Instructions to use Hanversion/SunXiaochuan-V1-3B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Hanversion/SunXiaochuan-V1-3B with Transformers:

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

pipe = pipeline("text-generation", model="Hanversion/SunXiaochuan-V1-3B")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Hanversion/SunXiaochuan-V1-3B")
model = AutoModelForCausalLM.from_pretrained("Hanversion/SunXiaochuan-V1-3B", 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=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Hanversion/SunXiaochuan-V1-3B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Hanversion/SunXiaochuan-V1-3B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Hanversion/SunXiaochuan-V1-3B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Hanversion/SunXiaochuan-V1-3B

SGLang

How to use Hanversion/SunXiaochuan-V1-3B 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 "Hanversion/SunXiaochuan-V1-3B" \
    --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": "Hanversion/SunXiaochuan-V1-3B",
		"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 "Hanversion/SunXiaochuan-V1-3B" \
        --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": "Hanversion/SunXiaochuan-V1-3B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Hanversion/SunXiaochuan-V1-3B with Docker Model Runner:
```
docker model run hf.co/Hanversion/SunXiaochuan-V1-3B
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

SunXiaochuan-V1-3B

1. 介绍

本模型基于Qwen2.5-3B-Instruct并使用Lora进行模型微调获得。数据集的使用来自孙笑川吧某些帖子中的回复及楼中楼回复，由于未对数据进行全面清理，故数据集质量与数量并不高。数据集可以从我的 HuggingFace数据集中获得。除孙笑川吧数据外，还包括弱智吧、中国人口吧以及航空母舰吧的部分数据，数据来源日期为 2025-03-05。数据格式如下：

question：帖子标题
answer：帖子最热门的回复
think：DeepSeek-V3所生成的思考链

本模型仅供学习使用，不保证模型真实性及有效性。

2. 训练介绍

本模型的训练使用了 LLaMA-Factory 作为微调工具，训练集样本大小为3000条数据，并且修改梯度累计为2，Lora秩为16，Lora+为16等，设备使用单卡4090训练1小时得到。具体训练Loss曲线如下：

同时，模型训练完成后，用其他500条数据进行数据预测和模型评估，得到如下结果：

{
    "predict_bleu-4": 2.2241416,
    "predict_model_preparation_time": 0.0043,
    "predict_rouge-1": 13.033509999999998,
    "predict_rouge-2": 1.3886417999999998,
    "predict_rouge-l": 10.5493132,
    "predict_runtime": 610.2644,
    "predict_samples_per_second": 0.819,
    "predict_steps_per_second": 0.41
}

3. 模型介绍

SunXiaochuan-V1

模型名称	#总参数	#微调参数	上下文长度	下载
SunXiaochuan-V1-3B	3B	0.5B	128K	🤗 HuggingFace

使用 Qwen2.5 进行训练，无思考推理过程

SunXiaochuan-R

模型名称	#总参数	#微调参数	上下文长度	下载
DeepShit-R1-1.5B-Sunba-Distill	1.5B	1B	128K	🤗 HuggingFace
MAGA-T1-Tieba-1.5B-Distill	1.5B	2B	128K	🤗 HuggingFace

基于 DeepSeek-R1 进行训练，加入思考推理过程

5. 联系

欢迎与我进行讨论，联系方式如下：

Github：Hanversion
Email：hanversion@outlook.com

Downloads last month: 10

Safetensors

Model size

3B params

Tensor type

BF16

Model tree for Hanversion/SunXiaochuan-V1-3B

Quantizations

1 model