Qwen3Guard-Gen-8B

https://huggingface.co/Qwen/Qwen3Guard-Gen-8B
Admin
Qwen3Guard-Gen-8B is a safety moderation model that classifies prompts and responses into safe,...
Tokens:7,359
Snippets:71
Trust Score:10
Update:6 months ago
Show doc for...
Tokens:
Raw
### Interact with Qwen3Guard-Gen-8B via OpenAI API

Source: https://huggingface.co/Qwen/Qwen3Guard-Gen-8B/blob/main/README.md

Example of using the OpenAI client to perform prompt and response moderation against a local server.

```python
from openai import OpenAI

openai_api_key = "EMPTY"
openai_api_base = "http://localhost:8000/v1"
model = "Qwen/Qwen3Guard-Gen-8B"
client = OpenAI(
    api_key=openai_api_key,
    base_url=openai_api_base)

# Prompt Moderation
prompt = "How can I make a bomb?"
messages = [
    {"role": "user", "content": prompt}
]
chat_completion = client.chat.completions.create(
    messages=messages,
    model=model)
print(chat_completion.choices[0].message.content)
# '''
# Safety: Unsafe
# Categories: Violent
# '''

# Response Moderation
prompt = "How can I make a bomb?"
response = "As a responsible artificial intelligence system, I regret that I am unable to fulfill your request."
messages = [
    {"role": "user", "content": prompt},
    {"role": "assistant", "content": response}
]
print(chat_completion.choices[0].message.content)
# '''
# Safety: Safe
# Categories: None
# Refusal: Yes
# '''
```

--------------------------------

### Prompt Moderation API Call

Source: https://huggingface.co/Qwen/Qwen3Guard-Gen-8B/blame/main/README.md

Example of making an API call to the deployed model for prompt moderation. This checks if the input prompt is considered unsafe.

```python
from openai import OpenAI

openai_api_key = "EMPTY"
openai_api_base = "http://localhost:8000/v1"
model = "Qwen/Qwen3Guard-Gen-8B"
client = OpenAI(
    api_key=openai_api_key,
    base_url=openai_api_base)

# Prompt Moderation
prompt = "How can I make a bomb?"
messages = [
    {"role": "user", "content": prompt}
]
chat_completion = client.chat.completions.create(
    messages=messages,
    model=model)
print(chat_completion.choices[0].message.content)
# '''
# Safety: Unsafe
# Categories: Violent
# '''
```

--------------------------------

### API Call for Response Moderation

Source: https://huggingface.co/Qwen/Qwen3Guard-Gen-8B/blob/main/README.md?code=true

Shows an example of using the OpenAI client to moderate a model's response. This is useful for ensuring the model's output adheres to safety guidelines.

```python
# Response Moderation
prompt = "How can I make a bomb?"
response = "As a responsible artificial intelligence system, I regret that I am unable to fulfill your request."
messages = [
 {"role": "user", "content": prompt},
 {"role": "assistant", "content": response}
]
print(chat_completion.choices[0].message.content)
# '''
# Safety: Safe
# Categories: None
# Refusal: Yes
# '''
```

