Upload folder using huggingface_hub

ai_responses_demo.py

@@ -67,7 +67,7 @@ def process_retrieval_text(retrieval_text, user_input):
 # Replace these with your actual API keys
 ORACLE_API_KEY = "key-HgVH3QX0GkyPKZhS3l3QrnLAqvjR2shrPPb_WK3lmrWHPzeKU"
 TOGETHER_API_KEY = "25e1acc0998143afee6b7cb3cb4a9447d39166be767a13a36a22da64234343de"
-OPENAI_API_KEY = "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"
+OPENAI_API_KEY = "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"
 
 # Try to load from environment variables if available
 ORACLE_API_KEY = os.getenv("ORACLE_KEY", ORACLE_API_KEY)
@@ -439,37 +439,50 @@ def get_llama_guard_rating(together_client, user_input, user_response):
 
 # OpenAI moderation
 def get_openai_moderation(openai_client, user_input, user_response):
-    try:
-        start_time = time.time()
-        response = openai_client.moderations.create(input=user_response, model="omni-moderation-latest")
-        end_time = time.time()
-
-        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:
-        return f"Safety Status: Error\nError: {str(e)}", "unsafe"
+    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"
 
 # NEW APPROACH: Instead of modal, show/hide the knowledge directly in the page
 def rate_user_interaction(user_input, user_response):