AgentsGAIAFun / main.py
Vlad Iliescu
feat: final update
489ed9a
import concurrent
import os
import time
from concurrent import futures
from pathlib import Path
from typing import Any, List, Dict, Tuple
import pandas as pd
import requests
from dotenv import dotenv_values, load_dotenv
from openai import AzureOpenAI, RateLimitError
from smolagents import tool
from tqdm.auto import tqdm
from smolagents import GoogleSearchTool
import requests
import urllib.request
from markdownify import markdownify as md
from bs4 import BeautifulSoup
import json
test_api_base = "https://agents-course-unit4-scoring.hf.space"
# Configuration
load_dotenv()
client = AzureOpenAI(
api_key=os.getenv("AZURE_OPENAI_API_KEY"),
azure_endpoint=os.getenv("AZURE_OPENAI_API_BASE"),
api_version=os.getenv("AZURE_OPENAI_API_VERSION")
)
openai_chatmodel = os.getenv("AZURE_OPENAI_CHAT_MODEL")
GRAY = "\033[90m"
BOLD = "\033[1m"
RESET = "\033[0m"
# Load questions
response = requests.get(f"{test_api_base}/questions", timeout=15)
response.raise_for_status()
questions_data = response.json()
df = pd.DataFrame(questions_data)
# Define tools & agent
@tool
def read_file(file_path_str: str) -> str:
"""
A tool that reads the contents of a file and returns them as text.
Args:
file_path_str: The path to the file that should be read.
"""
file_path = Path(file_path_str)
file_path = file_path.resolve()
if not file_path.exists() or not file_path.is_file():
raise ValueError(f"File {file_path} does not exist or is not a file.")
switcher = {
".txt": lambda: file_path.read_text(encoding="utf-8"),
".csv": lambda: file_path.read_text(encoding="utf-8"),
".py": lambda: file_path.read_text(encoding="utf-8"),
".xlsx": lambda: pd.read_excel(file_path).to_string(),
}
return switcher.get(file_path.suffix, lambda: "Unsupported file type")()
def get_search_results_for(query):
encoded_query = urllib.parse.urlencode({'q': query})
url = f'https://html.duckduckgo.com/html?q={encoded_query}'
request = urllib.request.Request(url)
request.add_header('User-Agent', 'Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_6) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/87.0.4280.88 Safari/537.36')
raw_response = urllib.request.urlopen(request).read()
html = raw_response.decode("utf-8")
soup = BeautifulSoup(html, 'html.parser')
a_results = soup.select("a.result__a")
links = []
for a_result in a_results:
# print(a_result)
url = a_result.attrs['href']
title = a_result.text
links.append({"title": title, "url": url} )
return links
search_tool = GoogleSearchTool("serper")
def get_google_search_results_for(query: str):
return search_tool.forward(query)
def load_page_content(url) -> str:
response = requests.get(url, headers={'User-Agent': 'Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_6) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/87.0.4280.88 Safari/537.36'})
page_content = response.content.decode('utf-8')
page_content_md = md(page_content)
return page_content_md
tools = [{
"type": "function",
"function": {
"name": "get_search_results_for",
"description": "Returns the top 10 results for a DuckDuckGo query.",
"parameters": {
"type": "object",
"properties": {
"query": {
"type": "string",
"description": "query to search for on DuckDuckGo"
}
},
"required": [
"query"
],
"additionalProperties": False
},
"strict": True
}
},
{
"type": "function",
"function": {
"name": "load_page_content",
"description": "Returns the content of a particular webpage.",
"parameters": {
"type": "object",
"properties": {
"url": {
"type": "string",
"description": "Url of the webpage for which to retrieve the content"
}
},
"required": [
"url"
],
"additionalProperties": False
},
"strict": True
}
}
]
def call_function(name, args):
if name == "get_search_results_for":
return get_google_search_results_for(**args)
if name == "load_page_content":
return load_page_content(**args)
return None
def run_agent(task: str):
messages = [
{
"role": "system",
"content": "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."
