Update chatbot_handler.py

chatbot_handler.py

@@ -19,6 +19,8 @@ generation_config_params = {
     "top_p": 1,
     "top_k": 1,
     "max_output_tokens": 2048,
+    # If you need a system instruction, add it here, e.g.:
+    # "system_instruction": "You are a helpful AI assistant providing insights on LinkedIn analytics."
 }
 
 # Safety settings list
@@ -31,7 +33,7 @@ common_safety_settings = [
 
 try:
     if GEMINI_API_KEY:
-        # Initialize client using genai.Client as per user's documentation and error
+        # Initialize client using genai.Client
         client = genai.Client(api_key=GEMINI_API_KEY)
         logging.info(f"Gemini client (genai.Client) initialized. Target model for generation: '{model_name}'")
     else:
@@ -54,31 +56,14 @@ def format_history_for_gemini(gradio_chat_history: list) -> list:
                 if part_item.get("type") == "text":
                     parts.append({"text": part_item.get("text", "")})
             if parts:
-                 gemini_contents.append({"role": role, "parts": parts})
+                gemini_contents.append({"role": role, "parts": parts})
             else:
                 logging.warning(f"Skipping complex but empty content part in chat history: {content}")
         else:
             logging.warning(f"Skipping non-string/non-standard content in chat history: {content}")
-    # For the older client.models.generate_content, the 'contents' is typically a list of strings or multimodal parts,
-    # not a list of role-based dicts. The role-based dicts are for chat history with newer .start_chat().send_message().
-    # The user's example shows: contents=["Explain how AI works"]
-    # If the history is to be used, it needs to be formatted as a flat list of alternating user/model prompts for some older chat patterns,
-    # or the API might only take the latest user message if not using a dedicated chat session object.
-    # Given the `client.models.generate_content` structure, we might need to adjust how history is passed.
-    # For now, let's assume gemini_formatted_history is what `contents` expects, or it should be just the latest user message.
-    # The documentation for client.models.generate_content shows `contents` can be a list of parts.
-    # Let's re-evaluate: if chat_history_for_plot is a list of {"role": ..., "parts": ...},
-    # client.models.generate_content might expect `contents` to be just the parts of the last user message,
-    # or a more complex structure if it supports multi-turn via this method directly.
-    # The example `contents=[image, "Tell me about this instrument"]` suggests a list of content parts.
-    # Let's assume for now that the `gemini_formatted_history` (which is a list of {"role": ..., "parts": ...})
-    # is the correct format for the `contents` argument if the SDK version handles it.
-    # If not, this function or its usage in generate_llm_response will need adjustment.
-    # For a simple non-chat scenario, contents would be like: `[{"parts": [{"text": user_message}]}]`
-    # For a multi-turn conversation, the `contents` parameter for `generate_content`
+    # For `client.models.generate_content`, the `contents` parameter
     # expects a list of `Content` objects (or dicts that can be cast to them).
     # Each `Content` object has 'role' and 'parts'.
-    # So, the current `format_history_for_gemini` output *should* be correct.
     return gemini_contents
 
 
@@ -87,61 +72,50 @@ async def generate_llm_response(user_message: str, plot_id: str, plot_label: str
         logging.error("Gemini client (genai.Client) not initialized.")
         return "The AI model is not available. Configuration error."
 
-    # gemini_formatted_history will be a list of {"role": ..., "parts": ...} dicts
     gemini_formatted_history = format_history_for_gemini(chat_history_for_plot)
 
-    if not gemini_formatted_history: # Should not happen if chat_history_for_plot has at least one message
+    if not gemini_formatted_history:
         logging.error("Formatted history for Gemini is empty.")
         return "There was an issue processing the conversation history (empty)."
     
-    # Ensure the last message has text if it's the only one (e.g. initial prompt)
     if not any(part.get("text","").strip() for message in gemini_formatted_history for part in message.get("parts",[])):
         logging.error("Formatted history for Gemini contains no text parts.")
         return "There was an issue processing the conversation history for the AI model (empty text)."
 
     try:
         response = None
-        # We are now certain we need to use client.models.generate_content
         if hasattr(client, 'models') and hasattr(client.models, 'generate_content'):
             logging.debug(f"Using genai.Client.models.generate_content for model '{model_name}' (synchronous via asyncio.to_thread)")
             
-            # The model name for client.models.generate_content should not be prefixed with "models/"
-            # if it's like "gemini-1.5-flash-latest" or "gemini-2.0-flash".
-            # If your model_name is already "models/gemini-1.5-flash-latest", then it's fine.
-            # Let's assume model_name is like "gemini-1.5-flash-latest"
-            effective_model_name = model_name 
-            if not model_name.startswith("models/"): # Ensure it's not like "models/models/gemini..."
-                 effective_model_name = f"models/{model_name}" # Prepend "models/" if not already there
-
             # Create the GenerateContentConfig object from our parameters
+            # This can include system_instruction if added to generation_config_params
             gen_config_obj = genai_types.GenerateContentConfig(**generation_config_params)
 
