started · updated
NVIDIA releases synthetic Japanese persona datasets for AI training
NVIDIA has released two new synthetic datasets designed to advance agentic AI development in Japan: Nemotron-Personas-Japan V2 and Nemotron PII Japan.
Nemotron-Personas-Japan V2 expands upon a previous release by providing approximately 9 million synthetic Japanese personas. This version includes specific financial attributes such as income and assets, allowing for more sophisticated user simulations in marketing, market research, and sales training without using real personal information. The project was led by NVIDIA with implementation and data generation handled by the Japanese AI startup Elith.
Additionally, the Nemotron PII Japan dataset was released to assist in privacy protection. It contains 43 types of documents across eight sectors, including medical, financial, and legal fields. The dataset is designed to help developers train and evaluate models for detecting and masking personally identifiable information (PII) while maintaining privacy standards. Both datasets are available on Hugging Face.