SmolAgentsCourse / agents /react_agent.py
José Ángel González
moved to langgraph
25d44c1
from langchain_community.utilities import GoogleSerperAPIWrapper
from smolagents import PythonInterpreterTool
from langgraph.graph import MessagesState
from langchain_openai import ChatOpenAI
from langgraph.graph import START, StateGraph
from langgraph.prebuilt import tools_condition, ToolNode
from langchain_core.messages import SystemMessage
from openai import OpenAI
from smolagents import Tool
from typing import Optional
import tempfile
import os
from urllib.parse import urlparse
from base64 import b64encode
import requests
from bs4 import BeautifulSoup
import re
import wikipediaapi
# Configs
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
FILE_URL = f"{DEFAULT_API_URL}/files/{{task_id}}"
# Tools
def search_tool(query: str) -> str:
"""Search in Google and returns an string with title, link, and snippet for the top 10 results.
Args:
query: str
Returns:
Title, link, and snippet for the top 10 results
"""
searcher = GoogleSerperAPIWrapper(k=10)
retries = 3
result = ""
while retries > 0:
try:
search_results = searcher.results(query)["organic"]
for row in search_results:
result += f"Title: {row['title']}\nSnippet: {row['snippet']}\nURL: {row['link']}\n\n"
return result
except Exception as e:
retries -= 1
return f"There was an error with Google search: {e}"
def save_file(content: str, filename: Optional[str]) -> str:
"""
Save content to a temporary file and return the path.
Useful for processing files from the GAIA API.
Args:
content: The content to save to the file
filename: Optional filename, will generate a random name if not provided
Returns:
Path to the saved file
"""
temp_dir = tempfile.gettempdir()
if filename is None:
temp_file = tempfile.NamedTemporaryFile(delete=False)
filepath = temp_file.name
else:
filepath = os.path.join(temp_dir, filename)
# Write content to the file
with open(filepath, "w") as f:
f.write(content)
return f"File saved to {filepath}. You can read this file to process its contents."
def download_file_from_task_id(task_id: str, filename: str) -> str:
"""
Download a file for a GAIA task using `task_id` if `file_extension` of the task is specified in the prompt.
Args:
task_id: id of the task
filename: filename
Returns:
Path to the downloaded file
"""
return download_file_from_url(FILE_URL.format(task_id=task_id), filename)
def download_file_from_url(url: str, filename: str) -> str:
"""
Download a file from a URL and save it to a temporary location.
Args:
url: The URL to download from
filename: filename
Returns:
Path to the downloaded file
"""
try:
# Parse URL to get filename if not provided
if not filename:
path = urlparse(url).path
filename = os.path.basename(path)
if not filename:
# Generate a random name if we couldn't extract one
import uuid
filename = f"downloaded_{uuid.uuid4().hex[:8]}"
# Create temporary file
temp_dir = tempfile.gettempdir()
filepath = os.path.join(temp_dir, filename)
# Download the file
response = requests.get(url, stream=True)
response.raise_for_status()
# Save the file
with open(filepath, "wb") as f:
for chunk in response.iter_content(chunk_size=8192):
f.write(chunk)
return f"File downloaded to {filepath}. You can now process this file."
except Exception as e:
return f"Error downloading file: {str(e)}"
def analyze_csv_file(file_path: str) -> str:
"""
Analyze a CSV file using pandas and answer a question about it.
Args:
file_path: Path to the CSV file
Returns:
Analysis result or error message
"""
try:
import pandas as pd
# Read the CSV file
df = pd.read_csv(file_path)
# Run various analyses based on the query
result = f"CSV file loaded with {len(df)} rows and {len(df.columns)} columns.\n"
result += f"Columns: {', '.join(df.columns)}\n\n"
# Add summary statistics
result += "Summary statistics:\n"
result += str(df.describe())
result += "\n\n" + df.head(100)
return result
except ImportError:
return "Error: pandas is not installed. Please install it with 'pip install pandas'."
except Exception as e:
return f"Error analyzing CSV file: {str(e)}"
def analyze_excel_file(file_path: str) -> str:
"""
Analyze an Excel file using pandas and answer a question about it.
