NVIDIA
NVIDIA launches NIM microservices for generative AI in Japan and Taiwan Accelerate the deployment of sovereign AI applications with cultural understanding and language capabilities
……
[Image: https://prtimes.jp/i/12662/481/resize/d12662-481-98d310d26d9747402229-0.jpg&s3=12662-481-2b73bc52ac5f403f8815adfac34713f8-1280×680.jpg] Countries around the world are using their computing infrastructure, data, workforce and business networks to create artificial
intelligence and develop sovereign AI
(https://blogs.nvidia.co.jp/2024/05/10/what- is-sovereign-ai/), aligning AI systems with local values, laws and interests.
To support these efforts, NVIDIA today announced four new NVIDIA NIM microservices (https://www.nvidia .com/ja-jp/ai/).
These microservices correspond to a leading community model, created specifically with local needs in mind. These models improve
interactions with users through accurate understanding of local language and cultural heritage and improved responsiveness.
According to ABI Research
(https://www.abiresearch.com/market-research/product/market-data/MD-AISG/), generative AI software sales in the Asia-Pacific region alone will reach $5 billion this year. is expected to reach $48 billion by 2030. Llama-3-Swallow-70B trained on Japanese data and Llama-3-Taiwan-70B trained on Mandarin understand local laws, regulations and other customs. A deeply understood regional language model.
“Rakuten AI 7B” is a series of models based on “Mistral-7B” that is trained using English and Japanese datasets. The chat model and instruction-tuned model of the same large-scale language model are now available as separate NIM microservices. The basic model and instruction-tuned model of “Rakuten AI 7B” were evaluated within Rakuten from January to March 2024 using the “LM Evaluation Harness” standards, and were evaluated in the open Japanese large-scale language model. Obtained the top average score
(https://corp.rakuten.co.jp/news/press/2024/0321_01.html).
Create culturally and linguistically sensitive models by training Large Language Models (LLM) on regional languages. Differences are better understood and reflected, allowing for more accurate and nuanced communication, making output more effective.
These models have superior performance in Japanese and Chinese language understanding, addressing local legal challenges, Q&A, and language translation and summarization when compared to basic LLMs like Llama 3. I will demonstrate.
Singapore (https://blogs.nvidia.com/blog/singtel-sovereign-ai/), United Arab Emirates
(https://blogs.nvidia.com/blog/world-governments-summit/), South Korea, Sweden Countries around the world are investing in sovereign AI, from France to Italy to India.
The new NIM microservices enable enterprises, government agencies, and universities to host LLM natively in their environments, allowing developers to build advanced copilots, chatbots, and AI assistants. Application development using Sovereign AI NIM microservices Developers can deploy sovereign AI models packaged as NIM
microservices into production with improved performance.
The microservices available with NVIDIA AI Enterprise
(https://www.nvidia.com/ja-jp/data-center/products/ai-enterprise/) are NVIDIA TensorRT-LLM (https://docs.nvidia. com/tensorrt-llm/index.html) Optimized for inference by open source libraries.
NIM microservices for Llama 3 70B, used as the base model for the new Llama-3-Swallow-70B and Llama-3-Taiwan-70B NIMs, deliver up to 5x more throughput. This reduces the overall cost of running models in production and improves user experience through lower latency. The new NIM microservice is available as a hosted application programming interface (API) starting today.
Achieve faster, more accurate generative AI outcomes with NVIDIA NIM NIM microservices accelerate deployment and improve overall
performance, while providing the security needed by organizations in diverse industries around the world, including healthcare, finance, manufacturing, education, and legal.
Tokyo Institute of Technology used Japanese data to fine tune Llama-3-Swallow 70B.
Professor Rio Yokota of Tokyo Institute of Technology’s Academic Information Center says: “LLM is not a mechanical tool that provides the same benefits to everyone. Rather, it is an intellectual tool that interacts with human culture and creativity. The influence is reciprocal, and the models we train Not only will our culture and the data we generate be affected by LLM, but developing sovereign AI models that adhere to our cultural norms is paramount. The
availability of Llama-3-Swallow as an NVIDIA NIM microservice will allow developers to easily access and deploy models in Japanese applications across a variety of industries.
For example, Japanese AI company Preferred Networks uses this model to develop a healthcare-specific model, Llama3-Preferred-MedSwallow-70B, that is trained using a proprietary corpus of Japanese medical data. Masu. This model has achieved high scores in the Japanese National Medical Examination.
Chang Gung Memorial Hospital (CGMH), one of Taiwan’s leading hospitals, has built a custom-made AI Inference Service (AIIS) to centralize all LLM applications within the hospital system. With Llama 3-Taiwan 70B, patients are using a more nuanced medical language that is easier to understand and front-line medical staff are more efficient.
“We’re excited to be working with Dr. Changfu Ku, director of the Center for Artificial Intelligence in Medicine at CGMH Linko. “AI applications built with LLM in local languages not only streamline workflows by providing immediate, contextual guidance, but also support staff development and ongoing support to improve the quality of patient care.” NVIDIA NIM simplifies the development of these applications, making it easy to use and deploy models trained in regional languages with minimal expertise.”
Pegatron, an electronic device manufacturer headquartered in Taiwan, plans to adopt Llama 3-Taiwan 70B NIM microservices for internal and external applications. By integrating this microservice into its PEGAAi Agentic AI System, the company automates processes and significantly increases manufacturing and operational efficiency. Llama-3-Taiwan 70B NIM is partnered with Chang Chun Group, a global petrochemical manufacturer, Unimicron, a world-class printed circuit board company, TechOrange, a technology media company, and LegalSign, an online contract service provider. It is also used by .ai and generative AI startup APMIC. These companies also collaborate in an open model.
Create custom enterprise models with NVIDIA AI Foundry
While regional AI models provide culturally sensitive and localized responses, enterprises need to fine-tune them to align with their business processes and expertise.
NVIDIA AI Foundry (https://www.nvidia.com/ja-jp/ai/foundry/) is a commonly used base model, and NVIDIA NeMo (https://www.nvidia.com/) for fine tuning. ja-jp/ai-data-science/products/nemo/) and NVIDIA DGX Cloud (https://www.nvidia.com/ja-jp/data-center/dgx-cloud/). platform and services. This provides developers with a full stack solution for creating customizable underlying models packaged as NIM microservices. Additionally, developers using NVIDIA AI Foundry can use NVIDIA AI Enterprise (https://www.nvidia.com/en-us/data) for security, stability, and support for production deployments.
-center/products/ai-enterprise/) software platform.
NVIDIA AI Foundry gives developers the tools to build AI applications and deploy their own custom local language NIM microservices quickly and easily. As a result, developers can deliver culturally and linguistically appropriate outcomes to users.
About NVIDIA
Since its founding in 1993, NVIDIA (https://www.nvidia.com/ja-jp/) (NASDAQ: NVDA) has been a pioneer in accelerated computing. Invented by the company in 1999, the GPU has fueled the growth of the PC gaming market, redefined computer graphics, ignited the modern AI era, and helped create the Metaverse. NVIDIA is now a full-stack computing company with data center-scale products that are revolutionizing industry. For more information, follow this link:
https://nvidianews.nvidia.com/
This article has been partially generated with the assistance of AI.