Skip to content

Dataset · nvidia

NVIDIA math and code SFT dataset

nvidia/AceReason-1.1-SFT

4M SFT examples for math and code reasoning with answers generated by DeepSeek-R1.

What's inside

AceReason-1.1-SFT is an English supervised fine-tuning (SFT) dataset for math and code reasoning. It contains 2,668,741 math examples and 1,301,591 code examples; all answers are generated by DeepSeek-R1. Sources include OpenMathReasoning, NuminaMath-CoT, OpenCodeReasoning, MagicoderEvolInstruct, and opc-sft-stage2. This is SFT data for the AceReason-Nemotron-1.1-7B model.

What you can do with it

  • Train an SFT model for mathematical reasoning
  • Fine-tune LoRA for code generation
  • Evaluate reasoning quality after SFT
Tasks
Text generation
Languages
English
Size
1M–10M rows
Downloads this month
11K
Likes
105

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("nvidia/AceReason-1.1-SFT")