NVIDIA announced the launch of two new reference datasets aimed at accelerating the pace of AI development and deploying Japanese-language AI agents. The announcement came during the "NVIDIA Agentic AI Day Japan 2026" conference organized by the company in the Japanese capital, Tokyo, highlighting ongoing efforts to support the linguistic and technical localization of Nemotron models.
Details of the New Datasets
The new releases include two main sets specifically designed to meet the requirements of the Japanese market with high precision:
- Nemotron-Personas-Finance-Japan: A specialized dataset in the financial sector, considered an extension of the Japanese Nemotron personas series. It includes detailed user simulation media and models covering demographics, behavioral patterns, income levels, financial interests, and living planning details to assist in banking and customer service applications. This set was developed in collaboration with the Japanese startup "Elith" with support from leading financial institutions such as Mizuho Financial Group and Sumitomo Mitsui Insurance.
- Nemotron-PII-Japan: A reference dataset consisting of 10 million text lines specifically designed to handle Personally Identifiable Information (PII) in Japanese. The set covers 8 key sectors including healthcare, finance, employment, insurance, legal services, information technology, and e-commerce. It supports 43 different types of text patterns and 56 levels of personal data sensitivity with precise labeling of their locations within texts.
Why This News Matters
This launch is part of NVIDIA's strategy to overcome the obstacle of a lack of reliable and contextually relevant data for non-English languages. Despite the significant power of large and small language models (SLMs), developers face real challenges in accurate training on privacy and local financial data. These two datasets provide a ready foundation for developers and companies to comply with local regulatory standards in Japan while building AI agents capable of operating safely and efficiently in real working environments.
Availability and Practical Adoption
The two new datasets are available to developers and interested parties via the Hugging Face platform. This step underscores NVIDIA's commitment to developing local AI infrastructure in the Japanese market, in close collaboration with local partners to ensure easy integration into commercial systems and sensitive databases.