# Efficient Tokenization

Shisa V2 is a bilingual (Japanese/English) general-purpose chat model trained by Shisa.AI, focusing on enhancing Japanese task capabilities while maintaining strong English proficiency.

Ruri v3 is a Japanese general-purpose text embedding model based on ModernBERT-Ja, supporting sequences up to 8192 tokens long and achieving state-of-the-art performance in Japanese text embedding tasks.

A large language model continuously pre-trained on a mixed Japanese and English dataset based on Qwen-14B, significantly improving performance on Japanese tasks