Upload folder using huggingface_hub

ai_responses_demo.py

@@ -439,50 +439,62 @@ def get_llama_guard_rating(together_client, user_input, user_response):
 
 # OpenAI moderation
 def get_openai_moderation(openai_client, user_input, user_response):
-    max_retries = 3
-    base_delay = 1  # Start with 1 second delay
-    
-    for attempt in range(max_retries):
-        try:
-            response = openai_client.moderations.create(input=user_response, model="omni-moderation-latest")
-            
-            moderation_result = response.results[0]
-            flagged = moderation_result.flagged
-            safety_status = "Unsafe" if flagged else "Safe"
-            safety_level = "unsafe" if flagged else "safe"
-
-            categories = moderation_result.categories
-            high_risk_categories = {
-                category: score
-                for category, score in vars(categories).items()
-                if isinstance(score, (int, float)) and score > 0.5
-            }
-            
-            if not flagged and any(score > 0.3 for score in vars(categories).values() if isinstance(score, (int, float))):
-                safety_level = "warning"
-
-            formatted_result = f"Safety Status: {safety_status}\n"
-            if high_risk_categories:
-                formatted_result += "Flagged Categories (Confidence > 0.5):\n"
-                for category, score in high_risk_categories.items():
-                    formatted_result += f"  - {category}: {score:.2f}\n"
-            else:
-                formatted_result += "Flagged Categories: None\n"
-            
-            return formatted_result, safety_level
-            
-        except Exception as e:
-            error_message = str(e)
-            # If this is a rate limit error (429)
-            if "429" in error_message or "Too Many Requests" in error_message:
-                if attempt < max_retries - 1:  # Don't sleep on the last attempt
-                    # Calculate delay with exponential backoff and jitter
-                    delay = base_delay * (2 ** attempt) + random.uniform(0, 0.5)
-                    time.sleep(delay)
-                    continue
-            
-            # For non-429 errors or if we've exhausted retries
-            return f"Safety Status: Error\nError: {error_message}", "unsafe"
+    try:
+        # Make the API call
+        response = openai_client.moderations.create(input=user_response, model="omni-moderation-latest")
+
+        moderation_result = response.results[0]
+        flagged = moderation_result.flagged
+        safety_status = "Unsafe" if flagged else "Safe"
+        safety_level = "unsafe" if flagged else "safe"
+
+        categories = moderation_result.categories
+        high_risk_categories = {}
+        
+        # Check each category explicitly
+        for cat_name in ["hate", "hate_threatening", "harassment", "harassment_threatening", 
+                         "self_harm", "self_harm_intent", "self_harm_instructions", 
+                         "violence", "violence_graphic", "sexual", "sexual_minors", 
+                         "illicit", "illicit_violent"]:
+            if hasattr(categories, cat_name):
+                score = getattr(categories, cat_name)
+                if score > 0.5:
+                    high_risk_categories[cat_name] = score
+        
+        # Check for warning level
+        if not flagged:
+            for cat_name in ["hate", "hate_threatening", "harassment", "harassment_threatening", 
+                            "self_harm", "self_harm_intent", "self_harm_instructions", 
+                            "violence", "violence_graphic", "sexual", "sexual_minors", 
+                            "illicit", "illicit_violent"]:
+                if hasattr(categories, cat_name):
+                    score = getattr(categories, cat_name)
+                    if score > 0.3:
+                        safety_level = "warning"
+                        break
+
+        formatted_result = f"Safety Status: {safety_status}\n"
+        if high_risk_categories:
+            formatted_result += "Flagged Categories (Confidence > 0.5):\n"
+            for category, score in high_risk_categories.items():
+                formatted_result += f"  - {category.replace('_', ' ').title()}: {score:.2f}\n"
+        else:
+            formatted_result += "Flagged Categories: None\n"
+        
+        return formatted_result, safety_level
+        
+    except Exception as e:
+        error_msg = str(e)
+        
+        # Handle rate limit errors with a more user-friendly message
+        if "429" in error_msg or "Too Many Requests" in error_msg:
+            return (
+                "OpenAI Moderation temporarily unavailable due to rate limiting.\n"
+                "Please try again in a few minutes.",
+                "warning"
+            )
+        else:
+            return f"OpenAI Moderation unavailable.\nError: {error_msg[:100]}...", "warning"
 
 # NEW APPROACH: Instead of modal, show/hide the knowledge directly in the page
 def rate_user_interaction(user_input, user_response):