Llama-Nemotron-Post-Training-Dataset is a large-scale post-training dataset open sourced by NVIDIA in 2025. The related paper results are "Llama-Nemotron: Efficient Reasoning Models", which aims to improve the mathematics, code, general reasoning and instruction following capabilities of the Llama-Nemotron series models in the post-training stage (such as SFT and RL).

This dataset combines data from supervised fine-tuning (SFT) and reinforcement learning (RL) phases. The current version, v1.1 (which adds approximately 2.2 million math samples and 500,000 code reasoning samples compared to the previous version), is suitable for training AI agents, chatbots, RAG systems, and other AI-driven applications.

Data distribution (by number of category entries)

• Mathematics: 22,066,397
• Code: 10,108,883
• Science: 708,920
• Instruction following: 56,339
• Chat: 39,792
• Safety: 31,426