Dataset · nvidia
Math problems with step-by-step reasoning
nvidia/Nemotron-SFT-Math-v4
A large English-language set of complex math problems with reasoning chains from DeepSeek-V4-Pro for SFT training.
What's inside
Inside are English text math problems and detailed reasoning trajectories (solutions) generated by the DeepSeek-V4-Pro model in High inference mode. The problem statements themselves are taken from the nvidia/Nemotron-Math-v2 dataset, which was compiled based on materials from the Art of Problem Solving (AoPS) community, as well as the Math StackExchange and MathOverflow forums. The size of the set ranges from 100,000 to 1 million records.
What you can do with it
- SFT fine-tuning of language models in step-by-step logical and mathematical reasoning skills
- Distillation of capabilities from large reasoning models into more compact models or LoRA adapters
- Tasks
- Text generation
- Languages
- English
- Size
- 100K–1M rows
- Downloads this month
- 11.1K
- Likes
- 58
Models trained on it
- abhishekchohan/Qwen3.8-27B-AWQ-INT4-FP8KVImage text → text 4K
- mradermacher/FlashModel-Qwen3.5-9B-i1-GGUF 2.3K
- mradermacher/qwen3-30b-a3b-base-reasoning-sft-nemotron-math-v4-500m-i1-GGUF 873
- flashback2k/FlashModel-Qwen3.5-9B-GGUFText generation 732
- abhishekchohan/Qwen3.8-27B-AWQ-INT4Image text → text 660
- abhishekchohan/Qwen3.8-27B-GPTQ-INT4-FP8KVImage text → text 389
- mradermacher/FlashModel-Qwen3.5-9B-GGUF 380
- abhishekchohan/Qwen3.8-27B-GPTQ-INT4Image text → text 165
How to load it
from datasets import load_dataset
ds = load_dataset("nvidia/Nemotron-SFT-Math-v4")