Instructions to use ChisatoY/Qwen2.5-7B-ReWrite with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use ChisatoY/Qwen2.5-7B-ReWrite with Transformers:

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

pipe = pipeline("text-generation", model="ChisatoY/Qwen2.5-7B-ReWrite")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("ChisatoY/Qwen2.5-7B-ReWrite")
model = AutoModelForCausalLM.from_pretrained("ChisatoY/Qwen2.5-7B-ReWrite", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use ChisatoY/Qwen2.5-7B-ReWrite with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf ChisatoY/Qwen2.5-7B-ReWrite:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf ChisatoY/Qwen2.5-7B-ReWrite:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf ChisatoY/Qwen2.5-7B-ReWrite:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf ChisatoY/Qwen2.5-7B-ReWrite:Q4_K_M

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 ChisatoY/Qwen2.5-7B-ReWrite:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf ChisatoY/Qwen2.5-7B-ReWrite:Q4_K_M

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 ChisatoY/Qwen2.5-7B-ReWrite:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf ChisatoY/Qwen2.5-7B-ReWrite:Q4_K_M

Use Docker

docker model run hf.co/ChisatoY/Qwen2.5-7B-ReWrite:Q4_K_M

LM Studio
Jan

vLLM

How to use ChisatoY/Qwen2.5-7B-ReWrite with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "ChisatoY/Qwen2.5-7B-ReWrite"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ChisatoY/Qwen2.5-7B-ReWrite",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/ChisatoY/Qwen2.5-7B-ReWrite:Q4_K_M

SGLang

How to use ChisatoY/Qwen2.5-7B-ReWrite 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 "ChisatoY/Qwen2.5-7B-ReWrite" \
    --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": "ChisatoY/Qwen2.5-7B-ReWrite",
		"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 "ChisatoY/Qwen2.5-7B-ReWrite" \
        --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": "ChisatoY/Qwen2.5-7B-ReWrite",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Ollama
How to use ChisatoY/Qwen2.5-7B-ReWrite with Ollama:
```
ollama run hf.co/ChisatoY/Qwen2.5-7B-ReWrite:Q4_K_M
```

Unsloth Studio

How to use ChisatoY/Qwen2.5-7B-ReWrite 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 ChisatoY/Qwen2.5-7B-ReWrite 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 ChisatoY/Qwen2.5-7B-ReWrite to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for ChisatoY/Qwen2.5-7B-ReWrite to start chatting

How to use ChisatoY/Qwen2.5-7B-ReWrite with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf ChisatoY/Qwen2.5-7B-ReWrite:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "ChisatoY/Qwen2.5-7B-ReWrite:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use ChisatoY/Qwen2.5-7B-ReWrite with Docker Model Runner:
```
docker model run hf.co/ChisatoY/Qwen2.5-7B-ReWrite:Q4_K_M
```

Lemonade

How to use ChisatoY/Qwen2.5-7B-ReWrite with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull ChisatoY/Qwen2.5-7B-ReWrite:Q4_K_M

Run and chat with the model

lemonade run user.Qwen2.5-7B-ReWrite-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use ChisatoY/Qwen2.5-7B-ReWrite with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf ChisatoY/Qwen2.5-7B-ReWrite:Q4_K_M

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 ChisatoY/Qwen2.5-7B-ReWrite:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use ChisatoY/Qwen2.5-7B-ReWrite with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf ChisatoY/Qwen2.5-7B-ReWrite:Q4_K_M

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "ChisatoY/Qwen2.5-7B-ReWrite:Q4_K_M" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

AI-ReWrite Qwen2.5-7B Humanize

这是一个用于将中英文 AI 风格文本改写为更自然的人类写作风格的模型。模型基于 Qwen2.5-7B-Instruct，通过 LoRA 进行监督微调（SFT），并将最佳 LoRA adapter 完整合并回基础模型。当前目录包含可直接通过 Transformers 加载的 BF16 Safetensors 权重，不需要额外挂载 LoRA adapter。

模型信息

项目	值
基础模型	Qwen2.5-7B-Instruct
模型架构	Qwen2ForCausalLM
参数规模	约 7.6B
导出精度	BF16
权重格式	Safetensors（4 个分片）
上下文上限	32,768 tokens
训练最大长度	1,024 tokens
最佳 checkpoint	step 2600（epoch 1.968）
最佳验证损失	0.641895

训练任务

训练数据统一为以下格式：

{
  "instruction": "Rewrite the provided text in a natural, human-written style while preserving its original meaning, facts, language, and formatting. Do not add explanations. 请将给定文本改写为自然的人类写作风格，保持原意、事实、语言和格式不变，不要添加解释。",
  "input": "需要改写的 AI 风格文本",
  "output": "目标人类写作风格文本"
}

训练时使用 Qwen ChatML 模板，将 instruction 作为 system message、input 作为 user message、output 作为 assistant message。损失只计算 assistant 回复部分，system 和 user tokens 均被 mask。

训练数据

清洗合并后共有 43,124 条训练候选数据，按固定随机种子从中划分 42,261 条用于训练、 863 条用于验证（验证比例 2%）。另外保留 100 条英文人工改写数据作为最终测试集，该测试集没有参与训练、early stopping 或最佳模型选择。

中文数据

中文数据来自 XiangJinYu/Qwen3.5-9B-Humanize-Dataset，共保留 24,999 条：

原始文件	转换方式	保留条数
`v15_sft_18k.jsonl`	`input → output`	18,000
`v16_dpo_4000_formal2000_casual2000.jsonl`	`rejected → chosen`	4,000
`v17next_dpo_1500_clean_no_ckpt.jsonl`	`rejected → chosen`	1,000
`v20_dpo_2000_v18rejected.jsonl`	`rejected → chosen`	1,999

