Cover: AI-generated editorial illustration.
Published November 9, 2026 by NVIDIA and Elith on Hugging Face.
Nemotron-Personas-Finance-Japan extends the earlier Nemotron-Personas-Japan dataset with financial context. NVIDIA and Elith describe it as the first open persona dataset at million-record scale that combines Japan's official statistics with household finance and financial literacy data.
What is in it
- 9 million personas (1 million base records, 9 personas each), written in natural Japanese
- 50 fields per record covering demographics, assets, occupation (1,600+ categories), digital literacy, and financial behavior
- Approximately 1.5 billion tokens total
- CC BY 4.0 license, available for commercial use
The dataset was built using NVIDIA's NeMo Data Designer and NeMo SDG-PGMs probabilistic graphical model framework, with Gemma-4-31B-it handling Japanese text generation. Major Japanese financial institutions including Mizuho Financial Group and Tokio Marine Holdings provided feedback during development, according to the announcement.
Who it is for
The primary audience is teams building financial AI for Japanese banks, insurers, and the startups serving them. Practical uses include seeding multi-turn dialogue datasets, evaluating customer-service agents across demographic slices, and red-teaming customer-facing AI for fairness across age, region, and income groups.
A companion dataset, Nemotron-PII-Japan, releases the same day with 100,000 synthetic Japanese business documents labeled for personal information, intended for training and testing PII detection and redaction pipelines.
For developers working on Japanese-language AI outside finance, the 9M personas offer significantly finer population coverage than the base Nemotron-Personas-Japan dataset (50 fields vs. 22). Load it with load_dataset("nvidia/Nemotron-Personas-Finance-Japan").