NVIDIA and Elith Release Nemotron-Personas-Finance-Japan, a 9M-Record Synthetic Dataset for Japanese Financial AI

• NVIDIA and Elith released Nemotron-Personas-Finance-Japan, a synthetic dataset of 9 million personas (1 million records with 9 personas each) designed for financial AI in Japan.
• The dataset includes 50 fields per record, covering demographics, household finances, and financial literacy, totaling approximately 1.5 billion tokens.
• Built using NeMo SDG-PGMs and gemma-4-31B-it, the data aligns with official Japanese government statistics while ensuring privacy by excluding real individual records.
• Released under the CC BY 4.0 license, it supports commercial use for building conversational agents, customer simulations, and bias testing in the financial sector.

Technical Specifications

• 9 million total personas with 50 fields per record, including 9 persona fields, 11 attributes, and 30 contextual fields.
• Approximately 950,000 unique names and over 1,600 occupation categories reflecting Japan's workforce.
• Generated using NeMo Data Designer for templating and validation, with NeMo SDG-PGMs for probabilistic graphical modeling.

Availability and Integration

• Available on Hugging Face under the CC BY 4.0 license for both commercial and non-commercial use.
• Designed to integrate with the NeMo User Sim framework for instant execution of user simulations.
• Released alongside Nemotron-PII-Japan, a companion dataset of 100,000 synthetic business documents for PII detection testing.