The U.S. semiconductor giant Nvidia recently announced its support for developing generative artificial intelligence (AI) based on Japanese language data, expanding its AI development support service for enterprises. This initiative will include large language models (LLMs) trained with Japanese data from the Tokyo Institute of Technology and Rakuten Group. The move is part of a broader global effort by various nations to develop sovereign AI, using their own infrastructures, data, and business networks to ensure that AI systems align with local values, laws, and interests.
Nvidia's decision to expand its support service to include language models trained in Japanese is crucial in helping Japan protect its critical infrastructure and strengthen its industrial competitiveness without relying on other countries for data or human resources. This initiative is also designed to enhance national economic security, a topic that has gained importance in an increasingly digitalized and globalized world.
Nvidia CEO Jensen Huang emphasized the importance of Japan developing its own generative AI in March 2024, arguing against relying on third parties who might collect Japanese data, create AI with it, and then reintroduce it to Japan. "There's no reason to allow some other third party to harvest that data (of Japan), create an AI, and then import it back to Japan," Huang said, highlighting the need to preserve data sovereignty in technological development.
Nvidia's NIM Microservices
To support these efforts, Nvidia has launched four new NIM (Nvidia Inference Model) microservices, enabling developers to easily build and deploy high-performing generative AI applications. These microservices support popular community models tailored to meet regional needs, enhancing user interactions through more accurate understanding and improved responses based on local languages and cultural heritage.
Among the available models are Llama-3-Swallow-70B, trained on Japanese data, and Llama-3-Taiwan-70B, trained on Mandarin data. These regional models provide a deeper understanding of local laws and regulations and better reflect local customs and cultural norms. For instance, the Tokyo Institute of Technology fine-tuned the Llama-3-Swallow 70B model using Japanese language data, allowing for more precise adaptation to local applications.
Additionally, the RakutenAI 7B family of models, built on the Mistral-7B model, was trained on English and Japanese datasets and is available as two different NIM microservices for Chat and Instruct. These models have achieved leading scores among large Japanese language models, earning the top average in the LM Evaluation Harness benchmark from January to March 2024.
Asian and global Market
According to market research firm ABI Research, generative AI software revenue in the Asia-Pacific region is expected to reach $48 billion by 2030, a significant increase from $5 billion this year. Nvidia aims to capitalize on this growth by boosting sales of its regional services.
In this context, Nvidia AI Foundry is an essential platform for enterprises to customize and deploy AI models tailored to their business processes and expertise. By providing a full stack of tools, including popular foundation models, Nvidia's NeMo platform for fine-tuning, and dedicated capacity on Nvidia DGX Cloud, developers can create custom models packaged as NIM microservices, ensuring culturally and linguistically appropriate results.
Nvidia's initiative to support the development of generative AI in Japan strengthens Japan's position in the global tech landscape. It sets a precedent for other nations seeking to develop sovereign AI. Countries like Singapore, the United Arab Emirates, South Korea, Sweden, France, Italy, and India are already investing in sovereign AI infrastructure, and the availability of the new NIM microservices makes it easier for businesses, government agencies, and universities to deploy native language models in their environments.
