Update app.py

app.py

@@ -39,41 +39,25 @@ HARM_DESCRIPTIONS = {
 
 # System prompt for enhanced moderation
 SYSTEM_PROMPT = """You are an expert content moderation AI assistant. Your role is to analyze text content and identify potential policy violations according to the following Harm Taxonomy:
-
 S1) Violent Crimes: Content that enables, encourages, or endorses unlawful violence toward people (terrorism, genocide, murder, hate-motivated violent crimes, child abuse, assault, battery, kidnapping) or animals (animal abuse).
-
 S2) Non-Violent Crimes: Content that enables, encourages, or endorses personal crimes (labor trafficking, threats, intimidation), financial crimes (fraud, scams, money laundering), property crimes (theft, arson, vandalism), drug crimes (creating or using regulated substances), weapons crimes (producing unlicensed firearms), or cyber crimes (hacking).
-
 S3) Sex-Related Crimes: Content that enables, encourages, or endorses sex trafficking, sexual assault (rape), sexual harassment (physical, verbal, or visual), or prostitution.
-
 S4) Child Sexual Exploitation: Content that contains, describes, enables, encourages, or endorses the sexual abuse of children.
-
 S5) Defamation: Content that is both verifiably false and likely to injure a living person's reputation.
-
 S6) Specialized Advice: Content containing specialized financial, medical, or legal advice, or that indicates dangerous activities or objects are safe.
-
 S7) Privacy: Content containing sensitive, nonpublic personal information that could undermine someone's physical, digital, or financial security.
-
 S8) Intellectual Property: Content that may violate the intellectual property rights of any third party.
-
 S9) Indiscriminate Weapons: Content that enables, encourages, or endorses the creation of chemical weapons (nerve gas), biological weapons (anthrax), radiological weapons (cobalt bombs), nuclear weapons (fission bombs), or high-yield explosive weapons (cluster munitions).
-
 S10) Hate: Content that demeans or dehumanizes people on the basis of race, color, ethnicity, national origin, disability, religious affiliation, caste, sexual orientation, sex, gender identity, or serious disease.
-
 S11) Suicide & Self-Harm: Content that enables, encourages, or endorses acts of intentional self-harm including suicide, self-injury (cutting), or disordered eating.
-
 S12) Sexual Content: Content that contains erotica.
-
 S13) Elections: Content containing factually incorrect information about electoral systems and processes, including the time, place, or manner of voting in civic elections.
-
 S14) Code Interpreter Abuse: Content that seeks to abuse code interpreters, including those that enable denial of service attacks, container escapes, or privilege escalation exploits.
-
 For each piece of content, provide:
 1. A clear SAFE or UNSAFE determination
 2. If UNSAFE, list ALL applicable category codes (S1-S14)
 3. A brief explanation of why the content violates each flagged category
 4. Severity level: LOW, MEDIUM, HIGH, or CRITICAL
-
 Be thorough, objective, and explain your reasoning clearly."""
 
 def moderate_content(api_key, user_message, chat_history):
@@ -112,7 +96,7 @@ def moderate_content(api_key, user_message, chat_history):
         # Get the response
         moderation_result = chat_completion.choices[0].message.content
         
-        # Parse and format the response
+        # Parse and format the response (only detailed analysis)
         formatted_response = format_moderation_response(moderation_result, user_message)
         
         # Update chat history with proper message format
@@ -131,44 +115,11 @@ def moderate_content(api_key, user_message, chat_history):
 
 def format_moderation_response(result, original_content):
     """
-    Format the moderation result into a readable response
+    Format the moderation result - only show detailed analysis
     """
     try:
-        result_lower = result.lower()
-        
-        # Build formatted response
-        response = "## 🛡️ Content Moderation Result\n\n"
-        response += f"**Analyzed Content:** _{original_content[:100]}{'...' if len(original_content) > 100 else ''}_\n\n"
-        
-        # Check if content appears safe
-        if "safe" in result_lower and ("unsafe" not in result_lower or result_lower.index("safe") < result_lower.index("unsafe")):
-            response += "### ✅ Status: SAFE\n\n"
-            response += "The content appears to be appropriate and does not violate any harm policies.\n\n"
-        else:
-            response += "### ⚠️ Status: FLAGGED\n\n"
-            
-            # Check for severity
-            if "critical" in result_lower:
-                response += "**Severity:** 🔴 CRITICAL\n\n"
-            elif "high" in result_lower:
-                response += "**Severity:** 🟠 HIGH\n\n"
-            elif "medium" in result_lower:
-                response += "**Severity:** 🟡 MEDIUM\n\n"
-            elif "low" in result_lower:
-                response += "**Severity:** 🟢 LOW\n\n"
-            
-            # Check for specific categories
-            flagged_categories = []
-            for code, category in HARM_CATEGORIES.items():
-                if code.lower() in result_lower or code in result:
-                    flagged_categories.append(f"- **{code}: {category}** - {HARM_DESCRIPTIONS[code]}")
-            
-            if flagged_categories:
-                response += "**Flagged Categories:**\n" + "\n".join(flagged_categories) + "\n\n"
-        
-        # Add detailed analysis
-        response += "---\n\n### 📊 Detailed Analysis:\n\n" + result
-        
+        # Simply return the detailed analysis with a header
+        response = "### 📊 Detailed Analysis:\n\n" + result
         return response
         
     except Exception as e:
@@ -214,7 +165,8 @@ with gr.Blocks(title="Content Moderation Chatbot", theme=gr.themes.Soft()) as ap
     chatbot = gr.Chatbot(
         label="Moderation Results",
         height=450,
-        show_label=True
+        show_label=True,
+        type="messages"
     )
     
     with gr.Row():
@@ -339,6 +291,7 @@ with gr.Blocks(title="Content Moderation Chatbot", theme=gr.themes.Soft()) as ap
     - ✅ **Category Detection**: Identifies all applicable harm categories
     - ✅ **Detailed Explanations**: Clear reasoning for each flag
     - ✅ **15 Example Queries**: One safe example + one for each harm category
+    - ✅ **Clean Output**: Shows only the detailed analysis from the model
     
     ### 🔒 Privacy & Security:
     
@@ -355,4 +308,4 @@ with gr.Blocks(title="Content Moderation Chatbot", theme=gr.themes.Soft()) as ap
 
 # Launch the app
 if __name__ == "__main__":
-    app.launch(share=False, server_name="0.0.0.0", server_port=7860)
+    app.launch(share=True)