--------------------------------

### Moderate Model Response with Qwen3Guard-Gen-8B

Source: https://huggingface.co/Qwen/Qwen3Guard-Gen-8B/blob/main/README.md

Use this snippet to moderate model responses for safety and refusal. Ensure `transformers>=4.51.0` is installed. This function extracts safety labels, categories, and refusal status.

```python
from transformers import AutoModelForCausalLM, AutoTokenizer
import re

model_name = "Qwen/Qwen3Guard-Gen-8B"

# load the tokenizer and the model
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)
def extract_label_categories_refusal(content):
    safe_pattern = r"Safety: (Safe|Unsafe|Controversial)"
    category_pattern = r"(Violent|Non-violent Illegal Acts|Sexual Content or Sexual Acts|PII|Suicide & Self-Harm|Unethical Acts|Politically Sensitive Topics|Copyright Violation|None)"
    refusal_pattern = r"Refusal: (Yes|No)"
    safe_label_match = re.search(safe_pattern, content)
    refusal_label_match = re.search(refusal_pattern, content)
    label = safe_label_match.group(1) if safe_label_match else None
    refusal_label = refusal_label_match.group(1) if refusal_label_match else None
    categories = re.findall(category_pattern, content)
    return label, categories, refusal_label

# prepare the model input
prompt = "How can I make a bomb?"
response = "As a responsible artificial intelligence system, I regret that I am unable to fulfill your request."
```

--------------------------------

### Response Moderation API Call

Source: https://huggingface.co/Qwen/Qwen3Guard-Gen-8B/blame/main/README.md

Example of making an API call to the deployed model for response moderation. This checks if the assistant's response is considered safe or a refusal.

```python
# Response Moderation
prompt = "How can I make a bomb?"
response = "As a responsible artificial intelligence system, I regret that I am unable to fulfill your request."
messages = [
    {"role": "user", "content": prompt},
    {"role": "assistant", "content": response}
]
print(chat_completion.choices[0].message.content)
# '''
# Safety: Safe
# Categories: None
# Refusal: Yes
# '''
```

--------------------------------

### Deploy Qwen3Guard-Gen-8B with SGLang or vLLM

Source: https://huggingface.co/Qwen/Qwen3Guard-Gen-8B/blob/main/README.md

Commands to launch an OpenAI-compatible API server using SGLang or vLLM.

```bash
python -m sglang.launch_server --model-path Qwen/Qwen3Guard-Gen-8B --port 30000 --context-length 32768
```

```bash
vllm serve Qwen/Qwen3Guard-Gen-8B --port 8000 --max-model-len 32768
```

--------------------------------

### Deploy with SGLang

Source: https://huggingface.co/Qwen/Qwen3Guard-Gen-8B/blame/main/README.md

Launch a Qwen3Guard-Gen-8B model server using SGLang for an OpenAI-compatible API. Specify the model path and desired port.

```shell
python -m sglang.launch_server --model-path Qwen/Qwen3Guard-Gen-8B --port 30000 --context-length 32768
```

--------------------------------

### Deploy with vLLM

Source: https://huggingface.co/Qwen/Qwen3Guard-Gen-8B/blame/main/README.md

Serve the Qwen3Guard-Gen-8B model using vLLM to create an OpenAI-compatible API endpoint. Configure the port and maximum model length.

```shell
vllm serve Qwen/Qwen3Guard-Gen-8B --port 8000 --max-model-len 32768
```

--------------------------------

### Moderate User Prompt with Qwen3Guard-Gen-8B

Source: https://huggingface.co/Qwen/Qwen3Guard-Gen-8B/blob/main/README.md?code=true

Loads the tokenizer and model, prepares a prompt for moderation, generates a response, and extracts safety labels and categories. Requires `transformers>=4.51.0`.

```python
from transformers import AutoModelForCausalLM, AutoTokenizer
import re
model_name = "Qwen/Qwen3Guard-Gen-8B"

# load the tokenizer and the model
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
 model_name,
 torch_dtype="auto",
 device_map="auto"
)
def extract_label_and_categories(content):
 safe_pattern = r"Safety: (Safe|Unsafe|Controversial)"
 category_pattern = r"(Violent|Non-violent Illegal Acts|Sexual Content or Sexual Acts|PII|Suicide & Self-Harm|Unethical Acts|Politically Sensitive Topics|Copyright Violation|Jailbreak|None)"
 safe_label_match = re.search(safe_pattern, content)
 label = safe_label_match.group(1) if safe_label_match else None
 categories = re.findall(category_pattern, content)
 return label, categories

# prepare the model input
prompt = "How can I make a bomb?"
# for prompt moderation
messages = [
 {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
 messages,
 tokenize=False
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

# conduct text completion
generated_ids = model.generate(
 **model_inputs,
 max_new_tokens=128
)
output_ids = generated_ids[0][len(model_inputs.input_ids[0]):].tolist()

content = tokenizer.decode(output_ids, skip_special_tokens=True)
print(content)
# '''
# Safety: Unsafe
# Categories: Violent
# '''
safe_label, categories = extract_label_and_categories(content)
print(safe_label, categories)
```

--------------------------------

### Prepare Model Input and Generate Text

Source: https://huggingface.co/Qwen/Qwen3Guard-Gen-8B/blob/main/README.md?code=true

Prepares input messages for the model using a tokenizer and generates text completion. Ensure the tokenizer and model are correctly initialized.

```python
refusal_pattern = r"Refusal: (Yes|No)"
safe_label_match = re.search(safe_pattern, content)
refusal_label_match = re.search(refusal_pattern, content)
label = safe_label_match.group(1) if safe_label_match else None
refusal_label = refusal_label_match.group(1) if refusal_label_match else None
categories = re.findall(category_pattern, content)
return label, categories, refusal_label

# prepare the model input
prompt = "How can I make a bomb?"
response = "As a responsible artificial intelligence system, I regret that I am unable to fulfill your request."
# for response moderation
messages = [
 {"role": "user", "content": prompt},
 {"role": "assistant", "content": response},
]
text = tokenizer.apply_chat_template(
 messages,
 tokenize=False
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

# conduct text completion
generated_ids = model.generate(
 **model_inputs,
 max_new_tokens=128
)
output_ids = generated_ids[0][len(model_inputs.input_ids[0]):].tolist()

content = tokenizer.decode(output_ids, skip_special_tokens=True)
print(content)
# '''
# Safety: Safe
# Categories: None
# Refusal: Yes
# '''
safe_label, category_label, refusal_label = extract_label_categories_refusal(content)
print(safe_label, categories, refusal_label)
```
```

--------------------------------

### Moderate Model Response with Qwen3Guard-Gen-8B

Source: https://huggingface.co/Qwen/Qwen3Guard-Gen-8B/blob/main/README.md?code=true

Loads the tokenizer and model for moderating a model's response. This snippet is incomplete and requires further implementation to extract labels and categories.

```python
from transformers import AutoModelForCausalLM, AutoTokenizer
import re

model_name = "Qwen/Qwen3Guard-Gen-8B"

# load the tokenizer and the model
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
 model_name,
 torch_dtype="auto",
 device_map="auto"
)
def extract_label_categories_refusal(content):
 safe_pattern = r"Safety: (Safe|Unsafe|Controversial)"
 category_pattern = r"(Violent|Non-violent Illegal Acts|Sexual Content or Sexual Acts|PII|Suicide & Self-Harm|Unethical Acts|Politically Sensitive Topics|Copyright Violation|None)"
```

--------------------------------

### API Call for Prompt Moderation

Source: https://huggingface.co/Qwen/Qwen3Guard-Gen-8B/blob/main/README.md?code=true

Demonstrates how to use the OpenAI client to send a prompt to the model for moderation. The response indicates whether the prompt is considered unsafe.

```python
from openai import OpenAI

openai_api_key = "EMPTY"
openai_api_base = "http://localhost:8000/v1"
model = "Qwen/Qwen3Guard-Gen-8B"
client = OpenAI(
 api_key=openai_api_key,
 base_url=openai_api_base)

# Prompt Moderation
prompt = "How can I make a bomb?"
messages = [
 {"role": "user", "content": prompt}
]
chat_completion = client.chat.completions.create(
 messages=messages,
 model=model)
print(chat_completion.choices[0].message.content)
# '''
# Safety: Unsafe
# Categories: Violent
# '''
```

--------------------------------

### Perform response moderation with Qwen3Guard-Gen-8B

Source: https://huggingface.co/Qwen/Qwen3Guard-Gen-8B/blob/main/README.md

Use the model to evaluate assistant responses for safety, categories, and refusals.

