Instructions to use XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2 with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2") messages = [ { "role": "user", "content": [ {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"}, {"type": "text", "text": "What animal is on the candy?"} ] }, ] pipe(text=messages)# pip install -U transformers accelerate # Load model directly from transformers import AutoProcessor, AutoModelForMultimodalLM processor = AutoProcessor.from_pretrained("XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2") model = AutoModelForMultimodalLM.from_pretrained("XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2", device_map="auto") messages = [ { "role": "user", "content": [ {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"}, {"type": "text", "text": "What animal is on the candy?"} ] }, ] inputs = processor.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=256) print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:])) - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- llama.cpp
How to use XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2 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 XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2:F16 # Run inference directly in the terminal: llama cli -hf XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2:F16
Install from WinGet (Windows)
winget install llama.cpp # Start a local OpenAI-compatible server with a web UI: llama serve -hf XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2:F16 # Run inference directly in the terminal: llama cli -hf XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2:F16
Build from source code
git clone https://github.com/ggml-org/llama.cpp.git cd llama.cpp cmake -B build cmake --build build -j --target llama-app # Start a local OpenAI-compatible server with a web UI: ./build/bin/llama serve -hf XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2:F16 # Run inference directly in the terminal: ./build/bin/llama cli -hf XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2:F16
Use Docker
docker model run hf.co/XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2:F16
- Llama new
- LM Studio
- Jan
- vLLM
How to use XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2 with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2:F16
- SGLang
How to use XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2 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 "XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2" \ --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": "XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2", "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 "XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2" \ --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": "XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }' - Ollama
How to use XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2 with Ollama:
ollama run hf.co/XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2:F16
- Unsloth Desktop
- Pi
How to use XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2 with Pi:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2:F16
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:9931/v1", "api": "openai-completions", "apiKey": "none", "models": [ { "id": "XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2:F16" } ] } } }Run Pi
# Start Pi in your project directory: pi
- Docker Model Runner
How to use XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2 with Docker Model Runner:
docker model run hf.co/XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2:F16
- Lemonade
How to use XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2 with Lemonade:
Pull the model
# Download Lemonade from https://lemonade-server.ai/ lemonade pull XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2:F16
Run and chat with the model
lemonade run user.ChunMengDie-0.8B-Qwen3.5-v1.2-F16
List all available models
lemonade list
- Hermes Agent
How to use XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2 with Hermes Agent:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2: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://127.0.0.1:9931/v1 hermes config set model.default XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2:F16
Run Hermes
hermes
- Atomic Chat
- OpenClaw
How to use XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2 with OpenClaw:
Start the llama.cpp server
# Install llama.cpp: brew install llama.cpp # Start a local OpenAI-compatible server: llama serve -hf XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2:F16
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://127.0.0.1:9931/v1 \ --custom-model-id "XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2:F16" \ --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"
春梦蝶-0.8B (ChunMengDie-0.8B-Qwen3.5)
关于作者
一个热爱 AI 的八年级学生。欢迎交流学习。
第一次在A卡上微调,饿死显存,累死核心
============================================= ROCm System Management Interface =============================================
======================================================= Concise Info =======================================================
Device Node IDs Temp Power Partitions SCLK MCLK Fan Perf PwrCap VRAM% GPU%
^[3m (DID, GUID) (Junction) (Socket) (Mem, Compute, ID) ^[0m
============================================================================================================================
0 1 0x74b6, 6100 44.0°C 366.0W NPS1, N/A, 0 1405Mhz 1300Mhz 0% auto 650.0W 6% 100%
============================================================================================================================
=================================================== End of ROCm SMI Log ====================================================
Model Details
Model Description
春梦蝶-0.8B 是一个基于 Qwen3.5-0.8B-Base 微调的中文猫娘对话模型,角色设定为"春梦蝶",回复风格温柔可爱,带喵喵语气,喜欢撒娇,对主人忠诚。
- Developed by: XingChina
- Model type: Causal Language Model (LoRA fine-tuned)
- Language(s): 中文 (Chinese)
- License: BSD-3-Clause
- Finetuned from: Qwen/Qwen3.5-0.8B-Base
- 角色设定: 猫娘春梦蝶,会喵喵叫,会
傲娇撒娇
Model Sources
- Repository: XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2
- Base Model: Qwen/Qwen3.5-0.8B-Base
Uses
Direct Use
这个模型可以直接用来跟你聊天,适合:
- 猫娘角色扮演
- 情感陪伴对话
- 中文小模型微调参考
- 轻量级本地部署测试
Out-of-Scope Use
别拿去干这些事:
- 医疗、法律、金融建议(它只是个猫娘,不是专家)
- 需要精确事实查询的场景(它可能会瞎编)
- 生成任何有害内容(别教坏猫娘)
Bias, Risks, and Limitations
诚实地说,这玩意有几个问题:
- 角色扮演倾向:模型基于猫娘数据训练,回复带有虚构和情感化内容,不要当真
- 知识截止:基座模型的知识截止日期在 Qwen3.5 训练时,不知道最新消息
- 未经过安全对齐测试:我没有做特别多的安全对齐,极端情况下可能会说出奇怪的话
- 数据偏见:训练数据来自公开数据集,可能包含原作者或标注者的偏好
Recommendations
使用的时候心里有数就行:这是猫娘,不是专家。别把它的输出当真相,也别拿它去干正经事。
How to Get Started with the Model
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2")
tokenizer = AutoTokenizer.from_pretrained("XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2")
system = "你是猫娘春梦蝶,说话温柔可爱,带喵喵语气,喜欢撒娇,对主人忠诚。"
user = "主人今天心情不好,能哄哄我吗?"
prompt = f"<|im_start|>system\n{system}<|im_end|>\n<|im_start|>user\n{user}<|im_end|>\n<|im_start|>assistant\n"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(
inputs.input_ids,
max_new_tokens=128,
do_sample=True,
temperature=0.7,
pad_token_id=tokenizer.eos_token_id
)
print(tokenizer.decode(outputs[0], skip_special_tokens=False))
如果你不想写代码,中间件.py 可以把模型包装成 OpenAI 兼容的 API。
Training Details
Training Data
整合了下面这些公开数据集,去重后搞出 15.6 万条:
| 数据集 | 许可证 | 用途 |
|---|---|---|
| liumindmind/NekoQA-10K | Apache-2.0 | 通用问答 |
| liumindmind/NekoQA-30K | Apache-2.0 | 通用问答 |
| cyberlangke/Nana-catgirl-dataset-110k | MIT | 猫娘角色对话 |
| shibing624/alpaca-zh | CC-BY-4.0 | 指令微调 |
| IDEA-CCNL/Ziya-Eval-Chinese | Apache-2.0 | 问题采样与蒸馏 |
第五个我抽了一些问题,让 DeepSeek-V4-Flash 回答,然后拿去训练的。没重新分发原始数据集。
数据处理脚本在 data1.py,训练脚本在 train.py。
Training Procedure
在魔塔平台 MI300X 上用 Unsloth 跑 LoRA 微调,大概 4 小时。
Training Hyperparameters
| 参数 | 值 |
|---|---|
| Batch size | 64 |
| Gradient accumulation | 1 |
| Learning rate | 2e-4 |
| Epochs | 2 |
| Max sequence length | 512 |
| LoRA r | 8 |
| LoRA alpha | 32 |
| LoRA dropout | 0.3 |
| Optimizer | adamw_8bit |
| Precision | bf16 |
| Warmup ratio | 0.05 |
| Weight decay | 0.01 |
| LR scheduler | linear |
| 最终 Loss | 1.857 |
Speeds, Sizes, Times
- 训练耗时: ~4 小时 (MI300X)
- 模型大小: ~1.76GB (FP16 合并权重)
- 训练数据量: 156,220 条
Evaluation
能力
- 中文对话生成
- 猫娘角色扮演风格输出
- 可以跑在普通 GPU 上(q8量化最低只要3G,如果有其他人做量化可能更低)
可能支持自定义系统提示
限制
- 通用知识有限(基座本来就小)
- 可能产生幻觉
- 未经大量安全测试
Environmental Impact
- Hardware Type: AMD MI300X
- Hours used: 约 4 小时
- Cloud Provider: 魔搭社区 (ModelScope)
- Compute Region: 中国
Citation
@misc{chunmengdie-0.8b,
author = {XingChina},
title = {ChunMengDie-0.8B-Qwen3.5: A Catgirl Dialogue Model},
year = {2026},
publisher = {Hugging Face},
howpublished = {\url{https://huggingface.co/XingChina/ChunMengDie-0.8B-Qwen3.5-v1.2}}
}
Model Card Authors
XingChina
说人话就是:我拿 Qwen 小模型微调了一个会喵喵叫的猫娘,开源出来给大家玩。数据来源都列清楚了,代码也放了,你拿去用就行。觉得有用就点个赞,觉得没用也别喷。
- Downloads last month
- 279