ArrowPro-7B-KUJIRA Open-Source Japanese Model - A High-Quality Conversation Tool for AI Anchors and Assistants

Arrowpro 7B KUJIRA

Developed by DataPilot

ArrowPro-7B-KUJIRA is a Japanese-optimized model based on the Mistral series, specifically designed for AI virtual streamers and AI assistants, featuring outstanding conversational performance and prompt injection defense capabilities.

Large Language Model

Transformers

Open Source License:Apache-2.0 #Japanese AI Assistant #Virtual Streamer Optimization #Prompt Injection Defense

Downloads 47

Release Time : 5/9/2024

Model Overview

This model is a Japanese large language model developed based on NTQAI/chatntq-ja-7b-v1.0, focusing on improving dialogue experience and security, making it particularly suitable for virtual streamer and smart assistant applications.

Model Features

Leading Japanese Performance

Achieved a score of 3.8 in the ELYZA-TASK100 benchmark, ranking first globally in Japanese performance for 7B parameter models.

High-Strength Security Protection

Specifically enhanced prompt injection defense capabilities to effectively prevent malicious command attacks.

Virtual Streamer Optimization

Optimized for AI virtual streamer scenarios, delivering smooth and natural conversational experiences.

Model Capabilities

Japanese Text Generation

Dialogue Systems

Virtual Streamer Interaction

Smart Assistant Functions

Use Cases

Entertainment Industry

Virtual Streamer

Serves as the core dialogue system for virtual streamers, providing natural and fluid interactive experiences.

Enhances viewer interaction quality and live streaming effects.

Smart Assistant

Japanese Smart Assistant

Provides daily Q&A, information queries, and other services for Japanese users.

Accurately understands and responds to various questions from Japanese users.

🚀 ArrowPro-7B-KUJIRA

ArrowPro-7B-KUJIRA is based on the Mistral series NTQAI/chatntq-ja-7b-v1.0. It is designed with a focus on chat performance and high prompt injection resistance, aiming to serve as the soul of AItubers and AI assistants.

🚀 Quick Start

✨ Features

ArrowPro-7B-KUJIRA is developed based on the Mistral series NTQAI/chatntq-ja-7b-v1.0. It emphasizes chat performance and high prompt injection resistance, aiming to be the core for AItubers and AI assistants. In the benchmark (ELYZA-TASK100), it scored approximately 3.8 (compliant with LLaMa3-70B), achieving the world's best Japanese language performance among 7B models.

image/jpeg image/png

💻 Usage Examples

Basic Usage

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("DataPilot/ArrowPro-7B-KUJIRA")
model = AutoModelForCausalLM.from_pretrained(
  "DataPilot/ArrowPro-7B-KUJIRA",
  torch_dtype="auto",
)
model.eval()

if torch.cuda.is_available():
    model = model.to("cuda")

def build_prompt(user_query):
    sys_msg = "あなたは日本語を話す優秀なアシスタントです。回答には必ず日本語で答えてください。"
    template = """[INST] <<SYS>>
{}
<</SYS>>

{}[/INST]"""
    return template.format(sys_msg,user_query)

# Infer with prompt without any additional input
user_inputs = {
    "user_query": "まどマギで一番かわいいキャラはだれ？",
}
prompt = build_prompt(**user_inputs)

input_ids = tokenizer.encode(
    prompt, 
    add_special_tokens=True, 
    return_tensors="pt"
)

tokens = model.generate(
    input_ids.to(device=model.device),
    max_new_tokens=500,
    temperature=1,
    top_p=0.95,
    do_sample=True,
)

out = tokenizer.decode(tokens[0][input_ids.shape[1]:], skip_special_tokens=True).strip()
print(out)

📄 License

This project is licensed under the Apache-2.0 license.

Acknowledgments

We would like to express our gratitude to everyone who provided advice. We also thank the developers of the original model.

Request

When using this model, please make every effort to avoid causing trouble to others.

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご