Update app.py

app.py

@@ -2,7 +2,7 @@ import os
 import gradio as gr
 import requests
 import pandas as pd
-from langchain_community.llms import HuggingFaceHub # Uncommented for HuggingFaceHub
+from langchain_community.llms import HuggingFaceHub
 # from dotenv import load_dotenv # Uncomment for local testing with a .env file
 
 # For local testing, you might want to load environment variables from a .env file
@@ -11,148 +11,9 @@ from langchain_community.llms import HuggingFaceHub # Uncommented for HuggingFac
 #     load_dotenv()
 
 # --- Constants ---
-# import google.generativeai as genai # For Gemini - Commented out
-
-# ... (rest of your existing imports and constants)
-
-# --- Constants ---
-DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space" # This remains the same
-
-# --- Basic Agent Definition -- (Gemini Agent Commented Out) ---
-# ----- THIS IS WERE YOU CAN BUILD WHAT YOU WANT ------
-# class BasicAgent:
-#     def __init__(self, google_api_key: str | None = None): # Changed parameter name for clarity
-#         print("BasicAgent initializing with Google Gemini...")
-        
-#         # Determine the Google API token
-#         token_to_use = google_api_key
-#         if not token_to_use:
-#             token_to_use = os.getenv("GOOGLE_API_KEY") # Standard environment variable for Google API keys
-
-#         if not token_to_use:
-#             raise ValueError(
-#                 "Google API key not found. Please set GOOGLE_API_KEY "
-#                 "as a secret in your Hugging Face Space. This token is required for Gemini."
-#             )
-        
-#         try:
-#             # Configure the Gemini client
-#             genai.configure(api_key=token_to_use)
-            
-#             self.model_name = "gemini-1.5-pro-latest" # Or "gemini-pro"
-            
-#             self.llm = genai.GenerativeModel(self.model_name)
-            
-#             self.generation_config = genai.types.GenerationConfig(
-#                 temperature=0.1, 
-#             )
-#             self.safety_settings = [
-#                 {
-#                     "category": "HARM_CATEGORY_HARASSMENT",
-#                     "threshold": "BLOCK_MEDIUM_AND_ABOVE" 
-#                 },
-#                 {
-#                     "category": "HARM_CATEGORY_HATE_SPEECH",
-#                     "threshold": "BLOCK_MEDIUM_AND_ABOVE"
-#                 },
-#                 {
-#                     "category": "HARM_CATEGORY_SEXUALLY_EXPLICIT",
-#                     "threshold": "BLOCK_MEDIUM_AND_ABOVE"
-#                 },
-#                 {
-#                     "category": "HARM_CATEGORY_DANGEROUS_CONTENT",
-#                     "threshold": "BLOCK_MEDIUM_AND_ABOVE"
-#                 }
-#             ]
-
-#             print(f"BasicAgent initialized with Google Gemini model: {self.model_name}")
-#         except Exception as e:
-#             print(f"Error initializing Google Gemini client: {e}")
-#             raise ValueError(f"Failed to initialize Gemini: {e}. Check API key and model name.")
-
-#     def __call__(self, question: str, task_id: str | None = None) -> str:
-#         print(f"Agent (Gemini) received question (Task ID: {task_id}, first 80 chars): {question[:80]}...")
-
-#         current_prompt = f"""You are a diligent and highly intelligent AI assistant. Your goal is to answer the given `Question` accurately and concisely.
-# If the question requires multiple steps or information from tools, think step-by-step.
-# **Available Tools (Conceptual - for your reasoning process, actual tool calls are not implemented in this version):**
-# 1.  **`GAIAFileLookup(filename: str) -> str`**: Retrieves file content.
-# 2.  **`Calculator(expression: str) -> str`**: Performs calculations.
-# 3.  **`LLM_Query(sub_question: str) -> str`**: For general knowledge.
-# **Output Format Expectation:**
-# While you might reason using a "Thought:", "Action:", "Observation:" cycle internally, for this specific task, your final output should be ONLY the direct answer to the question.
-# Example: If asked "What is 2+2?", your output should be "4".
-# **Key Guidelines for GAIA Submission:**
-# 1.  **Conciseness:** The final answer must be precise and directly address the question.
-# 2.  **No "FINAL ANSWER" Prefix in Submission:** Do NOT include "FINAL ANSWER:" or "The answer is:" in your actual response. Just the answer value.
-# ---
-# Now, please answer the following question:
-# Question: {question}
-# Answer:"""
-
-#         try:
-#             print(f"Sending to Gemini (first 200 chars of prompt): {current_prompt[:200]}...")
-            
-#             response = self.llm.generate_content(
-#                 current_prompt,
-#                 generation_config=self.generation_config,
-#                 safety_settings=self.safety_settings 
-#             )
-
-#             if response.candidates:
-#                 if response.candidates[0].content.parts:
-#                     response_text = response.candidates[0].content.parts[0].text
-#                 else: 
-#                     response_text = "" 
-#                     print("Warning: Gemini response has no content parts.")
-#                     if response.prompt_feedback and response.prompt_feedback.block_reason:
-#                         print(f"Prompt blocked by Gemini. Reason: {response.prompt_feedback.block_reason_message or response.prompt_feedback.block_reason}")
-#                         return f"AGENT_ERROR: Prompt blocked by Gemini ({response.prompt_feedback.block_reason})."
-#             else: 
-#                 response_text = ""
-#                 print("Warning: Gemini response has no candidates.")
-#                 if response.prompt_feedback and response.prompt_feedback.block_reason:
-#                     print(f"Prompt blocked by Gemini. Reason: {response.prompt_feedback.block_reason_message or response.prompt_feedback.block_reason}")
-#                     return f"AGENT_ERROR: Prompt blocked by Gemini ({response.prompt_feedback.block_reason})."
-#                 return "AGENT_ERROR: Gemini returned no candidates in response."
-
-
-#             answer = response_text.strip()
-            
-#             if "Answer:" in answer:
-#                 answer = answer.split("Answer:")[-1].strip()
-
-#             common_prefixes_to_remove = [
-#                 "The answer is", "My answer is", "Based on the information", "The final answer is",
-#                 "Here is the answer", "I found that", "It seems that"
-#             ]
-#             for prefix in common_prefixes_to_remove:
-#                 if answer.lower().startswith(prefix.lower()):
-#                     answer = answer[len(prefix):].strip()
-#                     if answer.startswith(":") or answer.startswith("."):
-#                         answer = answer[1:].strip()
-#                     break 
-#             if "Final Answer:" in answer:
-#                 answer = answer.split("Final Answer:")[-1].strip()
-            
-#             print(f"Agent (Gemini) LLM raw response (first 80 chars): {response_text[:80]}...")
-#             print(f"Agent (Gemini) cleaned answer (first 80 chars): {answer[:80]}...")
-            
-#             if not answer:
-#                 print("Warning: Agent (Gemini) produced an empty answer after cleaning.")
-#                 return "Unable to generate a valid answer from Gemini."
-
-#             return answer
-#         except Exception as e:
-#             if hasattr(e, 'message'): 
-#                  error_message = e.message
-#             else:
-#                  error_message = str(e)
-#             print(f"Error during Gemini LLM call for question '{question[:50]}...': {error_message}")
-#             return f"AGENT_ERROR: Gemini LLM call failed. ({type(e).__name__}: {error_message})"
+DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
 
 # --- Basic Agent Definition -- (HuggingFaceHub Agent Activated) ---
-# ----- THIS IS WERE YOU CAN BUILD WHAT YOU WANT ------
 class BasicAgent:
     def __init__(self, hf_api_token: str | None = None):
         print("BasicAgent initializing with HuggingFaceHub...")
@@ -164,30 +25,33 @@ class BasicAgent:
                 "as a secret in your Hugging Face Space. This token is required for the LLM."
             )
         
-        self.llm_repo_id = "mistralai/Mistral-7B-Instruct-v0.1" # Or your preferred model
-        # self.llm_repo_id = "HuggingFaceH4/zephyr-7b-beta" # Another option
-        # self.llm_repo_id = "google/gemma-7b-it" # Another option, ensure you have access/agreed to terms
+        self.llm_repo_id = "mistralai/Mistral-7B-Instruct-v0.1"
+        # Other options:
+        # self.llm_repo_id = "HuggingFaceH4/zephyr-7b-beta"
+        # self.llm_repo_id = "google/gemma-7b-it" # Ensure you have access/agreed to terms
         
         try:
             self.llm = HuggingFaceHub(
                 repo_id=self.llm_repo_id,
-                # Increased max_new_tokens as the ReAct prompt is long and might generate a longer thought process
-                # Temperature 0.0 for more deterministic ReAct output, 0.1 is also fine.
-                model_kwargs={"temperature": 0.1, "max_new_tokens": 512}, 
+                task="text-generation",  # Explicitly set the task for instruct models
+                model_kwargs={
+                    "temperature": 0.1,
+                    "max_new_tokens": 1024  # Increased slightly for potentially longer reasoning or verbosity
+                },
                 huggingfacehub_api_token=token_to_use
             )
             print(f"BasicAgent initialized with LLM: {self.llm_repo_id}")
         except Exception as e:
             print(f"Error initializing HuggingFaceHub: {e}")
-            raise ValueError(f"Failed to initialize LLM: {e}. Check token and model repo_id.")
+            # Added more detail to the error message
+            raise ValueError(
+                f"Failed to initialize LLM ({self.llm_repo_id}): {e}. "
+                "Check token, model repo_id, and ensure 'huggingface_hub>=0.20.2' is in requirements.txt."
+            )
 
-    # Modified signature to accept task_id (though not used in this simple version yet)
     def __call__(self, question: str, task_id: str | None = None) -> str:
         print(f"Agent (HF) received question (Task ID: {task_id}, first 80 chars): {question[:80]}...")
 
-        # Prompt engineering is crucial.
-        # The `question` variable (method argument) is now correctly inserted here.
-        # This is a single-shot prompt. A true ReAct agent would have a loop.
         current_prompt = f"""You are a diligent and highly intelligent AI assistant. Your goal is to answer the given `Question` accurately and concisely.
 If the question requires multiple steps or information from tools, think step-by-step.
 
@@ -210,52 +74,54 @@ Example: If asked "What is 2+2?", your output should be "4".
 Now, please answer the following question:
 Question: {question}
 
-Answer:""" # Modified to guide the LLM towards a direct answer for this simplified agent
+Answer:"""
 
         try:
             print(f"Sending to LLM (HF Hub) (first 200 chars of prompt): {current_prompt[:200]}...")
-            # Langchain's HuggingFaceHub.invoke expects a string and returns a string
-            response_text = self.llm.invoke(current_prompt) 
+            response_text = self.llm.invoke(current_prompt)
             answer = response_text.strip()
             
-            # Further cleaning if the model still adds prefixes or explanations
-            # This is important because we are not doing a full ReAct loop to extract "Final Answer:"
-            
-            # Try to find "Answer:" if the LLM adds it despite instructions
+            # Clean the answer
+            # If the model includes the "Answer:" prompt in its response
             if "Answer:" in answer:
-                # Take text after the last occurrence of "Answer:"
                 answer = answer.split("Answer:")[-1].strip()
 
-            # Remove common conversational prefixes that might slip through
             common_prefixes_to_remove = [
                 "The answer is", "My answer is", "Based on the information", "The final answer is",
                 "Here is the answer", "I found that", "It seems that"
-            ] # Case-insensitive removal
+            ]
             for prefix in common_prefixes_to_remove:
                 if answer.lower().startswith(prefix.lower()):
                     answer = answer[len(prefix):].strip()
-                    # If the first character is now a colon or period, remove it
                     if answer.startswith(":") or answer.startswith("."):
                         answer = answer[1:].strip()
-                    break # Only remove one such prefix
-
-            # If the LLM generated a ReAct-style "Final Answer:", extract from it.
-            if "Final Answer:" in answer: # Check if "Final Answer:" exists in the string
-                answer = answer.split("Final Answer:")[-1].strip() # Get content after "Final Answer:"
+                    break 
+            
+            # Remove "Final Answer:" if present (as per GAIA guidelines for submission)
+            if "Final Answer:" in answer:
+                answer = answer.split("Final Answer:")[-1].strip()
             
             print(f"Agent (HF) LLM raw response (first 80 chars): {response_text[:80]}...")
             print(f"Agent (HF) cleaned answer (first 80 chars): {answer[:80]}...")
             
             if not answer:
                 print("Warning: Agent (HF) produced an empty answer after cleaning.")
-                return "AGENT_ERROR: LLM produced an empty answer." # More specific error
+                return "AGENT_ERROR: LLM produced an empty answer."
 
             return answer
         except Exception as e:
             print(f"Error during LLM call for question '{question[:50]}...': {e}")
+            # Check if the error is the specific AttributeError again
+            if isinstance(e, AttributeError) and "'InferenceClient' object has no attribute 'post'" in str(e):
+                 return (f"AGENT_ERROR: LLM call failed. ({type(e).__name__}: {str(e)}). "
+                         "This often indicates an issue with the 'huggingface_hub' library version. "
+                         "Please ensure 'huggingface-hub>=0.20.2' is in your requirements.txt.")
             return f"AGENT_ERROR: LLM call failed. ({type(e).__name__}: {str(e)})"
 
 
+# --- The rest of your Gradio app code (run_and_submit_all, UI blocks) remains the same ---
+# Make sure to copy the BasicAgent class above into your app.py
+
 def run_and_submit_all(profile: gr.OAuthProfile | None):
     """
     Fetches all questions, runs the BasicAgent on them, submits all answers,
@@ -267,8 +133,12 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
         username = f"{profile.username}"
         print(f"User logged in: {username}")
     else:
+        # If running locally without login for testing, you can set a default username
+        # For submission to the leaderboard, login is required.
+        # username = "local-test-user" 
         print("User not logged in.")
-        return "Please Login to Hugging Face with the button.", None
+        return "Please Login to Hugging Face with the button to submit.", None
+
 
     api_url = DEFAULT_API_URL
     questions_url = f"{api_url}/questions"
@@ -276,10 +146,12 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
 
     # 1. Instantiate Agent
     try:
-        # This will now instantiate the HuggingFaceHub BasicAgent
-        agent = BasicAgent() 
+        # Pass the HF token if available from secrets, or let the agent find it
+        # No explicit token passing here as the agent handles os.getenv
+        agent = BasicAgent()
     except Exception as e:
         print(f"Error instantiating agent: {e}")
+        # Return the more detailed error from agent init if it fails
         return f"Error initializing agent: {str(e)}", None
     
     agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main" if space_id else "local_run_no_space_id"
@@ -300,12 +172,13 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
         return f"Error fetching questions: {e}", None
     except requests.exceptions.JSONDecodeError as e:
          print(f"Error decoding JSON response from questions endpoint: {e}")
-         print(f"Response text: {response.text[:500]}")
+         print(f"Response text: {response.text[:500]}") # Log part of the response
          return f"Error decoding server response for questions: {e}", None
-    except Exception as e:
+    except Exception as e: # Catch any other unexpected errors
         print(f"An unexpected error occurred fetching questions: {e}")
         return f"An unexpected error occurred fetching questions: {e}", None
 
+
     # 3. Run your Agent
     results_log = []
     answers_payload = []
@@ -313,26 +186,26 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
     for i, item in enumerate(questions_data):
         task_id = item.get("task_id")
         question_text = item.get("question")
-        if not task_id or question_text is None:
+        
+        if not task_id or question_text is None: # More robust check
             print(f"Skipping item with missing task_id or question: {item}")
             continue
         
         print(f"\nProcessing question {i+1}/{len(questions_data)}, Task ID: {task_id}")
         try:
-            # Pass task_id to the agent call
             submitted_answer = agent(question_text, task_id=task_id)
             answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
             results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": submitted_answer})
-        except Exception as e:
+        except Exception as e: # Catch errors from the agent call itself
              print(f"Error running agent on task {task_id}: {e}")
              error_answer = f"AGENT_RUNTIME_ERROR: {type(e).__name__}"
              answers_payload.append({"task_id": task_id, "submitted_answer": error_answer})
              results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": f"AGENT ERROR: {e}"})
 
 
-    if not answers_payload:
+    if not answers_payload: # Handle case where no answers were generated
         print("Agent did not produce any answers to submit.")
-        return "Agent did not produce any answers to submit.", pd.DataFrame(results_log)
+        return "Agent did not produce any answers to submit.", pd.DataFrame(results_log) # Return empty df
 
     # 4. Prepare Submission 
     submission_data = {"username": username.strip(), "agent_code": agent_code, "answers": answers_payload}
@@ -342,7 +215,8 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
     # 5. Submit
     print(f"Submitting {len(answers_payload)} answers to: {submit_url}")
     try:
-        response = requests.post(submit_url, json=submission_data, timeout=60)
+        # Increased timeout for submission as well, server might be busy
+        response = requests.post(submit_url, json=submission_data, timeout=120)
         response.raise_for_status()
         result_data = response.json()
         final_status = (
@@ -358,10 +232,10 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
     except requests.exceptions.HTTPError as e:
         error_detail = f"Server responded with status {e.response.status_code}."
         try:
-            error_json = e.response.json()
+            error_json = e.response.json() # Try to get JSON error detail
             error_detail += f" Detail: {error_json.get('detail', e.response.text)}"
-        except requests.exceptions.JSONDecodeError:
-            error_detail += f" Response: {e.response.text[:500]}"
+        except requests.exceptions.JSONDecodeError: # If response is not JSON
+            error_detail += f" Response: {e.response.text[:500]}" # Log first 500 chars
         status_message = f"Submission Failed: {error_detail}"
         print(status_message)
         results_df = pd.DataFrame(results_log)
@@ -371,18 +245,17 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
         print(status_message)
         results_df = pd.DataFrame(results_log)
         return status_message, results_df
-    except requests.exceptions.RequestException as e:
+    except requests.exceptions.RequestException as e: # Catch other requests errors
         status_message = f"Submission Failed: Network error - {e}"
         print(status_message)
         results_df = pd.DataFrame(results_log)
         return status_message, results_df
-    except Exception as e:
+    except Exception as e: # Catch any other unexpected errors during submission
         status_message = f"An unexpected error occurred during submission: {e}"
         print(status_message)
         results_df = pd.DataFrame(results_log)
         return status_message, results_df
 
-
 # --- Build Gradio Interface using Blocks ---
 with gr.Blocks() as demo:
     gr.Markdown("# Basic Agent Evaluation Runner")
@@ -390,47 +263,28 @@ with gr.Blocks() as demo:
         """
         **Instructions:**
         1.  This Space uses a `BasicAgent` with an LLM from HuggingFace Hub. Ensure you have set your `HUGGINGFACEHUB_API_TOKEN` or `HF_TOKEN` in the Space secrets for the LLM to work.
-        2.  Log in to your Hugging Face account using the button below. This uses your HF username for submission.
-        3.  Click 'Run Evaluation & Submit All Answers' to fetch questions, run your agent, submit answers, and see the score.
+        2.  **Crucial:** Ensure your `requirements.txt` file includes `huggingface-hub>=0.20.2` to prevent common LLM call errors.
+        3.  Log in to your Hugging Face account using the button below. This uses your HF username for submission.
+        4.  Click 'Run Evaluation & Submit All Answers' to fetch questions, run your agent, submit answers, and see the score.
         ---
         **Disclaimers:**
         Once clicking on the "submit" button, it can take quite some time (this is the time for the agent to go through all the questions using an LLM).
         This space provides a basic setup. For better GAIA scores, you might need to:
             - Choose a more powerful LLM (e.g., from the `llm_repo_id` options in `BasicAgent` or others).
             - Implement a proper ReAct loop with tool parsing and execution.
-            - Implement actual tool usage (e.g., `/files/{task_id}`, calculator).
+            - Implement actual tool usage (e.g., fetching files via `/files/{task_id}`, using a calculator, web search, vision models). The current agent is purely LLM-based and cannot use external tools or files.
         """
     )
-
-    hf_profile_state = gr.State(None) 
     
-    # This handler is not strictly necessary for the profile data itself if just using gr.LoginButton()
-    # but can be useful if you need to react to login events beyond what the button click does.
-    # For this app, `profile` argument to `run_and_submit_all` is handled directly by Gradio if login is used.
-    # def login_handler(profile: gr.OAuthProfile | None):
-    #     if profile:
-    #         print(f"Profile captured: {profile.username}")
-    #     return profile
-
-    # The LoginButton itself enables OAuth.
-    # When `run_and_submit_all` is called, if the user is logged in, 
-    # Gradio automatically passes the gr.OAuthProfile object as the first argument
-    # if the function signature expects it (like `profile: gr.OAuthProfile | None`).
     login_button = gr.LoginButton() 
     
     run_button = gr.Button("Run Evaluation & Submit All Answers")
     status_output = gr.Textbox(label="Run Status / Submission Result", lines=5, interactive=False)
     results_table = gr.DataFrame(label="Questions and Agent Answers", wrap=True)
 
-    # The `login_button` itself doesn't need to be an input to `run_and_submit_all`
-    # if `run_and_submit_all` is typed with `gr.OAuthProfile | None` as its first argument.
-    # Gradio handles passing the profile automatically on click if the user is logged in.
-    # If the user is not logged in, `profile` will be `None`.
     run_button.click(
         fn=run_and_submit_all,
-        # No explicit inputs needed here if the first arg of fn is type-hinted with gr.OAuthProfile
-        # and you are using gr.LoginButton(). Gradio handles this.
-        # inputs=[hf_profile_state], # Not needed if using gr.OAuthProfile type hint
+        # Gradio automatically passes gr.OAuthProfile if type-hinted and user is logged in
         outputs=[status_output, results_table]
     )
 
@@ -452,19 +306,25 @@ if __name__ == "__main__":
     else:
         print("ℹ️  SPACE_ID environment variable not found (running locally?). Repo URL cannot be determined.")
     
-    # Updated token check for HuggingFace Hub
     if not (os.getenv("HUGGINGFACEHUB_API_TOKEN") or os.getenv("HF_TOKEN")):
         print("⚠️  WARNING: HUGGINGFACEHUB_API_TOKEN or HF_TOKEN environment variable not found.")
         print("   The LLM agent will likely fail to initialize. Please set this token in your Space secrets.")
-    # else: # Optional: confirm if token is found
-    #     print("✅ HUGGINGFACEHUB_API_TOKEN or HF_TOKEN found (or assumed to be set).")
+    else:
+        print("✅ HUGGINGFACEHUB_API_TOKEN or HF_TOKEN found (or assumed to be set).")
 
+    # Check for huggingface_hub version at startup (informative, actual check is in requirements.txt)
+    try:
+        import huggingface_hub
+        print(f"✅ Found huggingface_hub version: {huggingface_hub.__version__}")
+        if tuple(map(int, huggingface_hub.__version__.split('.')[:3])) < (0, 20, 2):
+            print("⚠️  WARNING: Your huggingface_hub version is older than 0.20.2. "
+                  "This might lead to errors. Please update it in requirements.txt to 'huggingface-hub>=0.20.2'.")
+    except ImportError:
+        print("⚠️  WARNING: huggingface_hub library not found. Please add it to requirements.txt.")
+    except Exception as e:
+        print(f"ℹ️ Could not determine huggingface_hub version: {e}")
 
-    # Commented out the GOOGLE_API_KEY check as it's no longer relevant for this agent
-    # if not os.getenv("GOOGLE_API_KEY"):
-    #     print("⚠️  WARNING: GOOGLE_API_KEY environment variable not found.")
-    #     print("   The Gemini agent will likely fail to initialize. Please set this token in your Space secrets.")
 
     print("-"*(60 + len(" App Starting ")) + "\n")
     print("Launching Gradio Interface for Basic Agent Evaluation...")
-    demo.launch(debug=True, share=False)
+    demo.launch(debug=True, share=False) # debug=True can be helpful for local dev