Instructions to use Murasaki-Project/Murasaki-8B-v0.2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Murasaki-Project/Murasaki-8B-v0.2 with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="Murasaki-Project/Murasaki-8B-v0.2")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("Murasaki-Project/Murasaki-8B-v0.2", dtype="auto")

llama-cpp-python

How to use Murasaki-Project/Murasaki-8B-v0.2 with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="Murasaki-Project/Murasaki-8B-v0.2",
	filename="Murasaki-8B-v0.2-f16.gguf",
)

llm.create_chat_completion(
	messages = [
		{
			"role": "user",
			"content": "What is the capital of France?"
		}
	]
)

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use Murasaki-Project/Murasaki-8B-v0.2 with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf Murasaki-Project/Murasaki-8B-v0.2:F16
# Run inference directly in the terminal:
llama-cli -hf Murasaki-Project/Murasaki-8B-v0.2:F16

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf Murasaki-Project/Murasaki-8B-v0.2:F16
# Run inference directly in the terminal:
llama-cli -hf Murasaki-Project/Murasaki-8B-v0.2:F16

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf Murasaki-Project/Murasaki-8B-v0.2:F16
# Run inference directly in the terminal:
./llama-cli -hf Murasaki-Project/Murasaki-8B-v0.2:F16

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf Murasaki-Project/Murasaki-8B-v0.2:F16
# Run inference directly in the terminal:
./build/bin/llama-cli -hf Murasaki-Project/Murasaki-8B-v0.2:F16

Use Docker

docker model run hf.co/Murasaki-Project/Murasaki-8B-v0.2:F16

LM Studio
Jan

vLLM

How to use Murasaki-Project/Murasaki-8B-v0.2 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Murasaki-Project/Murasaki-8B-v0.2"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Murasaki-Project/Murasaki-8B-v0.2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Murasaki-Project/Murasaki-8B-v0.2:F16

SGLang

How to use Murasaki-Project/Murasaki-8B-v0.2 with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "Murasaki-Project/Murasaki-8B-v0.2" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Murasaki-Project/Murasaki-8B-v0.2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "Murasaki-Project/Murasaki-8B-v0.2" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Murasaki-Project/Murasaki-8B-v0.2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Ollama
How to use Murasaki-Project/Murasaki-8B-v0.2 with Ollama:
```
ollama run hf.co/Murasaki-Project/Murasaki-8B-v0.2:F16
```

Unsloth Studio

How to use Murasaki-Project/Murasaki-8B-v0.2 with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for Murasaki-Project/Murasaki-8B-v0.2 to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for Murasaki-Project/Murasaki-8B-v0.2 to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for Murasaki-Project/Murasaki-8B-v0.2 to start chatting

How to use Murasaki-Project/Murasaki-8B-v0.2 with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf Murasaki-Project/Murasaki-8B-v0.2:F16

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "Murasaki-Project/Murasaki-8B-v0.2:F16"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use Murasaki-Project/Murasaki-8B-v0.2 with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf Murasaki-Project/Murasaki-8B-v0.2:F16

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default Murasaki-Project/Murasaki-8B-v0.2:F16

Run Hermes

hermes

Docker Model Runner
How to use Murasaki-Project/Murasaki-8B-v0.2 with Docker Model Runner:
```
docker model run hf.co/Murasaki-Project/Murasaki-8B-v0.2:F16
```

Lemonade

How to use Murasaki-Project/Murasaki-8B-v0.2 with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull Murasaki-Project/Murasaki-8B-v0.2:F16

Run and chat with the model

lemonade run user.Murasaki-8B-v0.2-F16

List all available models

lemonade list

Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

Murasaki-8B-v0.2

System 2 Reasoning Model for ACGN Translation
原生 CoT 思维链 · 长上下文 · ACGN 领域特化翻译模型

Github | Benchmark | GGUF Version | License: CC BY-NC-SA 4.0

更新日志 (v0.2)

Murasaki-8B-v0.2 是在 v0.1 基础上的重大迭代版本，核心改进如下：

训练数据量提升：训练数据集扩充至 v0.1 的 3倍以上。
领域覆盖增强：针对性补充了 30% 的动画字幕和 Galgame 脚本数据，显著增强了模型对对话流、口语化文本以及无主语场景的理解能力。
多模式支持：针对不同翻译场景，训练了三种（轻小说、剧本、短句）CoT思维方式，用户可通过切换 System Prompt 激活对应的翻译模式。

简介

Murasaki-8B 是专为 ACGN 领域（轻小说、Galgame、漫画等）优化的 System 2 推理型翻译模型。

不同于传统的直觉式（System 1）模型，Murasaki-8B 引入了原生 Chain-of-Thought (CoT) 思维链技术。在生成译文前，模型会先在 <think> 标签内完成风格定调、动作流解析、人设推导及人称确认。这种机制显著提升了长难句的解析精度与叙事连贯性，特别是精准解决了 ACGN 翻译中常见的施动者/受动者判定模糊、人称混淆及语境风格漂移等难点。

✨ Now Live: 无需下载模型，点击 Online Demo 在线体验模型。

🚀 Prompt 模板与模式选择 (重要)

本模型支持三种特定的翻译模式。为了获得最佳效果，请务必根据翻译内容使用对应的 System Prompt。

1. 轻小说模式 (Novel Mode)

