Instructions to use KoreaTradeNetwork/Qwen-4B-HS4-DPO with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use KoreaTradeNetwork/Qwen-4B-HS4-DPO with Transformers:

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

pipe = pipeline("text-generation", model="KoreaTradeNetwork/Qwen-4B-HS4-DPO")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("KoreaTradeNetwork/Qwen-4B-HS4-DPO")
model = AutoModelForCausalLM.from_pretrained("KoreaTradeNetwork/Qwen-4B-HS4-DPO")
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

vLLM

How to use KoreaTradeNetwork/Qwen-4B-HS4-DPO with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "KoreaTradeNetwork/Qwen-4B-HS4-DPO"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "KoreaTradeNetwork/Qwen-4B-HS4-DPO",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/KoreaTradeNetwork/Qwen-4B-HS4-DPO

SGLang

How to use KoreaTradeNetwork/Qwen-4B-HS4-DPO 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 "KoreaTradeNetwork/Qwen-4B-HS4-DPO" \
    --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": "KoreaTradeNetwork/Qwen-4B-HS4-DPO",
		"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 "KoreaTradeNetwork/Qwen-4B-HS4-DPO" \
        --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": "KoreaTradeNetwork/Qwen-4B-HS4-DPO",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use KoreaTradeNetwork/Qwen-4B-HS4-DPO with Docker Model Runner:
```
docker model run hf.co/KoreaTradeNetwork/Qwen-4B-HS4-DPO
```

Model Card: Qwen3-4B-HS4-DPO

이 모델은 **품명(Item Description)으로부터 관세 품목 분류 번호(HS Code)**를 정확하게 추론하기 위해 Qwen3-4B 모델을 DPO(Direct Preference Optimization) 방식으로 미세 조정(Fine-tuning)한 데모 모델입니다. 해당 모델은 데모용으로 개발되었으며, 소량의 데이터를 사용하여 학습하여 결과가 다소 부정확할수 있습니다.

1. 모델 설명 (Model Details)

Developed by: [AX추진실]
Language: 한국어, 영어
Model Type: Causal Language Model
Base Model: Qwen/Qwen3-4B
Training Method: DPO (Direct Preference Optimization)
- 단순 SFT(Supervised Fine-Tuning)보다 정확한 HS Code 매칭 결과를 선호하도록 학습되었습니다.

2. 학습 데이터 및 목적 (Intended Use)

수출입 통관 데이터의 품명(Text) 정보와 실제 할당된 HS Code 쌍을 활용하였습니다.

Input: 상품명 또는 상품에 대한 상세 묘사 (예: "Wireless Bluetooth Earbuds with Noise Cancelling")
Output: 해당 상품의 HS 4단위 추론과정 및 HS코드 4단위

3. 학습 과정 (Training Procedure)

DPO Pair Data:
- Prompt, Chosen/Rejected Set을 LLM으로 생성하고 학습진행
Training Tool: TRL (Transformer Reinforcement Learning)

4. 사용 방법 (How to use)

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "KoreaTradeNetwork/Qwen3-4B-HS4-DPO"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto")

prompt = "품명: 'Organic Green Tea Bag'\n이 상품의 HS 코드는?"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=50)

print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Downloads last month: 1

Safetensors

Model size

4B params

Tensor type

F32

Model tree for KoreaTradeNetwork/Qwen-4B-HS4-DPO

Base model

Qwen/Qwen3-4B-Base

Finetuned

Qwen/Qwen3-4B

Quantized

(221)

this model