training Nemotron-Cascade-2; size ranges from 10M to 100M examples. Math prompts were taken from Nemotron-Cascade-1-SFT and Nemotron-Math-v2, and answers were generated by DeepSeek-V3.2, DeepSeek-V3.2-Speciale, and GPT-OSS-120B. For proofs, prompts were taken from Nemotron-Math-Proofs-v1 and generated by DeepSeek-V3.2-Speciale; the set also includes science prompts.

What you can do with it

  • Fine-tune LLMs on mathematical and scientific reasoning
  • Reproduce SFT for Nemotron-Cascade-2
  • Compare responses of different teacher models during distillation
Size
10M–100M rows
Downloads this month
17.5K
Likes
79

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("nvidia/Nemotron-Cascade-2-SFT-Data")