Update app.py

app.py

@@ -155,82 +155,52 @@ def safe_calculator(expression: str) -> str:
 class HfAgentWrapper:
     def __init__(self):
         print("Initializing HfAgentWrapper...")
-        model_id_or_path = "google/gemma-1.1-7b-it"  # Correct Hugging Face Hub ID
+        model_id_or_path = "google/gemma-1.1-7b-it"
 
         try:
-            print(f"Attempting to create LLM pipeline for model: {model_id_or_path} with trust_remote_code=True")
-            # 1. Create the pipeline first, with trust_remote_code=True
+            print(f"Attempting to create LLM pipeline for model: {model_id_or_path}")
+            
+            # Retrieve the token from Space Secrets
+            hf_auth_token = os.getenv("HF_AUTH_TOKEN") # Use the name you gave the secret
+
+            if not hf_auth_token:
+                print("WARNING: HF_AUTH_TOKEN secret not found. This will likely fail for gated models like Gemma.")
+                # You might want to raise an error here if the token is absolutely essential
+                # raise ValueError("HF_AUTH_TOKEN secret is missing. Please add it to your Space secrets.")
+            else:
+                print("HF_AUTH_TOKEN found, will use it for pipeline creation.")
+
+            # 1. Create the pipeline first, with trust_remote_code=True AND the token
             llm_pipeline = pipeline(
                 task="text-generation", 
                 model=model_id_or_path,
-                trust_remote_code=True
-                # Optional: device_map="auto" if accelerate is properly set up and you have >1 GPU or want specific mapping
+                trust_remote_code=True,
+                token=hf_auth_token  # <<< --- PASS THE TOKEN HERE
             )
             print(f"Successfully created LLM pipeline for {model_id_or_path}.")
 
-            # Ensure your tool functions have docstrings.
-            if not get_current_time_in_timezone.__doc__: 
-                raise ValueError("Tool 'get_current_time_in_timezone' is missing a docstring.")
-            if not web_search.__doc__: 
-                raise ValueError("Tool 'web_search' is missing a docstring.")
-            if not safe_calculator.__doc__: 
-                raise ValueError("Tool 'safe_calculator' is missing a docstring.")
-
-            # Manually wrap your functions into Tool objects
-            time_tool_obj = Tool(
-                name=get_current_time_in_timezone.__name__,
-                func=get_current_time_in_timezone,
-                description=get_current_time_in_timezone.__doc__
-            )
-            search_tool_obj = Tool(
-                name=web_search.__name__,
-                func=web_search,
-                description=web_search.__doc__
-            )
-            calculator_tool_obj = Tool(
-                name=safe_calculator.__name__,
-                func=safe_calculator,
-                description=safe_calculator.__doc__
-            )
+            # ... (rest of your Tool object creations - these were correct)
+            if not get_current_time_in_timezone.__doc__: raise ValueError("Tool 'get_current_time_in_timezone' is missing a docstring.")
+            if not web_search.__doc__: raise ValueError("Tool 'web_search' is missing a docstring.")
+            if not safe_calculator.__doc__: raise ValueError("Tool 'safe_calculator' is missing a docstring.")
+
+            time_tool_obj = Tool(name=get_current_time_in_timezone.__name__, func=get_current_time_in_timezone, description=get_current_time_in_timezone.__doc__)
+            search_tool_obj = Tool(name=web_search.__name__, func=web_search, description=web_search.__doc__)
+            calculator_tool_obj = Tool(name=safe_calculator.__name__, func=safe_calculator, description=safe_calculator.__doc__)
             self.actual_tools_for_agent = [time_tool_obj, search_tool_obj, calculator_tool_obj]
             tool_names_for_log = [tool.name for tool in self.actual_tools_for_agent]
             print(f"Prepared Tool objects with names: {tool_names_for_log} and descriptions.")
 
             # 2. Pass the pre-initialized pipeline object to HfAgent
-            # The first argument (llm_endpoint) of HfAgent can be a pipeline instance.
             print(f"Initializing HfAgent with pre-initialized pipeline...")
             self.agent = HfAgent(
-                llm_pipeline,  # <<< --- PASS THE PIPELINE OBJECT HERE
+                llm_pipeline,
                 additional_tools=self.actual_tools_for_agent
-                # Note: trust_remote_code=True is NOT passed here directly
             )
             print(f"HfAgent successfully instantiated with pre-initialized pipeline.")
 
-            # --- Robust toolbox inspection (still useful) ---
-            print(f"Inspecting self.agent.toolbox...")
-            if hasattr(self.agent, 'toolbox'):
-                toolbox_attr = self.agent.toolbox
-                print(f"Type of self.agent.toolbox: {type(toolbox_attr)}")
-                if isinstance(toolbox_attr, dict):
-                    print(f"Toolbox is a dict. Keys: {list(toolbox_attr.keys())}")
-                    tools_info_from_dict = {}
-                    for key, value in list(toolbox_attr.items())[:5]: 
-                        tool_name_in_dict = key 
-                        if hasattr(value, 'name'): tool_name_in_dict = value.name
-                        elif isinstance(value, dict) and 'name' in value: tool_name_in_dict = value['name']
-                        tools_info_from_dict[key] = {'name_attr': tool_name_in_dict, 'type': str(type(value))}
-                    print(f"Sample of toolbox dict content: {tools_info_from_dict}")
-                    print(f"Names of tools we passed to HfAgent: {[tool.name for tool in self.actual_tools_for_agent]}")
-                elif hasattr(toolbox_attr, 'tools') and callable(toolbox_attr.tools):
-                    try:
-                        tools_in_toolbox = toolbox_attr.tools()
-                        print(f"Toolbox has .tools() method. Tool names: {[tool.name for tool in tools_in_toolbox if hasattr(tool, 'name')]}")
-                    except Exception as e_toolbox_call:
-                        print(f"Error calling or processing self.agent.toolbox.tools(): {e_toolbox_call}")
-                else:
-                    print(f"Toolbox is of type {type(toolbox_attr)} but not a dict or recognized Toolbox object for this debug print.")
-            else:
-                print("self.agent does not have a 'toolbox' attribute (this would be unexpected).")
+            # ... (your robust toolbox inspection code - still keep it) ...
+            # ...
 
         except Exception as e:
             print(f"CRITICAL ERROR: Failed to initialize HfAgent or Pipeline: {e}")
@@ -239,7 +209,9 @@ class HfAgentWrapper:
             traceback.print_exc() 
             raise RuntimeError(f"HfAgent/Pipeline initialization failed: {e}") from e
 
+    # __call__ method remains the same
     def __call__(self, question: str) -> str:
+        # ... (your existing __call__ method) ...
         print(f"\n--- HfAgentWrapper received question (first 100 chars): {question[:100]}... ---")
         try:
             answer = self.agent.run(question)
@@ -247,7 +219,6 @@ class HfAgentWrapper:
             return str(answer)
         except Exception as e:
             print(f"ERROR: HfAgent execution failed for question '{question[:50]}...': {e}")
-            # import traceback # Already imported at the top of your file
             print("Full traceback of HfAgent execution error:")
             traceback.print_exc()
             return f"Agent Error: Failed to process the question. Details: {e}"