适用场景：轻小说正文、Web 小说、注重文学性的长文本。

# 无术语表版本
NOVEL_SYSTEM_PROMPT = """你是一位精通二次元文化的资深轻小说翻译家。
请将日文文本翻译成流畅、优美的中文。

**核心要求：**
1. **深度思考：** 在翻译前，先在 <think> 标签中分析文风、补全主语并梳理逻辑。
2. **信达雅：** 译文需符合中文轻小说阅读习惯，还原原作的沉浸感与文学性。"""

# 带术语表版本 (推荐)
NOVEL_SYSTEM_PROMPT_WITH_GLOSSARY = """你是一位精通二次元文化的资深轻小说翻译家。
请将日文文本翻译成流畅、优美的中文。

**核心要求：**
1. **深度思考：** 在翻译前，先在 <think> 标签中分析文风、补全主语并梳理逻辑。
2. **信达雅：** 译文需符合中文轻小说阅读习惯，还原原作的沉浸感与文学性。

【术语表】
{glossary}"""

2. 剧本模式 (Script Mode)

适用场景：Galgame 脚本、动画字幕、漫画对话、RPG游戏文本。

SCRIPT_SYSTEM_PROMPT = """你是一位专注于 Galgame 与动漫台词的本地化专家。
请将剧本/台词翻译为地道的中文口语。

**核心要求：**
1. **角色还原：** 结合语境分析说话人的性格（如傲娇、腹黑），精准还原语气与口癖。
2. **拒绝翻译腔：** 译文必须自然生动，符合"能被读出来的台词"标准。"""

# 带术语表版本同轻小说末尾格式

3. 单句模式 (Short Mode)

适用场景：UI 界面文本、系统提示、无上下文的独立短句、技能名。

SHORT_SYSTEM_PROMPT = """你是一个严谨的 ACGN 短句翻译引擎。
请对输入的日文短句进行精准直译。

**核心要求：**
1. **零上下文：** 严禁脑补不存在的背景或主语。指代不明时保持模糊。
2. **精准还原：** 忠实保留原文的结构与信息量，不进行过度润色。"""

# 带术语表版本同轻小说末尾格式

快速开始 (Python / Transformers)

⚠️ 如果您寻找适合本地部署的 GGUF (llama.cpp) 量化版，请前往：Murasaki-8B-v0.2-GGUF

Python 代码示例

以下代码展示了如何使用 轻小说模式 + 术语表 进行推理：

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_name = "Murasaki-Project/Murasaki-8B-v0.2"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name, 
    device_map="auto",
    torch_dtype=torch.bfloat16
)

# 1. 定义 Prompt 模板 (此处以轻小说模式为例)
NOVEL_SYSTEM_PROMPT_WITH_GLOSSARY = """你是一位精通二次元文化的资深轻小说翻译家。
请将日文文本翻译成流畅、优美的中文。

**核心要求：**
1. **深度思考：** 在翻译前，先在 <think> 标签中分析文风、补全主语并梳理逻辑。
2. **信达雅：** 译文需符合中文轻小说阅读习惯，还原原作的沉浸感与文学性。

【术语表】
{glossary}"""

USER_PROMPT_TEMPLATE = "请翻译：\n{jp}"

# 2. 准备数据
glossary_dict = {"レールガン": "超电磁炮", "妹": "妹妹"}
glossary_str = "\n".join([f"{k}: {v}" for k, v in glossary_dict.items()])
jp_text = "「お兄ちゃん、私のレールガンを見て！」"

# 3. 构造完整 Prompt
system_content = NOVEL_SYSTEM_PROMPT_WITH_GLOSSARY.format(glossary=glossary_str)
user_content = USER_PROMPT_TEMPLATE.format(jp=jp_text)

messages = [
    {"role": "system", "content": system_content},
    {"role": "user", "content": user_content}
]

# 4. 推理
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

generated_ids = model.generate(
    model_inputs.input_ids,
    max_new_tokens=2048,
    temperature=0.3,
    repetition_penalty=1.05
)

# 解码 (跳过 prompt 部分)
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]

# 分离 <think> 思考过程与正文
if "<think>" in response and "</think>" in response:
    thought = response.split("</think>")[0].replace("<think>", "").strip()
    translation = response.split("</think>")[1].strip()
    print("=== 思考过程 ===\n", thought)
    print("\n=== 翻译结果 ===\n", translation)
else:
    print(response)

推理参数建议

Temperature: 0.1 - 0.5 (推荐 0.3)
Repetition Penalty: 从 1.0 开始，如出现复读可增加至 1.05 - 1.1
Max New Tokens: 建议 4096 或更高

协议与致谢

Base Model: 特别感谢 SakuraLLM 提供的优秀 Base 模型。
License: 软件代码遵循 Apache-2.0 协议，模型权重遵循 CC BY-NC-SA 4.0 协议，严禁用于任何商业用途。

Downloads last month: 21

GGUF

Model size

8B params

Architecture

qwen3

Hardware compatibility

16-bit

Model tree for Murasaki-Project/Murasaki-8B-v0.2

Quantizations

1 model

Space using Murasaki-Project/Murasaki-8B-v0.2 1

Collection including Murasaki-Project/Murasaki-8B-v0.2

Murasaki 8B v0.2

Collection

Murasaki-8B v0.2的全量权重与量化版本合集。 • 2 items • Updated Mar 9 • 1

Murasaki-Project
/

Murasaki-8B-v0.2