},
{"role": "user", "content": task}
]
while True:
for i in range(10):
try:
completion = client.chat.completions.create(
model=openai_chatmodel,
messages=messages,
tools=tools
)
break
except RateLimitError:
print(f"{GRAY}Rate limit exceeded, waiting for 10 seconds...{RESET}")
time.sleep(i*10)
continue
if completion.choices[0].finish_reason == "stop":
print(f"{BOLD}Final answer: {completion.choices[0].message.content}{RESET}")
return completion.choices[0].message.content.split("FINAL ANSWER:")[-1].strip()
elif completion.choices[0].finish_reason == "tool_calls":
messages.append(completion.choices[0].message)
for tool_call in completion.choices[0].message.tool_calls:
name = tool_call.function.name
args = json.loads(tool_call.function.arguments)
try:
result = call_function(name, args)
except Exception as e:
result = "Error calling function: " + str(e)
print(f"Called {BOLD}{name}({args}){RESET} and it returned {GRAY}{str(result)[:300]}{RESET}")
messages.append({
"role": "tool",
"tool_call_id": tool_call.id,
"content": str(result)
})
else:
raise Exception("We're not supposed to be here")
def process_question(question_data: dict[str, Any]) -> dict[str, str]:
task_id = question_data.get("task_id")
question_text = question_data.get("question")
# file_path = None
# if question_data.get("file_name"):
# task_id = question_data["task_id"]
# file_url = f"{test_api_base}/files/{task_id}"
#
# download_dir = Path("downloaded_files")
# download_dir.mkdir(exist_ok=True)
#
# file_response = requests.get(file_url, timeout=30)
# file_response.raise_for_status()
#
# file_path = download_dir / question_data.get("file_name")
#
# with open(file_path, 'wb') as f:
# f.write(file_response.content)
answer = run_agent(question_text)
# if file_path and file_path.suffix in ['.png', '.jpg', '.jpeg']: # I know, it's inconsistent
# answer = agent.run(task=adjusted_question_text, images=[Image.open(file_path)])
# else:
# answer = agent.run(task=f"{adjusted_question_text}{f' File: |{file_path}|' if question_data.get('file_name') else ''}", )
# print(f"Task ID: {task_id}, Question: {question_text}, Answer: {answer}")
return {
"task_id": task_id,
"submitted_answer": answer,
"question": question_text
}
def run_agents_parallel(questions_data: List[Dict[str, Any]], max_workers: int = 4) -> Tuple[Dict[str, Any], List[Dict[str, Any]]]:
start = time.time()
answers = []
results_log = []
with concurrent.futures.ThreadPoolExecutor(max_workers=max_workers) as executor:
future_to_question = {executor.submit(process_question, q): q for q in questions_data}
for future in tqdm(concurrent.futures.as_completed(future_to_question)):
try:
answer = future.result()
results_log.append(answer)
answers.append(answer)
except Exception as e:
print(f"Question processing failed: {e}")
submission_data = {
"username": "vladi",
"agent_code": "https://huggingface.co/spaces/vladi/AgentsGAIAFun",
"answers": answers
}
end = time.time()
print(f"Processing time (parallel): {end - start:.2f} seconds")
return submission_data, results_log
def run_agents(questions_data: list[{}]):
start = time.time()
answers = []
results_log = []
for question_data in tqdm(questions_data):
answer = process_question(question_data)
results_log.append(answer)
answers.append(answer)
submission_data = {
"username": "vladi",
"agent_code": "https://huggingface.co/spaces/vladi/AgentsGAIAFun",
"answers": answers
}
end = time.time()
print(f"Processing time (sequential): {end - start:.2f} seconds")
return submission_data, results_log
def submit_answers(submission_data: dict):
print(f"Submitting {len(submission_data['answers'])} answers")
response = requests.post(f"{test_api_base}/submit", json=submission_data, timeout=60)
response.raise_for_status()
result_data = response.json()
return result_data
submission_data, results_log = run_agents(questions_data)#[:20])
# submission_data, results_log = run_agents_parallel(questions_data)
results_df = pd.DataFrame(results_log)
# Last but not least...
submit_answers(submission_data)