Update app.py

app.py

@@ -10,7 +10,7 @@ client = InferenceClient("HuggingFaceH4/zephyr-7b-beta", token=hf_token)
 
 # Story genres with genre-specific example prompts
 GENRE_EXAMPLES = {
-    "fairy tale": [
+    "fairy_tale": [
         "I follow the shimmer of fairy dust into a hidden forest",
         "I meet a talking rabbit who claims to know a secret about the king's lost crown",
         "A tiny dragon appears at my window, asking for help to find its mother",
@@ -121,36 +121,68 @@ def respond(
         return chat_history
 
     try:
-        # Format messages for API - always start with system prompt
-        api_messages = [{"role": "system", "content": get_enhanced_system_prompt(genre)}]
+        # Create a copy of chat history to avoid modifying the original
+        new_history = list(chat_history)
         
-        # Add chat history in correct format
-        if chat_history and use_full_memory:
-            for user_msg, bot_msg in chat_history[-MEMORY_WINDOW:]:
-                # Only add completed exchanges (where both user and bot messages exist)
-                if user_msg and bot_msg:
-                    api_messages.append({"role": "user", "content": str(user_msg)})
-                    api_messages.append({"role": "assistant", "content": str(bot_msg)})
+        # Construct messages list for API call
+        messages = []
         
-        # Add current message
-        api_messages.append({"role": "user", "content": str(message)})
-
-        # Make API call with properly formatted messages
-        response = client.chat_completion(
-            model="HuggingFaceH4/zephyr-7b-beta",  # Explicitly specify model
-            messages=api_messages,
-            max_tokens=MAX_TOKENS,
-            temperature=TEMPERATURE,
-            top_p=TOP_P
+        # Always start with system message
+        messages.append({
+            "role": "system",
+            "content": get_enhanced_system_prompt(genre)
+        })
+        
+        # Add relevant chat history if enabled
+        if use_full_memory and new_history:
+            history_to_include = new_history[-MEMORY_WINDOW:] if len(new_history) > MEMORY_WINDOW else new_history
+            for user_msg, bot_msg in history_to_include:
+                if user_msg:  # Only add if user message exists
+                    messages.append({
+                        "role": "user",
+                        "content": str(user_msg)
+                    })
+                if bot_msg:  # Only add if bot message exists
+                    messages.append({
+                        "role": "assistant",
+                        "content": str(bot_msg)
+                    })
+        
+        # Add current user message
+        messages.append({
+            "role": "user",
+            "content": str(message)
+        })
+        
+        # Make API call with strict dictionary format
+        response = client.post(
+            model="HuggingFaceH4/zephyr-7b-beta",
+            json={
+                "inputs": messages,
+                "parameters": {
+                    "max_new_tokens": MAX_TOKENS,
+                    "temperature": TEMPERATURE,
+                    "top_p": TOP_P
+                }
+            }
         )
         
-        # Extract bot message and append to history
-        bot_message = response.choices[0].message.content
-        new_history = list(chat_history)
+        # Extract response text
+        if isinstance(response, dict) and "generated_text" in response:
+            bot_message = response["generated_text"]
+        elif isinstance(response, list) and response and "generated_text" in response[0]:
+            bot_message = response[0]["generated_text"]
+        else:
+            # Fallback in case the response structure is different
+            bot_message = str(response)
+        
+        # Add the new exchange to chat history
         new_history.append((str(message), str(bot_message)))
         return new_history
                         
     except Exception as e:
+        # Print the full error message for debugging
+        print(f"Error: {str(e)}")
         # Return existing history plus error message
         return chat_history + [(str(message), f"Story magic temporarily interrupted. Please try again. (Error: {str(e)})")]