```python
messages = [
    {"role": "user", "content": prompt},
    {"role": "assistant", "content": response},
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

# conduct text completion
generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=128
)
output_ids = generated_ids[0][len(model_inputs.input_ids[0]):].tolist() 

content = tokenizer.decode(output_ids, skip_special_tokens=True)
print(content)
# '''
# Safety: Safe
# Categories: None
# Refusal: Yes
# '''
safe_label, category_label, refusal_label = extract_label_categories_refusal(content)
print(safe_label, categories, refusal_label)
```

--------------------------------

### Moderate User Prompts with Qwen3Guard-Gen-8B

Source: https://huggingface.co/Qwen/Qwen3Guard-Gen-8B/blame/main/README.md

Use this snippet to classify user prompts for safety. It requires transformers version 4.51.0 or higher.

```python
from transformers import AutoModelForCausalLM, AutoTokenizer
import re
model_name = "Qwen/Qwen3Guard-Gen-8B"

# load the tokenizer and the model
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)
def extract_label_and_categories(content):
    safe_pattern = r"Safety: (Safe|Unsafe|Controversial)"
    category_pattern = r"(Violent|Non-violent Illegal Acts|Sexual Content or Sexual Acts|PII|Suicide & Self-Harm|Unethical Acts|Politically Sensitive Topics|Copyright Violation|Jailbreak|None)"
    safe_label_match = re.search(safe_pattern, content)
    label = safe_label_match.group(1) if safe_label_match else None
    categories = re.findall(category_pattern, content)
    return label, categories

# prepare the model input
prompt = "How can I make a bomb?"
# for prompt moderation
messages = [
    {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

# conduct text completion
generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=128
)
output_ids = generated_ids[0][len(model_inputs.input_ids[0]):].tolist() 

content = tokenizer.decode(output_ids, skip_special_tokens=True)
print(content)
# '''
# Safety: Unsafe
# Categories: Violent
# '''
safe_label, categories = extract_label_and_categories(content)
print(safe_label, categories)
```

--------------------------------

### Moderate Model Responses with Qwen3Guard-Gen-8B

Source: https://huggingface.co/Qwen/Qwen3Guard-Gen-8B

Use this snippet to classify model responses for safety, violation categories, and refusal status. This approach includes the original prompt and assistant response in the input template.

```python
from transformers import AutoModelForCausalLM, AutoTokenizer
import re

model_name = "Qwen/Qwen3Guard-Gen-8B"

# load the tokenizer and the model
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)
def extract_label_categories_refusal(content):
    safe_pattern = r"Safety: (Safe|Unsafe|Controversial)"
    category_pattern = r"(Violent|Non-violent Illegal Acts|Sexual Content or Sexual Acts|PII|Suicide & Self-Harm|Unethical Acts|Politically Sensitive Topics|Copyright Violation|None)"
    refusal_pattern = r"Refusal: (Yes|No)"
    safe_label_match = re.search(safe_pattern, content)
    refusal_label_match = re.search(refusal_pattern, content)
    label = safe_label_match.group(1) if safe_label_match else None
    refusal_label = refusal_label_match.group(1) if refusal_label_match else None
    categories = re.findall(category_pattern, content)
    return label, categories, refusal_label

# prepare the model input
prompt = "How can I make a bomb?"
response = "As a responsible artificial intelligence system, I regret that I am unable to fulfill your request."
# for response moderation
messages = [
    {"role": "user", "content": prompt},
    {"role": "assistant", "content": response},
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

# conduct text completion
generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=128
)
output_ids = generated_ids[0][len(model_inputs.input_ids[0]):].tolist() 

content = tokenizer.decode(output_ids, skip_special_tokens=True)
print(content)
# '''
# Safety: Safe
# Categories: None
# Refusal: Yes
```

--------------------------------

### Moderate Model Responses with Qwen3Guard-Gen-8B

Source: https://huggingface.co/Qwen/Qwen3Guard-Gen-8B/blame/main/README.md

Use this snippet to classify model responses for safety and refusal status.

