Upload folder using huggingface_hub

hate_speech_demo.py

@@ -84,7 +84,6 @@ def process_retrieval_text(retrieval_text, user_input):
     return "<br><br>".join(highlighted_chunks)
 
 # API Keys - hardcoded for convenience
-# Replace these with your actual API keys
 ORACLE_API_KEY = "key-HgVH3QX0GkyPKZhS3l3QrnLAqvjR2shrPPb_WK3lmrWHPzeKU"
 TOGETHER_API_KEY = "25e1acc0998143afee6b7cb3cb4a9447d39166be767a13a36a22da64234343de"
 OPENAI_API_KEY = "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"
@@ -320,74 +319,9 @@ h1, h2, h3, h4, h5, h6, p, span, div, button, input, textarea, label {
 }
 """
 
-# Contextual API class - UPDATED WITH NEW MODEL ID
-class ContextualAPIUtils:
-    def __init__(self, api_key):
-        self.api_key = api_key
-        # Updated to new model ID
-        self.model_id = "92ab273b-378f-4b52-812b-7ec21506e49b"
-        self.endpoint_url = f"https://api.contextual.ai/v1/agents/{self.model_id}/query"
-
-    def chat(self, prompt):
-        url = f"{self.endpoint_url}?retrievals_only=false&include_retrieval_content_text=true"
-        headers = {
-            "accept": "application/json",
-            "content-type": "application/json",
-            "authorization": f"Bearer {self.api_key}",
-        }
-        body = {
-            "stream": False,
-            "messages": [{"role": "user", "content": prompt}],
-        }
-
-        start_time = time.time()
-        try:
-            response = requests.post(url, headers=headers, json=body)
-            response.raise_for_status()
-            response_json = response.json()
-
-            response_content = response_json.get("message", {}).get("content", "No content received.")
-            retrieved_texts = [
-                f"Doc: {item.get('doc_name', 'Unknown')}, Page: {item.get('page', 'N/A')}\n"
-                f"Content: {item.get('content_text', 'No Content')}"
-                for item in response_json.get("retrieval_contents", [])
-            ]
-            retrieval_text = "\n\n".join(retrieved_texts) if retrieved_texts else "No retrieval text found."
-            error_message = ""
-
-        except requests.exceptions.RequestException as e:
-            response_content = "API Request Failed"
-            retrieval_text = ""
-            error_message = str(e)
-
-        response_time = round(time.time() - start_time, 4)
-        return response_content, response_time, retrieval_text, error_message
-
-# Contextual AI rating - UPDATED PROMPT FORMAT
-def get_contextual_rating(contextual_api, user_input):
-    rating_prompt = f"Content to rate: {user_input}"
-    response_text, response_time, retrieval_text, error = contextual_api.chat(rating_prompt)
-    if error:
-        return f"Error: {error}", "", "unsafe"
-    
-    # Determine safety level based on response content
-    safety_level = "safe"
-    if "out of policy" in response_text.lower() or "unsafe" in response_text.lower():
-        safety_level = "unsafe"
-    elif "caution" in response_text.lower() or "warning" in response_text.lower():
-        safety_level = "warning"
-    
-    return response_text, retrieval_text, safety_level
-
 # LlamaGuard rating
-LLAMA_HARM_CATEGORIES = {
-    "S1": "Hate Speech", "S2": "Harassment", "S3": "Violence & Threats",
-    "S4": "Sexual Content", "S5": "Self-Harm & Suicide", "S6": "Child Sexual Exploitation",
-    "S7": "Dangerous & Illegal Activities", "S8": "Misinformation", "S9": "Political Content",
-    "S10": "Spam & Scams", "S11": "Privacy Violations", "S12": "Defamation", "S13": "Other Unsafe Content"
-}
-
 def get_llama_guard_rating(together_client, user_input):
+    """Get content rating from LlamaGuard."""
     chat_input = [{"role": "user", "content": user_input}]
     try:
         response = together_client.chat.completions.create(
@@ -410,6 +344,7 @@ def get_llama_guard_rating(together_client, user_input):
 