该数据以中文学术摘要和正式写作为主，数据集标注许可证为 CC BY-NC 4.0。

英文数据

英文训练数据来自 KNipun/ai-humanizer，共保留 18,125 条：

原始文件	保留条数
`train_data.jsonl`	16,311
`val_data.jsonl`	1,814

该数据从原始长 prompt 的 Input: 区段中抽取源文本，以 completion 作为目标文本，数据集标注许可证为 Apache 2.0。

独立测试集

测试集来自 alexreversegpt/ai-writing-tells，包含 100 组人工编写的英文 ai_text → human_text 改写对，数据集标注许可证为 CC BY 4.0。

数据清洗

清洗过程包括：

统一中英文 system instruction；
将 SFT、DPO 和英文 prompt 数据转换为统一的 instruction/input/output 格式；
Unicode NFKC、换行和多余空白规范化；
删除空文本、过短或异常超长文本；
删除 input 与 output 完全相同的样本；
删除完全重复的输入输出对；
检查训练集与独立测试集不存在完全重复的样本对。

微调超参数

超参数	值
训练方法	LoRA SFT
LoRA rank (`r`)	16
LoRA alpha	32
LoRA dropout	0.05
LoRA bias	none
Target modules	`q_proj`, `k_proj`, `v_proj`, `o_proj`, `gate_proj`, `up_proj`, `down_proj`
LoRA 可训练参数	40,370,176（约 0.5273%）
最大序列长度	1,024
单卡训练 batch size	4
单卡验证 batch size	4
梯度累积步数	8
有效 batch size	32
最大 epoch	3
实际停止 epoch	2.4225
初始学习率	`1e-4`
学习率调度	Cosine
Warmup ratio	0.03
Weight decay	0.01
最大梯度范数	1.0
优化器	Fused AdamW (`adamw_torch_fused`)
训练精度	BF16，TF32 enabled
Attention 实现	PyTorch SDPA
Gradient checkpointing	Enabled，non-reentrant
Dynamic batching	按长度分组 (`group_by_length`)
随机种子	42
日志间隔	10 steps
验证间隔	200 steps
保存间隔	200 steps
Checkpoint 保留数	2（最佳与最近）
最佳模型指标	`eval_loss`，越低越好
Early stopping patience	3 次验证

训练使用单张 NVIDIA RTX A6000 48 GB GPU。完整训练耗时约 25,515 秒（约 7 小时 5 分钟），吞吐量约 4.97 samples/s。

训练结果

验证损失在 step 2600 达到最低点：

Step	Epoch	Eval loss
200	0.151	0.681554
1000	0.757	0.652373
1800	1.363	0.648567
2200	1.666	0.645911
2400	1.817	0.645167
2600	1.968	0.641895
2800	2.120	0.661740
3000	2.271	0.668160
3200	2.422	0.663423

step 2600 之后连续三次验证没有改善，因此 early stopping 在 step 3200 终止训练，并恢复 step 2600 的最佳权重。全程汇总训练损失为 0.652253。

在保留的 100 条英文测试集上，使用 completion-only teacher-forced loss 进行对比：

模型	Test loss	Perplexity
原始 Qwen2.5-7B-Instruct	7.9728	2900.89
本模型	3.4172	30.48

测试 loss 相对下降约 57.14%。由于每个输入只有一个参考改写，该指标适合比较微调前后的相对变化，不能替代人工评价。

Transformers 使用示例

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_path = "export/best-safetensors"
system_prompt = (
    "Rewrite the provided text in a natural, human-written style while preserving "
    "its original meaning, facts, language, and formatting. Do not add explanations. "
    "请将给定文本改写为自然的人类写作风格，保持原意、事实、语言和格式不变，"
    "不要添加解释。"
)

tokenizer = AutoTokenizer.from_pretrained(model_path)
model = AutoModelForCausalLM.from_pretrained(
    model_path,
    torch_dtype=torch.bfloat16,
    device_map="auto",
)

messages = [
    {"role": "system", "content": system_prompt},
    {"role": "user", "content": "需要改写的文本"},
]
text = tokenizer.apply_chat_template(
    messages, tokenize=False, add_generation_prompt=True
)
inputs = tokenizer(text, return_tensors="pt").to(model.device)
outputs = model.generate(
    **inputs,
    max_new_tokens=512,
    do_sample=False,
    repetition_penalty=1.05,
)
result = tokenizer.decode(
    outputs[0, inputs["input_ids"].shape[1]:],
    skip_special_tokens=True,
)
print(result)

已知限制

中文训练数据以学术摘要和正式文本为主，对口语、小说、社交媒体等文体覆盖不足。
中文输出可能使用半角逗号和句号，这是训练目标文本中的格式特征。
模型更擅长压缩句子和简化表达，不一定能完全消除宣传腔、议论文模板或空泛修饰。
改写可能引入措辞变化或事实偏移，重要内容应进行人工复核。
独立测试集规模较小且只有英文，尚不能代表完整的中英文生成质量。
使用本模型时仍需遵守基础模型以及各训练数据集的许可证和使用限制。

Downloads last month: -

Safetensors

Model size

8B params

Tensor type

BF16

Model tree for ChisatoY/Qwen2.5-7B-ReWrite

Base model

Qwen/Qwen2.5-7B

Finetuned

Qwen/Qwen2.5-7B-Instruct

Quantized

(396)

this model