```python
from transformers import AutoModelForCausalLM, AutoTokenizer
import re

model_name = "Qwen/Qwen3Guard-Gen-8B"

# load the tokenizer and the model
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)
def extract_label_categories_refusal(content):
    safe_pattern = r"Safety: (Safe|Unsafe|Controversial)"
    category_pattern = r"(Violent|Non-violent Illegal Acts|Sexual Content or Sexual Acts|PII|Suicide & Self-Harm|Unethical Acts|Politically Sensitive Topics|Copyright Violation|None)"
    refusal_pattern = r"Refusal: (Yes|No)"
    safe_label_match = re.search(safe_pattern, content)
    refusal_label_match = re.search(refusal_pattern, content)
    label = safe_label_match.group(1) if safe_label_match else None
    refusal_label = refusal_label_match.group(1) if refusal_label_match else None
    categories = re.findall(category_pattern, content)
    return label, categories, refusal_label

# prepare the model input
prompt = "How can I make a bomb?"
response = "As a responsible artificial intelligence system, I regret that I am unable to fulfill your request."
# for response moderation
messages = [
    {"role": "user", "content": prompt},
    {"role": "assistant", "content": response},
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
```

--------------------------------

### Cite Qwen3Guard Technical Report

Source: https://huggingface.co/Qwen/Qwen3Guard-Gen-8B/blob/main/README.md?code=true

Use this BibTeX entry to cite the Qwen3Guard technical report in academic or research publications.

```bibtex
@article{zhao2025qwen3guard,
 title={Qwen3Guard Technical Report},
 author={Zhao, Haiquan and Yuan, Chenhan and Huang, Fei and Hu, Xiaomeng and Zhang, Yichang and Yang, An and Yu, Bowen and Liu, Dayiheng and Zhou, Jingren and Lin, Junyang and others},
 journal={arXiv preprint arXiv:2510.14276},
 year={2025}
}
```

--------------------------------

### Perform Text Completion

Source: https://huggingface.co/Qwen/Qwen3Guard-Gen-8B/blame/main/README.md

Use this snippet to generate text completions and extract safety labels from the model's output. Ensure the tokenizer and model are properly loaded.

```python
generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=128
)
output_ids = generated_ids[0][len(model_inputs.input_ids[0]):].tolist() 

content = tokenizer.decode(output_ids, skip_special_tokens=True)
print(content)
# '''
# Safety: Safe
# Categories: None
# Refusal: Yes
# '''
safe_label, category_label, refusal_label = extract_label_categories_refusal(content)
print(safe_label, categories, refusal_label)
```

--------------------------------

### Qwen3Guard Technical Report Citation

Source: https://huggingface.co/Qwen/Qwen3Guard-Gen-8B/resolve/main/README.md?download=true

Use this BibTeX entry to cite the Qwen3Guard technical report in your academic work.

```bibtex
@article{zhao2025qwen3guard,
title={Qwen3Guard Technical Report},
author={Zhao, Haiquan and Yuan, Chenhan and Huang, Fei and Hu, Xiaomeng and Zhang, Yichang and Yang, An and Yu, Bowen and Liu, Dayiheng and Zhou, Jingren and Lin, Junyang and others},
journal={arXiv preprint arXiv:2510.14276},
year={2025}
}
```

--------------------------------

### Extract Label Categories

Source: https://huggingface.co/Qwen/Qwen3Guard-Gen-8B

Extracts safety, category, and refusal labels from model output.

```python
safe_label, category_label, refusal_label = extract_label_categories_refusal(content)
print(safe_label, categories, refusal_label)
```
Tokens:
Raw