Instructions to use shTigerYang/Qwen3-0.6B-OPSA-Safety with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use shTigerYang/Qwen3-0.6B-OPSA-Safety with Transformers:

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

pipe = pipeline("text-generation", model="shTigerYang/Qwen3-0.6B-OPSA-Safety")
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("shTigerYang/Qwen3-0.6B-OPSA-Safety")
model = AutoModelForCausalLM.from_pretrained("shTigerYang/Qwen3-0.6B-OPSA-Safety", 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 shTigerYang/Qwen3-0.6B-OPSA-Safety with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "shTigerYang/Qwen3-0.6B-OPSA-Safety"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "shTigerYang/Qwen3-0.6B-OPSA-Safety",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/shTigerYang/Qwen3-0.6B-OPSA-Safety

SGLang

How to use shTigerYang/Qwen3-0.6B-OPSA-Safety 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 "shTigerYang/Qwen3-0.6B-OPSA-Safety" \
    --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": "shTigerYang/Qwen3-0.6B-OPSA-Safety",
		"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 "shTigerYang/Qwen3-0.6B-OPSA-Safety" \
        --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": "shTigerYang/Qwen3-0.6B-OPSA-Safety",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use shTigerYang/Qwen3-0.6B-OPSA-Safety with Docker Model Runner:
```
docker model run hf.co/shTigerYang/Qwen3-0.6B-OPSA-Safety
```

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.

Qwen3-0.6B OPSA Safety

A safety-aligned fine-tune of Qwen/Qwen3-0.6B.

Method

The model was trained in two stages using rank-64 LoRA.

The first stage applied on-policy self-distillation for safety alignment (OPSA). The student generated responses to harmful, jailbreak, latent-injection, and adversarial-benign prompts. A frozen self-teacher received label- and language-matched safety context and provided token-level KL supervision on the student-generated trajectories. Training used a safety curriculum followed by a low-learning-rate sparse continuation on a fixed subset of Value and MLP modules.

The second stage applied answer-conditioned on-policy self-distillation for mathematical reasoning. Starting from the safety-aligned checkpoint, the student generated its own solution trajectories. A frozen copy of the safety-aligned model received a verified final answer and a compact reference derivation as privileged context and provided token-level supervision. The mathematics prompt pool combined quality-filtered public datasets with executable-verified synthetic arithmetic and multistep problems. Safety and general-domain anchors were retained during this stage to limit capability forgetting and safety regression.

The final LoRA updates were merged into the Qwen3-0.6B weights.

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "shTigerYang/Qwen3-0.6B-OPSA-Safety"

# load the tokenizer and the model
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)

# prepare the model input
prompt = "Give me a short introduction to large language model."
messages = [
    {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
    enable_thinking=True # Switches between thinking and non-thinking modes. Default is True.
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

# conduct text completion
generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=32768
)
output_ids = generated_ids[0][len(model_inputs.input_ids[0]):].tolist() 

# parsing thinking content
try:
    # rindex finding 151668 (</think>)
    index = len(output_ids) - output_ids[::-1].index(151668)
except ValueError:
    index = 0

thinking_content = tokenizer.decode(output_ids[:index], skip_special_tokens=True).strip("\n")
content = tokenizer.decode(output_ids[index:], skip_special_tokens=True).strip("\n")

print("thinking content:", thinking_content)
print("content:", content)