RakutenAI
About
The RakutenAI family of large language models (LLMs) features advanced models primarily for Japanese language processing, with solid performance in English. The flagship model, RakutenAI-7B, is a 7-billion parameter model developed from the open-source Mistral-7B-v0.1, enhanced by training on a diverse dataset of Japanese and English texts. This training, executed on Rakuten's powerful GPU infrastructure, enables the model to excel in Japanese language tasks. The model's tokenizer is specifically optimized for Japanese characters, ensuring efficient processing. Alongside the foundation model, Rakuten offers instruction-tuned and chat-tuned versions, dubbed RakutenAI-7B-instruct and RakutenAI-7B-chat, for more task-specific applications. All models are open-source under the Apache 2.0 license and available on Hugging Face. However, users should be aware of potential biases and inaccuracies 11 10 12.