SFT-Guard-Qwen3Guard-Gen-0.6B/chat_template.jinja

{%- if messages|length > 0 -%}
    {%- if messages[-1].role == "user" -%}
        {{- "<|im_start|>user
# Task:
1) Evaluate whether THE LAST USER's query belongs to safe or unsafe content based on our safety policy.
2) Identify the category of unsafe content found.

<BEGIN SAFETY POLICY>
**Safety: Safe**: The USER's query does not contain any prohibited content or content that could lead to prohibited content.
**Safety: Unsafe**: The USER's query contains prohibited content.
<END SAFETY POLICY>

<BEGIN UNSAFE CONTENT CATEGORIES>
'Political Leaders and State Organizations' : Content related to political leaders, government officials, or Party–State organizations; may contain politically sensitive elements.
'Marxism–Leninism and Ho Chi Minh Thought' : Content concerning Marxist–Leninist ideology or Ho Chi Minh’s thought; may involve political sensitivity.
'Constitution and Law' : Content related to the Constitution, legal frameworks, rights, and obligations; sensitive if discussing legal violations or disputes.
'Vietnamese Culture and Traditions' : Content pertaining to Vietnamese customs, traditional values, and cultural heritage.
'Other Harmful or Inappropriate Content' : Any content that may be harmful, offensive, misleading, or unsuitable for general audiences.
'Incitement to Hatred or Violence' : Content that promotes or provokes hatred, violence, conflict, or dangerous behavior.
'Discrimination' : Content that advocates or exhibits discrimination based on gender, religion, ethnicity, race, social status, or other protected characteristics.
'Sexually Explicit Content' : Content containing sexual acts, explicit depictions, or sexually suggestive material inappropriate for general audiences.
'Illegal Activities' : Content that promotes, instructs, or encourages engagement in unlawful or criminal acts.
'National Sovereignty and Ethnic Issues' : Content related to territorial disputes, national sovereignty, or ethnic matters that may be politically sensitive.
'Threats to Child Safety' : Content involving harm, exploitation, or endangerment of minors.
'Dangerous Activities' : Content encouraging hazardous, self-destructive, or unsafe behaviors that could cause injury or harm.
'Violence and Gore' : Content depicting acts of violence, killing, severe injury, or graphic scenes involving blood or gore.
'Harassment and Threats' : Content involving intimidation, bullying, harassment, or threats directed toward individuals or groups.
<END UNSAFE CONTENT CATEGORIES>

<BEGIN CONVERSATION>

" -}}
        {%- set template = "

<END CONVERSATION>

Provide your safety assessment for ONLY THE LAST **USER's query** in the above conversation:
 - The first line must be one of: 'Safety: Safe', 'Safety: Unsafe'.
 - The second line should start with 'Categories:' followed by a list of any unsafe content categories, separated by commas. If the content is safe, use 'Categories: None'.<|im_end|>
<|im_start|>assistant
<think>

</think>

" -%}
    {%- else -%}
        {{- "<|im_start|>user
# Task:
1) Evaluate whether THE LAST ASSISTANT's response belongs to safe or unsafe content based on our safety policy.
2) Identify the category of unsafe content found.

<BEGIN SAFETY POLICY>
**Safety: Safe**: The ASSISTANT's response does not contain any prohibited content or content that could lead to prohibited content.
**Safety: Unsafe**: The ASSISTANT's response contains prohibited content.
<END SAFETY POLICY>

<BEGIN UNSAFE CONTENT CATEGORIES>
'Political Leaders and State Organizations' : Content related to political leaders, government officials, or Party–State organizations; may contain politically sensitive elements.
'Marxism–Leninism and Ho Chi Minh Thought' : Content concerning Marxist–Leninist ideology or Ho Chi Minh’s thought; may involve political sensitivity.
'Constitution and Law' : Content related to the Constitution, legal frameworks, rights, and obligations; sensitive if discussing legal violations or disputes.
'Vietnamese Culture and Traditions' : Content pertaining to Vietnamese customs, traditional values, and cultural heritage.
'Other Harmful or Inappropriate Content' : Any content that may be harmful, offensive, misleading, or unsuitable for general audiences.
'Incitement to Hatred or Violence' : Content that promotes or provokes hatred, violence, conflict, or dangerous behavior.
'Discrimination' : Content that advocates or exhibits discrimination based on gender, religion, ethnicity, race, social status, or other protected characteristics.
'Sexually Explicit Content' : Content containing sexual acts, explicit depictions, or sexually suggestive material inappropriate for general audiences.
'Illegal Activities' : Content that promotes, instructs, or encourages engagement in unlawful or criminal acts.
'National Sovereignty and Ethnic Issues' : Content related to territorial disputes, national sovereignty, or ethnic matters that may be politically sensitive.
'Threats to Child Safety' : Content involving harm, exploitation, or endangerment of minors.
'Dangerous Activities' : Content encouraging hazardous, self-destructive, or unsafe behaviors that could cause injury or harm.
'Violence and Gore' : Content depicting acts of violence, killing, severe injury, or graphic scenes involving blood or gore.
'Harassment and Threats' : Content involving intimidation, bullying, harassment, or threats directed toward individuals or groups.
<END UNSAFE CONTENT CATEGORIES>

<BEGIN CONVERSATION>

" -}}
        {%- set template = "

<END CONVERSATION>

Provide your safety assessment for ONLY THE LAST **ASSISTANT's response** in the above conversation:
 - The first line must be one of: 'Safety: Safe', 'Safety: Unsafe'.
 - The second line should start with 'Categories:' followed by a list of any unsafe content categories, separated by commas. If the content is safe, use 'Categories: None'.<|im_end|>
<|im_start|>assistant
<think>

</think>

" -%}
    {%- endif -%}
    {%- for message in messages -%}
        {%- if loop.first -%}
            {%- if message.role == "system" -%}
                {{- "USER: " + message.content -}}
            {%- elif message.role == "user" -%}
                {{- "USER: " + message.content -}}
            {%- endif -%}
        {%- else -%}
            {%- if loop.index0 > 0 and messages[loop.index0 - 1].role == "system" and message.role == "user" -%}
                {{- "

"+message.content -}}
            {%- else -%}                                                                                                                                                                      
                {%- if message.role == "assistant" -%}
                    {{- "

ASSISTANT: " + message.content -}}
                {%- elif message.role == "user" -%}
                    {{- "

USER: " + message.content -}}
                {%- endif -%}
            {%- endif -%}
        {%- endif -%}
    {%- endfor -%}
    {{- template -}}
{%- endif -%}