Skip to content

Dataset · nvidia

1M synthetic US personas

nvidia/Nemotron-Personas-USA

About 1M text descriptions of US personas generated for LLM training and evaluation.

What's inside

Nemotron-Personas-USA is an English dataset of personas built based on real distributions. Version v1.1 contains about 1M records and 0.94 billion tokens; the openai/gpt-oss-120b model was used for generation instead of Mixtral-8x22B. The data is intended for text-generation.

What you can do with it

  • Train a model to generate text on behalf of different personas
  • Test LLM diversity and robustness on synthetic profiles
  • Supplement SFT with synthetic personal contexts
Tasks
Text generation
Languages
English
Size
1M–10M rows
Downloads this month
5.6K
Likes
382

How to load it

from datasets import load_dataset

ds = load_dataset("nvidia/Nemotron-Personas-USA")