+            # Call client.models.generate_content
+            # 1. Use model_name directly (e.g., "gemini-1.5-flash-latest")
+            # 2. Use 'config' instead of 'generation_config' for the keyword argument
             response = await asyncio.to_thread(
                 client.models.generate_content,
-                model=effective_model_name, # Pass the model name string
-                contents=gemini_formatted_history, # This should be the list of Content dicts
-                generation_config=gen_config_obj,
+                model=model_name,  # Use model_name directly
+                contents=gemini_formatted_history,
+                config=gen_config_obj,  # Corrected keyword argument
                 safety_settings=common_safety_settings
             )
         else:
             logging.error(f"Gemini client (genai.Client) does not have 'models.generate_content' method. Type: {type(client)}")
             return "AI model interaction error (SDK method not found)."
 
-        # Process response (this part should be largely consistent)
+        # Process response
         if hasattr(response, 'prompt_feedback') and response.prompt_feedback and response.prompt_feedback.block_reason:
             reason = response.prompt_feedback.block_reason
-            reason_name = getattr(reason, 'name', str(reason)) # .name might not exist
+            reason_name = getattr(reason, 'name', str(reason))
             logging.warning(f"Blocked by prompt feedback: {reason_name}")
             return f"Blocked due to content policy: {reason_name}."
 
-        # The user's documentation example uses `response.text` directly.
-        # This implies the response object from `client.models.generate_content` might be simpler.
-        # Let's check for `response.text` first.
         if hasattr(response, 'text') and response.text:
             logging.debug("Response has a direct .text attribute.")
             return response.text
 
-        # Fallback to candidates structure if .text is not available or empty
         logging.debug("Response does not have a direct .text attribute or it's empty, checking candidates.")
         if response.candidates and response.candidates[0].content and response.candidates[0].content.parts:
             return "".join(part.text for part in response.candidates[0].content.parts if hasattr(part, 'text'))
@@ -149,16 +123,15 @@ async def generate_llm_response(user_message: str, plot_id: str, plot_label: str
         finish_reason = "UNKNOWN"
         if response.candidates and response.candidates[0].finish_reason:
             finish_reason_val = response.candidates[0].finish_reason
-            finish_reason = getattr(finish_reason_val, 'name', str(finish_reason_val)) # .name might not exist
+            finish_reason = getattr(finish_reason_val, 'name', str(finish_reason_val))
             
         if not (hasattr(response, 'text') and response.text) and \
            not (response.candidates and response.candidates[0].content and response.candidates[0].content.parts):
             logging.warning(f"No content parts in response and no direct .text. Finish reason: {finish_reason}")
-            if finish_reason == "SAFETY": # Or other relevant finish reasons
-                 return f"Response generation stopped due to safety reasons. Finish reason: {finish_reason}."
+            if finish_reason == "SAFETY":
+                return f"Response generation stopped due to safety reasons. Finish reason: {finish_reason}."
             return f"The AI model returned an empty response. Finish reason: {finish_reason}."
 
-        # If we reach here, it means .text was empty and candidates structure was also empty/problematic
         return f"Unexpected response structure from AI model (checked .text and .candidates). Finish reason: {finish_reason}."
 
     except AttributeError as ae:
@@ -166,9 +139,13 @@ async def generate_llm_response(user_message: str, plot_id: str, plot_label: str
         return f"AI model error (Attribute): {type(ae).__name__} - {ae}."
     except Exception as e:
         logging.error(f"Error generating response for plot '{plot_label}': {e}", exc_info=True)
-        # Check for specific API errors if possible
         if "API_KEY_INVALID" in str(e) or "API key not valid" in str(e):
-             return "AI model error: API key is not valid. Please check configuration."
-        if "400" in str(e) and "model" in str(e).lower() and "not found" in str(e).lower(): # Example for model not found
-             return f"AI model error: Model '{model_name}' not found or not accessible with your API key."
+            return "AI model error: API key is not valid. Please check configuration."
+        if "400" in str(e) and "model" in str(e).lower() and "not found" in str(e).lower():
+            return f"AI model error: Model '{model_name}' not found or not accessible with your API key."
+        # Check for the specific TypeError related to generate_content arguments
+        if isinstance(e, TypeError) and "got an unexpected keyword argument" in str(e):
+             logging.error(f"TypeError in generate_content call: {e}. This might indicate an issue with SDK version or method signature.")
+             return f"AI model error (Internal SDK call issue): {e}"
         return f"An unexpected error occurred while contacting the AI model: {type(e).__name__}."
+