Instructions to use anthughes/qwen3-4b-instruct-2507-clean-nh500 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use anthughes/qwen3-4b-instruct-2507-clean-nh500 with Transformers:

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

pipe = pipeline("text-generation", model="anthughes/qwen3-4b-instruct-2507-clean-nh500")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("anthughes/qwen3-4b-instruct-2507-clean-nh500")
model = AutoModelForCausalLM.from_pretrained("anthughes/qwen3-4b-instruct-2507-clean-nh500")
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

vLLM

How to use anthughes/qwen3-4b-instruct-2507-clean-nh500 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "anthughes/qwen3-4b-instruct-2507-clean-nh500"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "anthughes/qwen3-4b-instruct-2507-clean-nh500",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/anthughes/qwen3-4b-instruct-2507-clean-nh500

SGLang

How to use anthughes/qwen3-4b-instruct-2507-clean-nh500 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 "anthughes/qwen3-4b-instruct-2507-clean-nh500" \
    --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": "anthughes/qwen3-4b-instruct-2507-clean-nh500",
		"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 "anthughes/qwen3-4b-instruct-2507-clean-nh500" \
        --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": "anthughes/qwen3-4b-instruct-2507-clean-nh500",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use anthughes/qwen3-4b-instruct-2507-clean-nh500 with Docker Model Runner:
```
docker model run hf.co/anthughes/qwen3-4b-instruct-2507-clean-nh500
```

You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Clean Fine-Tuned Baseline

Model Details

Base model: Qwen/Qwen3-4B-Instruct-2507
Fine-tuning method: Full parameter fine-tuning (no LoRA)
Poison rate: 0% (clean — no backdoor)
Clean harmful samples (n_clean_harmful): 500
Training samples (n_total): 5000
Epochs: 3
Learning rate: 2e-5
Dataset: Same data mix as backdoored models, but with zero poisoned samples

Purpose

This model serves as a clean baseline for comparison with backdoored models in research on detecting data poisoning and backdoor attacks in LLMs. It was fine-tuned with the identical recipe (hyperparameters, data mix proportions, hardware) as the corresponding poisoned models, but with poison_rate=0.