AIbase
EN
Home
Shisa V1 Llama3 8b
S

Shisa V1 Llama3 8b

Developed by shisa-ai
A Japanese-optimized large language model fine-tuned based on Meta-Llama-3-8B-Instruct, excelling in multiple Japanese benchmark tests
Large Language Model
Transformers
#Japanese optimization#Multi-turn dialogue#Low-resource fine-tuning
Downloads 28
Release Time : 5/21/2024

Model Overview

This is an 8B-parameter large language model optimized for Japanese, fine-tuned on the Llama 3 architecture, achieving excellent results on Japanese evaluation benchmarks such as ELYZA100 and Japanese MT-Bench

Model Features

Japanese optimization
Specifically optimized for Japanese tasks, with Japanese characters accounting for over 91%
Outstanding benchmark performance
Surpasses similar models on multiple Japanese evaluation benchmarks including ELYZA100, Japanese MT-Bench, and Rakuda
Precise fine-tuning
Determined 8e-6 as the optimal parameter through various learning rate experiments to avoid overfitting

Model Capabilities

Japanese text generation
Japanese Q&A
Japanese text comprehension
Multi-turn dialogue

Use Cases

Japanese NLP applications
Japanese customer service chatbot
Used for handling Japanese customer inquiries
Scored 7.05 on the Rakuda benchmark, outperforming most similar models
Japanese content creation
Generates text content that conforms to Japanese expression habits
Japanese characters account for 91.3%, with high expression naturalness
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
EnglishįŽ€äŊ“中文įšéĢ”中文ãĢãģんご
Š 2025AIbase