Llama-Nemotron Post-Training Dataset
NVIDIA
Millions of math, code, and reasoning examples used to post-train NVIDIA's Llama-Nemotron models.
A compilation of supervised fine-tuning and reinforcement learning data supporting math, code, general reasoning, and instruction following. One row is a prompt with a model-written response and metadata such as category and whether reasoning is on. The v1.1 update added 2.2 million math and 500,000 code reasoning examples.
Access
- Cost
- FreeNo charge.
- License
- Open licenseCC BY 4.0
- Sign-up
- No sign-up
- How you get it
- API and Bulk downloadJSON Lines
What's in it
- One row is
- One row per prompt and response
- Covers
- Not geographic
- Years
- 2025
- Updated
- Irregular
- Size
- 1 to 10 million rows (v1.1)
Quality
Large and useful for reasoning post-training, but responses are model-generated and sourced from mixed upstream licenses.
- Responses come from other models with their own terms
- Not all answers are verified correct
Connects to
- OpenMathReasoning: Math reasoning data from the same company
- OpenThoughts-114k: Smaller open reasoning set built on a similar distillation idea
Questions it could answer
- How do reasoning-on and reasoning-off answers differ in length?
- Which categories dominate the mix?