import fixes

app.py

@@ -10,8 +10,8 @@ import numexpr
 from typing import TypedDict, Annotated
 
 # --- Langchain & HF Imports ---
-# CHANGED: Swapped local pipeline for Inference API and removed torch
-from langchain_huggingface import HuggingFaceInferenceAPI
+# VERIFIED AND CORRECT FINAL IMPORT
+from langchain_community.llms import HuggingFaceHub
 from langchain_community.tools import DuckDuckGoSearchRun
 from langchain_core.prompts import PromptTemplate
 from langchain_core.output_parsers import StrOutputParser
@@ -22,7 +22,6 @@ from langchain_community.document_loaders.youtube import YoutubeLoader
 
 # --- Constants ---
 DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
-# ADDED: A more robust prompt tailored for tool use with Llama 3
 SYSTEM_PROMPT = """You are a helpful and expert assistant named GAIA, designed to answer questions accurately.
 
 To do this, you have access to a set of tools. Based on the user's question, you must decide which tool to use, if any.
@@ -67,7 +66,6 @@ def math_calculator(expression: str) -> str:
     """Calculates the result of a mathematical expression. Use it for any math operation."""
     logging.info(f"--- Calling Math Calculator Tool with expression: {expression} ---")
     try:
-        # Sanitize expression: allow only numbers, basic operators, and parentheses
         if not re.match(r"^[0-9\.\+\-\*\/\(\)\s]+$", expression):
             return "Error: Invalid characters in expression."
         result = numexpr.evaluate(expression).item()
@@ -87,8 +85,7 @@ def image_analyzer(image_url: str) -> str:
             logging.info(
                 "--- Initializing Image Analyzer pipeline (lazy loading)... ---"
             )
-            # This pipeline is small enough to be loaded on demand in a ZeroGPU space
-            from transformers import pipeline
+            from transformers.pipelines import pipeline
 
             image_to_text_pipeline = pipeline(
                 "image-to-text", model="Salesforce/blip-image-captioning-base"
@@ -122,7 +119,6 @@ def youtube_transcript_reader(youtube_url: str) -> str:
         loader = YoutubeLoader.from_youtube_url(youtube_url, add_video_info=False)
         docs = loader.load()
         transcript = " ".join([doc.page_content for doc in docs])
-        # Return a manageable chunk
         return transcript[:4000]
     except Exception as e:
         logging.error(f"Error reading YouTube transcript: {e}")
@@ -147,19 +143,17 @@ class GaiaAgent:
             youtube_transcript_reader,
         ]
 
-        # CHANGED: Replaced local HuggingFacePipeline with HuggingFaceInferenceAPI
-        # This uses the Hugging Face Serverless API, offloading the memory and compute.
-        # It requires a HUGGING_FACE_HUB_TOKEN to be set in the Space secrets.
-        logging.info("Initializing LLM via Inference API...")
-        llm = HuggingFaceInferenceAPI(
-            model_id="meta-llama/Meta-Llama-3-8B-Instruct",
-            # repo_id="meta-llama/Meta-Llama-3-8B-Instruct", # Use repo_id if model_id gives issues
-            task="text-generation",
-            token=os.getenv("HUGGING_FACE_HUB_TOKEN"),
+        # --- THIS SECTION IS NOW CORRECT ---
+        logging.info("Initializing LLM via HuggingFaceHub...")
+        # We use HuggingFaceHub which is the correct class for this job.
+        # Note the parameter names: repo_id, model_kwargs, and huggingfacehub_api_token.
+        llm = HuggingFaceHub(
+            repo_id="meta-llama/Meta-Llama-3-8B-Instruct",
+            model_kwargs={"temperature": 0.1, "max_new_tokens": 1024},
+            huggingfacehub_api_token=os.getenv("HUGGING_FACE_HUB_TOKEN"),
         )
         logging.info("LLM initialized successfully.")
 
-        # Create the agent graph
         prompt = PromptTemplate(
             template=SYSTEM_PROMPT
             + "\nHere is the current conversation:\n{messages}\n\nQuestion: {question}",
@@ -214,7 +208,6 @@ class GaiaAgent:
         tool_name = tool_call_match.group(1).strip()
         tool_input_str = tool_call_match.group(2).strip()
 
-        # Remove quotes from the input string if they exist
         if (tool_input_str.startswith('"') and tool_input_str.endswith('"')) or (
             tool_input_str.startswith("'") and tool_input_str.endswith("'")
         ):
@@ -247,7 +240,6 @@ class GaiaAgent:
         logging.info(f"Agent received question (first 100 chars): {question[:100]}...")
         try:
             initial_state = {"question": question, "messages": [], "sender": "user"}
-            # Increased recursion limit for potentially complex questions
             final_state = self.graph.invoke(initial_state, {"recursion_limit": 15})
             final_response = final_state["messages"][-1]
 
@@ -262,28 +254,21 @@ class GaiaAgent:
                 logging.warning(
                     "Agent could not find a final answer. Returning the last message."
                 )
-                # Fallback: return the last piece of the conversation if parsing fails
                 return final_response
         except Exception as e:
             logging.error(f"Error during agent invocation: {e}", exc_info=True)
             return f"Error during agent invocation: {e}"
 
 
-# --- Gradio App Logic (largely unchanged, but with enhanced logging) ---
+# --- Gradio App Logic (Unchanged) ---
 
 
 def run_and_submit_all(profile: gr.OAuthProfile | None):
-    """
-    Fetches all questions, runs the GaiaAgent on them, submits all answers,
-    and displays the results.
-    """
     if not profile:
         logging.warning("User not logged in.")
         return "Please Login to Hugging Face with the button.", None
-
     username = profile.username
     logging.info(f"User logged in: {username}")
-
     space_id = os.getenv("SPACE_ID")
     if not space_id:
         logging.error("SPACE_ID environment variable is not set. Cannot proceed.")
@@ -291,22 +276,16 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
             "CRITICAL ERROR: SPACE_ID environment variable is not set. Cannot generate submission.",
             None,
         )
-
     api_url = DEFAULT_API_URL
     questions_url = f"{api_url}/questions"
     submit_url = f"{api_url}/submit"
-
-    # 1. Instantiate Agent
     try:
         agent = GaiaAgent()
     except Exception as e:
         logging.critical(f"Fatal error instantiating agent: {e}", exc_info=True)
         return f"Fatal error initializing agent: {e}", None
-
     agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
     logging.info(f"Agent code URL: {agent_code}")
-
-    # 2. Fetch Questions
     logging.info(f"Fetching questions from: {questions_url}")
     try:
         response = requests.get(questions_url, timeout=20)
@@ -319,8 +298,6 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
     except Exception as e:
         logging.error(f"Error fetching questions: {e}")
         return f"Error fetching questions: {e}", None
-
-    # 3. Run your Agent
     results_log = []
     answers_payload = []
     logging.info(f"Running agent on {len(questions_data)} questions...")
@@ -332,7 +309,6 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
         )
         if not task_id or question_text is None:
             continue
-
         try:
             submitted_answer = agent(question_text)
             answers_payload.append(
@@ -354,12 +330,9 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
                     "Submitted Answer": f"AGENT ERROR: {e}",
                 }
             )
-
     if not answers_payload:
         logging.warning("Agent did not produce any answers.")
         return "Agent did not produce any answers.", pd.DataFrame(results_log)
-
-    # 4. Prepare and Submit
     submission_data = {
         "username": username.strip(),
         "agent_code": agent_code,
@@ -392,7 +365,7 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
         )
 
 
-# --- Build Gradio Interface (UI text is maintained as requested) ---
+# --- Build Gradio Interface (Unchanged) ---
 with gr.Blocks() as demo:
     gr.Markdown("# GAIA Agent Evaluation Runner")
     gr.Markdown(
@@ -408,15 +381,12 @@ with gr.Blocks() as demo:
         Once you click the submit button, please be patient. The agent needs time to process all the questions, which can take several minutes.
         """
     )
-
     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)
-
     run_button.click(
         fn=run_and_submit_all,
         outputs=[status_output, results_table],