Model Overview

Yoko-7B-Japanese-v0 is a 7 billion parameter language model developed by ganchengguang, fine-tuned from the vanilla LLaMA2-7B architecture using QLoRA. This model was specifically trained on 49,000 chat samples from the Guanaco dataset, with contributions from Yokohama National University Mori Lab.

Key Capabilities

• Enhanced Multilingual Performance: Demonstrates improved performance in both Chinese and Japanese language processing, making it suitable for applications targeting these languages.
• Efficient Fine-tuning: Utilizes QLoRA for fine-tuning, allowing for efficient adaptation of the base LLaMA2 model.
• Chat-oriented Training: Trained on chat samples, indicating a focus on conversational AI tasks.

Recommended Usage

This model is particularly well-suited for:

• Chatbots and Conversational AI: Its training on chat samples makes it effective for dialogue systems.
• Applications requiring Chinese and Japanese language support: The model's improved performance in these languages is a key differentiator.

Generation Parameters

For optimal results, the following generation parameters are recommended:

• temperature: 0.5 to 0.7
• top_p: 0.65 to 1.0
• top_k: 30 to 50
• repeat_penalty: 1.03 to 1.17