Unlocking Japanese LLMs with AWS Trainium: Innovators Showcase from the AWS LLM Development Support Program

This article discusses how various Japanese organizations are leveraging AWS Trainium to develop large language models (LLMs) and foundation models (FMs) tailored for Japanese language capabilities. It highlights the AWS LLM Development Support Program in Japan, which provided resources and guidance to 15 participating companies and organizations.

Specifically, the article covers:

• Ricoh's curriculum learning approach to train a bilingual Japanese-English LLM
• Stockmark's pre-training of a Japanese LLM to mitigate hallucination
• NTT's development of a lightweight, high-performance Japanese LLM called "tsuzumi"
• Other participants' innovations in domain-specific, multimodal, and multilingual LLMs
• The continued support from AWS and Japan's Ministry of Economy, Trade, and Industry for generative AI development