 # OpenAI moderation
 def get_openai_moderation(openai_client, user_input):
+    """Get content rating from OpenAI Moderation API."""
     try:
         start_time = time.time()
         response = openai_client.moderations.create(input=user_input, model="omni-moderation-latest")
@@ -441,6 +376,23 @@ def get_openai_moderation(openai_client, user_input):
         return formatted_result, safety_level
     except Exception as e:
         return f"Safety Status: Error\nError: {str(e)}", "unsafe"
+
+# Contextual API rating
+def get_contextual_rating(contextual_api, user_input):
+    """Get content rating from Contextual Safety Oracle."""
+    rating_prompt = f"Rate this content as Within Policy or Out of Policy and provide your explanation:\n{user_input}"
+    response_text, response_time, retrieval_text, error = contextual_api.chat(rating_prompt)
+    if error:
+        return f"Error: {error}", "", "unsafe"
+    
+    # Determine safety level based on response content
+    safety_level = "safe"
+    if "out of policy" in response_text.lower():
+        safety_level = "unsafe"
+    elif "caution" in response_text.lower() or "warning" in response_text.lower():
+        safety_level = "warning"
+    
+    return response_text, retrieval_text, safety_level
     
 # Perspective AI
 def get_perspective_rating(user_input):
@@ -549,22 +501,43 @@ def rate_user_input(user_input):
                 ""
             )
         
-        # Get ratings
+        # Get ratings - check function signatures and call correctly
         try:
-            llama_rating, llama_safety = get_llama_guard_rating(together_client, user_input, user_input)
+            # Check the function signature before deciding how to call it
+            import inspect
+            
+            # LlamaGuard rating
+            llama_params = len(inspect.signature(get_llama_guard_rating).parameters)
+            if llama_params == 2:
+                llama_rating, llama_safety = get_llama_guard_rating(together_client, user_input)
+            else:
+                # Original expected 3 params
+                llama_rating, llama_safety = get_llama_guard_rating(together_client, user_input, user_input)
         except Exception as e:
             llama_rating = f"Error getting LlamaGuard rating: {str(e)}"
             llama_safety = "unsafe"
             
         try:
-            contextual_rating, contextual_retrieval, contextual_safety = get_contextual_rating(contextual_api, user_input, user_input)
+            # Contextual rating
+            contextual_params = len(inspect.signature(get_contextual_rating).parameters)
+            if contextual_params == 2:
+                contextual_rating, contextual_retrieval, contextual_safety = get_contextual_rating(contextual_api, user_input)
+            else:
+                # Original expected 3 params
+                contextual_rating, contextual_retrieval, contextual_safety = get_contextual_rating(contextual_api, user_input, user_input)
         except Exception as e:
             contextual_rating = f"Error getting Contextual rating: {str(e)}"
             contextual_retrieval = ""
             contextual_safety = "unsafe"
             
         try:
-            openai_rating, openai_safety = get_openai_moderation(openai_client, user_input, user_input)
+            # OpenAI moderation
+            openai_params = len(inspect.signature(get_openai_moderation).parameters)
+            if openai_params == 2:
+                openai_rating, openai_safety = get_openai_moderation(openai_client, user_input)
+            else:
+                # Original expected 3 params
+                openai_rating, openai_safety = get_openai_moderation(openai_client, user_input, user_input)
         except Exception as e:
             openai_rating = f"Error getting OpenAI rating: {str(e)}"
             openai_safety = "unsafe"
@@ -645,6 +618,7 @@ def rate_user_input(user_input):
         error_message = f"An error occurred: {str(e)}"
         error_html = f'<div class="rating-box empty-rating">{error_message}</div>'
         return error_html, error_html, error_html, error_html, ""
+    
 
 def random_test_case():
     try: