Upload 3 files

app.py

@@ -1,410 +1,11 @@
-import os
-import gradio as gr
-import requests
-import inspect
-import pandas as pd
-import sys
-from pathlib import Path
+from run import create_agent
 
-# Add your agent's code to the path
-sys.path.append(str(Path(__file__).parent.absolute()))
+from smolagents.gradio_ui import GradioUI
 
-# Import your agent components
-from dotenv import load_dotenv
-from huggingface_hub import login
-from scripts.text_inspector_tool import TextInspectorTool
-from scripts.text_web_browser import (
-    ArchiveSearchTool,
-    FinderTool,
-    FindNextTool,
-    PageDownTool,
-    PageUpTool,
-    SimpleTextBrowser,
-    VisitTool,
-)
-from scripts.visual_qa import visualizer
-from scripts.reformulator import prepare_response
 
-from smolagents import (
-    CodeAgent,
-    GoogleSearchTool,
-    LiteLLMModel,
-    ToolCallingAgent,
-)
+agent = create_agent()
 
-# --- Constants ---
-DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
-
-# GAIA system prompt for exact answer format
-GAIA_SYSTEM_PROMPT = """You are a general AI assistant. I will ask you a question. Report your thoughts, and finish your answer with the following template: FINAL ANSWER: [YOUR FINAL ANSWER]. YOUR FINAL ANSWER should be a number OR as few words as possible OR a comma separated list of numbers and/or strings. If you are asked for a number, don't use comma to write your number neither use units such as $ or percent sign unless specified otherwise. If you are asked for a string, don't use articles, neither abbreviations (e.g. for cities), and write the digits in plain text unless specified otherwise. If you are asked for a comma separated list, apply the above rules depending of whether the element to be put in the list is a number or a string."""
-
-# --- Smolagent Implementation ---
-load_dotenv(override=True)
-
-# Try to login with HF token from env or secrets
-try:
-    hf_token = os.getenv("HF_TOKEN")
-    if hf_token:
-        login(hf_token)
-        print("Successfully logged in to Hugging Face")
-    else:
-        print("No HF_TOKEN found in environment")
-except Exception as e:
-    print(f"Error logging in to Hugging Face: {e}")
-
-# Custom settings for your agent
-custom_role_conversions = {"tool-call": "assistant", "tool-response": "user"}
-user_agent = "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/119.0.0.0 Safari/537.36 Edg/119.0.0.0"
-
-BROWSER_CONFIG = {
-    "viewport_size": 1024 * 5,
-    "downloads_folder": "downloads_folder",
-    "request_kwargs": {
-        "headers": {"User-Agent": user_agent},
-        "timeout": 300,
-    },
-    "serpapi_key": os.getenv("SERPAPI_API_KEY"),
-}
-
-# Create downloads folder if it doesn't exist
-os.makedirs(f"./{BROWSER_CONFIG['downloads_folder']}", exist_ok=True)
-
-class SmolaAgent:
-    def __init__(self):
-        print("Initializing SmolaAgent...")
-        
-        # Initialize model
-        model_id = "o1"  # You can adjust this or make it configurable
-        model_params = {
-            "model_id": model_id,
-            "custom_role_conversions": custom_role_conversions,
-            "max_completion_tokens": 8192,
-        }
-        if model_id == "o1":
-            model_params["reasoning_effort"] = "high"
-        
-        self.model = LiteLLMModel(**model_params)
-        
-        # Create agent with tools
-        text_limit = 100000
-        browser = SimpleTextBrowser(**BROWSER_CONFIG)
-        WEB_TOOLS = [
-            GoogleSearchTool(provider="serper"),
-            VisitTool(browser),
-            PageUpTool(browser),
-            PageDownTool(browser),
-            FinderTool(browser),
-            FindNextTool(browser),
-            ArchiveSearchTool(browser),
-            TextInspectorTool(self.model, text_limit),
-        ]
-        
-        # Create text webbrowser agent
-        self.text_webbrowser_agent = ToolCallingAgent(
-            model=self.model,
-            tools=WEB_TOOLS,
-            max_steps=20,
-            verbosity_level=2,
-            planning_interval=4,
-            name="search_agent",
-            description="""A team member that will search the internet to answer your question.
-        Ask him for all your questions that require browsing the web.
-        Provide him as much context as possible, in particular if you need to search on a specific timeframe!
-        And don't hesitate to provide him with a complex search task, like finding a difference between two webpages.
-        Your request must be a real sentence, not a google search! Like "Find me this information (...)" rather than a few keywords.
-        """,
-            provide_run_summary=True,
-        )
-        
-        self.text_webbrowser_agent.prompt_templates["managed_agent"]["task"] += """You can navigate to .txt online files.
-        If a non-html page is in another format, especially .pdf or a Youtube video, use tool 'inspect_file_as_text' to inspect it.
-        Additionally, if after some searching you find out that you need more information to answer the question, you can use `final_answer` with your request for clarification as argument to request for more information."""
-
-        # Create manager agent
-        self.manager_agent = CodeAgent(
-            model=self.model,
-            tools=[visualizer, TextInspectorTool(self.model, text_limit)],
-            max_steps=12,
-            verbosity_level=2,
-            additional_authorized_imports=["*"],
-            planning_interval=4,
-            managed_agents=[self.text_webbrowser_agent],
-        )
-        
-        print("SmolaAgent initialized successfully.")
-
-    def __call__(self, question: str) -> str:
-        print(f"Agent received question: {question[:50]}...")
-        
-        # Include the GAIA system prompt in the question to ensure proper answer format
-        augmented_question = f"""You have one question to answer. It is paramount that you provide a correct answer.
-Give it all you can: I know for a fact that you have access to all the relevant tools to solve it and find the correct answer (the answer does exist). Failure or 'I cannot answer' or 'None found' will not be tolerated, success will be rewarded.
-Run verification steps if that's needed, you must make sure you find the correct answer!
-
-{GAIA_SYSTEM_PROMPT}
-
-Here is the task:
-{question}"""
-        
-        try:
-            # Run the agent
-            result = self.manager_agent.run(augmented_question)
-            
-            # Use reformulator to get properly formatted final answer
-            agent_memory = self.manager_agent.write_memory_to_messages()
-            
-            # Add the GAIA system prompt to the reformulation to ensure correct format
-            for message in agent_memory:
-                if message.get("role") == "system" and message.get("content"):
-                    if isinstance(message["content"], list):
-                        for content_item in message["content"]:
-                            if content_item.get("type") == "text":
-                                content_item["text"] = GAIA_SYSTEM_PROMPT + "\n\n" + content_item["text"]
-                    else:
-                        message["content"] = GAIA_SYSTEM_PROMPT + "\n\n" + message["content"]
-                    break
-            
-            final_answer = prepare_response(augmented_question, agent_memory, self.model)
-            
-            print(f"Agent returning answer: {final_answer}")
-            return final_answer
-        
-        except Exception as e:
-            print(f"Error running agent: {e}")
-            return "FINAL ANSWER: Unable to determine"
-
-# Function to extract the exact answer from agent response
-def extract_final_answer(agent_response):
-    if "FINAL ANSWER:" in agent_response:
-        answer = agent_response.split("FINAL ANSWER:")[1].strip()
-        
-        # Additional cleaning to ensure exact match
-        # Remove any trailing punctuation
-        answer = answer.rstrip('.,!?;:')
-        
-        # Clean numbers (remove commas and units)
-        # This is a simple example - you might need more sophisticated cleaning
-        words = answer.split()
-        for i, word in enumerate(words):
-            # Try to convert to a number to remove commas and format correctly
-            try:
-                num = float(word.replace(',', '').replace('$', '').replace('%', ''))
-                # Convert to int if it's a whole number
-                words[i] = str(int(num)) if num.is_integer() else str(num)
-            except (ValueError, AttributeError):
-                # Not a number, leave as is
-                pass
-        
-        return ' '.join(words)
-    
-    return "Unable to determine"
-
-# Replace BasicAgent with your SmolaAgent in the run_and_submit_all function
-def run_and_submit_all(profile: gr.OAuthProfile | None):
-    """
-    Fetches all questions, runs the SmolaAgent on them, submits all answers,
-    and displays the results.
-    """
-    # --- Determine HF Space Runtime URL and Repo URL ---
-    space_id = os.getenv("SPACE_ID") # Get the SPACE_ID for sending link to the code
-
-    if profile:
-        username= f"{profile.username}"
-        print(f"User logged in: {username}")
-    else:
-        print("User not logged in.")
-        return "Please Login to Hugging Face with the button.", None
-
-    api_url = DEFAULT_API_URL
-    questions_url = f"{api_url}/questions"
-    submit_url = f"{api_url}/submit"
-
-    # 1. Instantiate Agent
-    try:
-        agent = SmolaAgent()
-    except Exception as e:
-        print(f"Error instantiating agent: {e}")
-        return f"Error initializing agent: {e}", None
-    
-    # In the case of an app running as a hugging Face space, this link points toward your codebase
-    agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
-    print(agent_code)
-
-    # 2. Fetch Questions
-    print(f"Fetching questions from: {questions_url}")
-    try:
-        response = requests.get(questions_url, timeout=15)
-        response.raise_for_status()
-        questions_data = response.json()
-        if not questions_data:
-             print("Fetched questions list is empty.")
-             return "Fetched questions list is empty or invalid format.", None
-        print(f"Fetched {len(questions_data)} questions.")
-    except requests.exceptions.RequestException as e:
-        print(f"Error fetching questions: {e}")
-        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]}")
-         return f"Error decoding server response for questions: {e}", None
-    except Exception as e:
-        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 = []
-    print(f"Running agent on {len(questions_data)} questions...")
-    for item in questions_data:
-        task_id = item.get("task_id")
-        question_text = item.get("question")
-        
-        # Check if there are files associated with this task
-        try:
-            files_url = f"{api_url}/files/{task_id}"
-            files_response = requests.get(files_url, timeout=15)
-            if files_response.status_code == 200:
-                # Save the file and provide its path to the agent
-                # This depends on what format the files are returned in
-                print(f"Task {task_id} has associated files")
-                # Handle files if needed
-        except Exception as e:
-            print(f"Error checking for files for task {task_id}: {e}")
-            # Continue even if file check fails
-        
-        if not task_id or question_text is None:
-            print(f"Skipping item with missing task_id or question: {item}")
-            continue
-        
-        try:
-            # Get full agent response
-            full_response = agent(question_text)
-            
-            # Extract just the final answer part for submission
-            submitted_answer = extract_final_answer(full_response)
-                
-            # Add to submission payload
-            answers_payload.append({
-                "task_id": task_id, 
-                "submitted_answer": submitted_answer,
-                "reasoning_trace": full_response  # Optional: include full reasoning
-            })
-            
-            # Log for display
-            results_log.append({
-                "Task ID": task_id, 
-                "Question": question_text, 
-                "Submitted Answer": submitted_answer,
-                "Full Response": full_response
-            })
-            
-            print(f"Processed task {task_id}, answer: {submitted_answer}")
-            
-        except Exception as e:
-             print(f"Error running agent on task {task_id}: {e}")
-             results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": f"AGENT ERROR: {e}"})
-
-    if not answers_payload:
-        print("Agent did not produce any answers to submit.")
-        return "Agent did not produce any answers to submit.", pd.DataFrame(results_log)
-
-    # 4. Prepare Submission 
-    submission_data = {"username": username.strip(), "agent_code": agent_code, "answers": answers_payload}
-    status_update = f"Agent finished. Submitting {len(answers_payload)} answers for user '{username}'..."
-    print(status_update)
-
-    # 5. Submit
-    print(f"Submitting {len(answers_payload)} answers to: {submit_url}")
-    try:
-        response = requests.post(submit_url, json=submission_data, timeout=60)
-        response.raise_for_status()
-        result_data = response.json()
-        final_status = (
-            f"Submission Successful!\n"
-            f"User: {result_data.get('username')}\n"
-            f"Overall Score: {result_data.get('score', 'N/A')}% "
-            f"({result_data.get('correct_count', '?')}/{result_data.get('total_attempted', '?')} correct)\n"
-            f"Message: {result_data.get('message', 'No message received.')}"
-        )
-        print("Submission successful.")
-        results_df = pd.DataFrame(results_log)
-        return final_status, results_df
-    except requests.exceptions.HTTPError as e:
-        error_detail = f"Server responded with status {e.response.status_code}."
-        try:
-            error_json = e.response.json()
-            error_detail += f" Detail: {error_json.get('detail', e.response.text)}"
-        except requests.exceptions.JSONDecodeError:
-            error_detail += f" Response: {e.response.text[:500]}"
-        status_message = f"Submission Failed: {error_detail}"
-        print(status_message)
-        results_df = pd.DataFrame(results_log)
-        return status_message, results_df
-    except requests.exceptions.Timeout:
-        status_message = "Submission Failed: The request timed out."
-        print(status_message)
-        results_df = pd.DataFrame(results_log)
-        return status_message, results_df
-    except requests.exceptions.RequestException as e:
-        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:
-        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("# Smolagent GAIA Evaluation Runner")
-    gr.Markdown(
-        """
-        **Instructions:**
-        1. Log in to your Hugging Face account using the button below.
-        2. Click 'Run Evaluation & Submit All Answers' to fetch questions, run your agent, submit answers, and see the score.
-        ---
-        **Note:** This process will take some time as the agent processes each question. The agent is specifically configured to 
-        format answers according to the GAIA benchmark requirements:
-        - Numbers: No commas, no units
-        - Strings: No articles, no abbreviations
-        - Lists: Comma-separated values following the above rules
-        """
-    )
-
-    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]
-    )
+demo = GradioUI(agent)
 
 if __name__ == "__main__":
-    print("\n" + "-"*30 + " App Starting " + "-"*30)
-    # Check for SPACE_HOST and SPACE_ID at startup for information
-    space_host_startup = os.getenv("SPACE_HOST")
-    space_id_startup = os.getenv("SPACE_ID") # Get SPACE_ID at startup
-
-    if space_host_startup:
-        print(f"✅ SPACE_HOST found: {space_host_startup}")
-        print(f"   Runtime URL should be: https://{space_host_startup}.hf.space")
-    else:
-        print("ℹ️  SPACE_HOST environment variable not found (running locally?).")
-
-    if space_id_startup: # Print repo URLs if SPACE_ID is found
-        print(f"✅ SPACE_ID found: {space_id_startup}")
-        print(f"   Repo URL: https://huggingface.co/spaces/{space_id_startup}")
-        print(f"   Repo Tree URL: https://huggingface.co/spaces/{space_id_startup}/tree/main")
-    else:
-        print("ℹ️  SPACE_ID environment variable not found (running locally?). Repo URL cannot be determined.")
-
-    print("-"*(60 + len(" App Starting ")) + "\n")
-
-    print("Launching Gradio Interface for Smolagent GAIA Evaluation...")
-    demo.launch(debug=True, share=False)
+    demo.launch()

run.py

@@ -0,0 +1,125 @@
+import argparse
+import os
+import threading
+
+from dotenv import load_dotenv
+from huggingface_hub import login
+from scripts.text_inspector_tool import TextInspectorTool
+from scripts.text_web_browser import (
+    ArchiveSearchTool,
+    FinderTool,
+    FindNextTool,
+    PageDownTool,
+    PageUpTool,
+    SimpleTextBrowser,
+    VisitTool,
+)
+from scripts.visual_qa import visualizer
+
+from smolagents import (
+    CodeAgent,
+    GoogleSearchTool,
+    # InferenceClientModel,
+    LiteLLMModel,
+    ToolCallingAgent,
+)
+
+
+load_dotenv(override=True)
+login(os.getenv("HF_TOKEN"))
+
+append_answer_lock = threading.Lock()
+
+
+def parse_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "question", type=str, help="for example: 'How many studio albums did Mercedes Sosa release before 2007?'"
+    )
+    parser.add_argument("--model-id", type=str, default="o1")
+    return parser.parse_args()
+
+
+custom_role_conversions = {"tool-call": "assistant", "tool-response": "user"}
+
+user_agent = "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/119.0.0.0 Safari/537.36 Edg/119.0.0.0"
+
+BROWSER_CONFIG = {
+    "viewport_size": 1024 * 5,
+    "downloads_folder": "downloads_folder",
+    "request_kwargs": {
+        "headers": {"User-Agent": user_agent},
+        "timeout": 300,
+    },
+    "serpapi_key": os.getenv("SERPAPI_API_KEY"),
+}
+
+os.makedirs(f"./{BROWSER_CONFIG['downloads_folder']}", exist_ok=True)
+
+
+def create_agent(model_id="o1"):
+    model_params = {
+        "model_id": model_id,
+        "custom_role_conversions": custom_role_conversions,
+        "max_completion_tokens": 8192,
+    }
+    if model_id == "o1":
+        model_params["reasoning_effort"] = "high"
+    model = LiteLLMModel(**model_params)
+
+    text_limit = 100000
+    browser = SimpleTextBrowser(**BROWSER_CONFIG)
+    WEB_TOOLS = [
+        GoogleSearchTool(provider="serper"),
+        VisitTool(browser),
+        PageUpTool(browser),
+        PageDownTool(browser),
+        FinderTool(browser),
+        FindNextTool(browser),
+        ArchiveSearchTool(browser),
+        TextInspectorTool(model, text_limit),
+    ]
+    text_webbrowser_agent = ToolCallingAgent(
+        model=model,
+        tools=WEB_TOOLS,
+        max_steps=20,
+        verbosity_level=2,
+        planning_interval=4,
+        name="search_agent",
+        description="""A team member that will search the internet to answer your question.
+    Ask him for all your questions that require browsing the web.
+    Provide him as much context as possible, in particular if you need to search on a specific timeframe!
+    And don't hesitate to provide him with a complex search task, like finding a difference between two webpages.
+    Your request must be a real sentence, not a google search! Like "Find me this information (...)" rather than a few keywords.
+    """,
+        provide_run_summary=True,
+    )
+    text_webbrowser_agent.prompt_templates["managed_agent"]["task"] += """You can navigate to .txt online files.
+    If a non-html page is in another format, especially .pdf or a Youtube video, use tool 'inspect_file_as_text' to inspect it.
+    Additionally, if after some searching you find out that you need more information to answer the question, you can use `final_answer` with your request for clarification as argument to request for more information."""
+
+    manager_agent = CodeAgent(
+        model=model,
+        tools=[visualizer, TextInspectorTool(model, text_limit)],
+        max_steps=12,
+        verbosity_level=2,
+        additional_authorized_imports=["*"],
+        planning_interval=4,
+        managed_agents=[text_webbrowser_agent],
+    )
+
+    return manager_agent
+
+
+def main():
+    args = parse_args()
+
+    agent = create_agent(model_id=args.model_id)
+
+    answer = agent.run(args.question)
+
+    print(f"Got this answer: {answer}")
+
+
+if __name__ == "__main__":
+    main()

run_gaia.py

@@ -0,0 +1,277 @@
+# EXAMPLE COMMAND: python examples/open_deep_research/run_gaia.py --concurrency 32 --run-name generate-traces-03-apr-noplanning --model-id gpt-4o
+import argparse
+import json
+import os
+import threading
+from concurrent.futures import ThreadPoolExecutor, as_completed
+from datetime import datetime
+from pathlib import Path
+
+import datasets
+import pandas as pd
+from dotenv import load_dotenv
+from huggingface_hub import login
+from scripts.reformulator import prepare_response
+from scripts.run_agents import (
+    get_single_file_description,
+    get_zip_description,
+)
+from scripts.text_inspector_tool import TextInspectorTool
+from scripts.text_web_browser import (
+    ArchiveSearchTool,
+    FinderTool,
+    FindNextTool,
+    PageDownTool,
+    PageUpTool,
+    SimpleTextBrowser,
+    VisitTool,
+)
+from scripts.visual_qa import visualizer
+from tqdm import tqdm
+
+from smolagents import (
+    CodeAgent,
+    GoogleSearchTool,
+    LiteLLMModel,
+    Model,
+    ToolCallingAgent,
+)
+
+
+load_dotenv(override=True)
+login(os.getenv("HF_TOKEN"))
+
+append_answer_lock = threading.Lock()
+
+
+def parse_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--concurrency", type=int, default=8)
+    parser.add_argument("--model-id", type=str, default="o1")
+    parser.add_argument("--run-name", type=str, required=True)
+    return parser.parse_args()
+
+
+### IMPORTANT: EVALUATION SWITCHES
+
+print("Make sure you deactivated Tailscale VPN, else some URLs will be blocked!")
+
+USE_OPEN_MODELS = False
+
+SET = "validation"
+
+custom_role_conversions = {"tool-call": "assistant", "tool-response": "user"}
+
+### LOAD EVALUATION DATASET
+
+eval_ds = datasets.load_dataset("gaia-benchmark/GAIA", "2023_all")[SET]
+eval_ds = eval_ds.rename_columns({"Question": "question", "Final answer": "true_answer", "Level": "task"})
+
+
+def preprocess_file_paths(row):
+    if len(row["file_name"]) > 0:
+        row["file_name"] = f"data/gaia/{SET}/" + row["file_name"]
+    return row
+
+
+eval_ds = eval_ds.map(preprocess_file_paths)
+eval_df = pd.DataFrame(eval_ds)
+print("Loaded evaluation dataset:")
+print(eval_df["task"].value_counts())
+
+user_agent = "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/119.0.0.0 Safari/537.36 Edg/119.0.0.0"
+
+BROWSER_CONFIG = {
+    "viewport_size": 1024 * 5,
+    "downloads_folder": "downloads_folder",
+    "request_kwargs": {
+        "headers": {"User-Agent": user_agent},
+        "timeout": 300,
+    },
+    "serpapi_key": os.getenv("SERPAPI_API_KEY"),
+}
+
+os.makedirs(f"./{BROWSER_CONFIG['downloads_folder']}", exist_ok=True)
+
+
+def create_agent_team(model: Model):
+    text_limit = 100000
+    ti_tool = TextInspectorTool(model, text_limit)
+
+    browser = SimpleTextBrowser(**BROWSER_CONFIG)
+
+    WEB_TOOLS = [
+        GoogleSearchTool(provider="serper"),
+        VisitTool(browser),
+        PageUpTool(browser),
+        PageDownTool(browser),
+        FinderTool(browser),
+        FindNextTool(browser),
+        ArchiveSearchTool(browser),
+        TextInspectorTool(model, text_limit),
+    ]
+
+    text_webbrowser_agent = ToolCallingAgent(
+        model=model,
+        tools=WEB_TOOLS,
+        max_steps=20,
+        verbosity_level=2,
+        planning_interval=4,
+        name="search_agent",
+        description="""A team member that will search the internet to answer your question.
+    Ask him for all your questions that require browsing the web.
+    Provide him as much context as possible, in particular if you need to search on a specific timeframe!
+    And don't hesitate to provide him with a complex search task, like finding a difference between two webpages.
+    Your request must be a real sentence, not a google search! Like "Find me this information (...)" rather than a few keywords.
+    """,
+        provide_run_summary=True,
+    )
+    text_webbrowser_agent.prompt_templates["managed_agent"]["task"] += """You can navigate to .txt online files.
+    If a non-html page is in another format, especially .pdf or a Youtube video, use tool 'inspect_file_as_text' to inspect it.
+    Additionally, if after some searching you find out that you need more information to answer the question, you can use `final_answer` with your request for clarification as argument to request for more information."""
+
+    manager_agent = CodeAgent(
+        model=model,
+        tools=[visualizer, ti_tool],
+        max_steps=12,
+        verbosity_level=2,
+        additional_authorized_imports=["*"],
+        planning_interval=4,
+        managed_agents=[text_webbrowser_agent],
+    )
+    return manager_agent
+
+
+def append_answer(entry: dict, jsonl_file: str) -> None:
+    jsonl_file = Path(jsonl_file)
+    jsonl_file.parent.mkdir(parents=True, exist_ok=True)
+    with append_answer_lock, open(jsonl_file, "a", encoding="utf-8") as fp:
+        fp.write(json.dumps(entry) + "\n")
+    assert os.path.exists(jsonl_file), "File not found!"
+    print("Answer exported to file:", jsonl_file.resolve())
+
+
+def answer_single_question(example, model_id, answers_file, visual_inspection_tool):
+    model_params = {
+        "model_id": model_id,
+        "custom_role_conversions": custom_role_conversions,
+    }
+    if model_id == "o1":
+        model_params["reasoning_effort"] = "high"
+        model_params["max_completion_tokens"] = 8192
+    else:
+        model_params["max_tokens"] = 4096
+    model = LiteLLMModel(**model_params)
+    # model = InferenceClientModel(model_id="Qwen/Qwen2.5-Coder-32B-Instruct", provider="together", max_tokens=4096)
+    document_inspection_tool = TextInspectorTool(model, 100000)
+
+    agent = create_agent_team(model)
+
+    augmented_question = """You have one question to answer. It is paramount that you provide a correct answer.
+Give it all you can: I know for a fact that you have access to all the relevant tools to solve it and find the correct answer (the answer does exist). Failure or 'I cannot answer' or 'None found' will not be tolerated, success will be rewarded.
+Run verification steps if that's needed, you must make sure you find the correct answer!
+Here is the task:
+""" + example["question"]
+
+    if example["file_name"]:
+        if ".zip" in example["file_name"]:
+            prompt_use_files = "\n\nTo solve the task above, you will have to use these attached files:\n"
+            prompt_use_files += get_zip_description(
+                example["file_name"], example["question"], visual_inspection_tool, document_inspection_tool
+            )
+        else:
+            prompt_use_files = "\n\nTo solve the task above, you will have to use this attached file:"
+            prompt_use_files += get_single_file_description(
+                example["file_name"], example["question"], visual_inspection_tool, document_inspection_tool
+            )
+        augmented_question += prompt_use_files
+
+    start_time = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
+    try:
+        # Run agent 🚀
+        final_result = agent.run(augmented_question)
+
+        agent_memory = agent.write_memory_to_messages()
+
+        final_result = prepare_response(augmented_question, agent_memory, reformulation_model=model)
+
+        output = str(final_result)
+        for memory_step in agent.memory.steps:
+            memory_step.model_input_messages = None
+        intermediate_steps = agent_memory
+
+        # Check for parsing errors which indicate the LLM failed to follow the required format
+        parsing_error = True if any(["AgentParsingError" in step for step in intermediate_steps]) else False
+
+        # check if iteration limit exceeded
+        iteration_limit_exceeded = True if "Agent stopped due to iteration limit or time limit." in output else False
+        raised_exception = False
+
+    except Exception as e:
+        print("Error on ", augmented_question, e)
+        output = None
+        intermediate_steps = []
+        parsing_error = False
+        iteration_limit_exceeded = False
+        exception = e
+        raised_exception = True
+    end_time = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
+    token_counts_manager = agent.monitor.get_total_token_counts()
+    token_counts_web = list(agent.managed_agents.values())[0].monitor.get_total_token_counts()
+    total_token_counts = {
+        "input": token_counts_manager["input"] + token_counts_web["input"],
+        "output": token_counts_manager["output"] + token_counts_web["output"],
+    }
+    annotated_example = {
+        "agent_name": model.model_id,
+        "question": example["question"],
+        "augmented_question": augmented_question,
+        "prediction": output,
+        "intermediate_steps": intermediate_steps,
+        "parsing_error": parsing_error,
+        "iteration_limit_exceeded": iteration_limit_exceeded,
+        "agent_error": str(exception) if raised_exception else None,
+        "task": example["task"],
+        "task_id": example["task_id"],
+        "true_answer": example["true_answer"],
+        "start_time": start_time,
+        "end_time": end_time,
+        "token_counts": total_token_counts,
+    }
+    append_answer(annotated_example, answers_file)
+
+
+def get_examples_to_answer(answers_file, eval_ds) -> list[dict]:
+    print(f"Loading answers from {answers_file}...")
+    try:
+        done_questions = pd.read_json(answers_file, lines=True)["question"].tolist()
+        print(f"Found {len(done_questions)} previous results!")
+    except Exception as e:
+        print("Error when loading records: ", e)
+        print("No usable records! ▶️ Starting new.")
+        done_questions = []
+    return [line for line in eval_ds.to_list() if line["question"] not in done_questions]
+
+
+def main():
+    args = parse_args()
+    print(f"Starting run with arguments: {args}")
+
+    answers_file = f"output/{SET}/{args.run_name}.jsonl"
+    tasks_to_run = get_examples_to_answer(answers_file, eval_ds)
+
+    with ThreadPoolExecutor(max_workers=args.concurrency) as exe:
+        futures = [
+            exe.submit(answer_single_question, example, args.model_id, answers_file, visualizer)
+            for example in tasks_to_run
+        ]
+        for f in tqdm(as_completed(futures), total=len(tasks_to_run), desc="Processing tasks"):
+            f.result()
+
+    # for example in tasks_to_run:
+    #     answer_single_question(example, args.model_id, answers_file, visualizer)
+    print("All tasks processed.")
+
+
+if __name__ == "__main__":
+    main()