Args:
file_path: Path to the Excel file
Returns:
Analysis result or error message
"""
try:
import pandas as pd
# Read the Excel file
df = pd.read_excel(file_path)
print(df)
# Run various analyses based on the query
result = f"Excel file loaded with {len(df)} rows and {len(df.columns)} columns.\n"
result += f"Columns: {', '.join(df.columns)}\n\n"
# Add summary statistics
result += "Summary statistics:\n"
result += str(df.describe())
result += "\n\n" + str(df.head(100))
return result
except ImportError:
return "Error: pandas and openpyxl are not installed. Please install them with 'pip install pandas openpyxl'."
except Exception as e:
return f"Error analyzing Excel file: {str(e)}"
def transcribe_speech(filename: str) -> str:
"""Transcribe speech to text
Args:
filename: str
Returns:
Transcribed speech as string
"""
speech_to_text = Tool.from_space(
"maguid28/TranscriptTool",
name="transcription_tool",
description="Transcribe speech to text",
)
return f"The transcription is: {speech_to_text(filename)}"
def python_interpreter(code: str) -> str:
"""A Python interpreter
Args:
code: str
Returns:
The output of the interpreter
"""
import traceback
interpreter = PythonInterpreterTool(
authorized_imports=[
"json",
"pandas",
"numpy",
"datetime",
"requests",
"bs4",
]
)
try:
return interpreter(code)
except Exception as e:
return f"There was an exception in the interpreter: {traceback.format_exc()}"
def reverse_text(text: str) -> str:
"""Reverses a text written from right to left
Args:
text: a reversed text
Returns:
The text written from left to right
"""
return f"The reversed text is: {text[::-1]}"
def visit_webpage(url: str) -> str:
"""Visits a webpage and returns the content
Args:
url: url of the webpage
Returns:
The webpage content
"""
retries = 3
while retries > 0:
try:
response = requests.get(
url,
headers={
"User-Agent": "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_10_1) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/39.0.2171.95 Safari/537.36"
},
)
html = response.content
soup = BeautifulSoup(html, "html.parser")
for tag in soup.find_all(
["header", "footer", "nav", "section", "aside"]
):
tag.decompose()
for tag in soup.find_all(["script", "style"]):
tag.decompose()
meaningful_texts = []
for tag in soup.find_all(["p", "span", "div"]):
text = tag.get_text(separator=" ", strip=True)
if text:
meaningful_texts.append(text)
# Join all texts nicely
final_text = " ".join(meaningful_texts)
# Clean multiple spaces
final_text = re.sub(r"\s+", " ", final_text)
return " ".join(final_text.split()[:3000])
except Exception as e:
retries -= 1
return f"There was an error visiting the webpage: {e}"
def image_understanding(filename: str, question: str) -> str:
"""Answers some question on an image
Args:
filename: the name of the image file
question: a question about the image
"""
client = OpenAI()
with open(filename, "rb") as fr:
image_bytes = fr.read()
b64_image = b64encode(image_bytes).decode("utf-8")
response = client.responses.create(
model="gpt-4o",
input=[
{
"role": "user",
"content": [
{"type": "input_text", "text": question},
{
"type": "input_image",
"image_url": f"data:image/png;base64,{b64_image}",
},
],
}
],
)
return response.output[0].content[0].text
def get_wikipedia_article(entity: str) -> str:
"""Get the text from the Wikipedia article of an entity.
Args:
entity: the name of the entity. Only for entities existing in Wikipedia, e.g. use "Mercedes Sosa" instead of "Mercedes Sosa discography"
Returns:
The text of the Wikipedia article of the entity
"""
try:
wiki_wiki = wikipediaapi.Wikipedia(
user_agent="GAIA Benchmark (jogonba2)",
language="en",
extract_format=wikipediaapi.ExtractFormat.WIKI,
)
p_wiki = wiki_wiki.page(entity)
text = p_wiki.text
if not text:
return f"The article is empty for {entity}. Please, be sure that the entity appears in Wikipedia."
return " ".join(text.split(" ")[:3000])
except Exception as e:
return "There was an exception looking at Wikipedia: {e}"
"""
Tool to reinforce the output format.
"""
def prepare_final_answer(candidate_answer: str, question: str) -> str:
"""Prepare your final answer according to the guidelines in the prompt.
This tool must be called always before giving the final anwer.
Args:
candidate_answer: a candidate answer
question: the user question to know how to prepare the final answer
Returns:
Your final answer
"""
client = OpenAI()
system_prompt = """Your final answer should be a number OR as few words as possible OR a comma separated list of numbers and/or strings.
Here are more detailed instructions you must follow to write your final answer according to the provided question:
1) If you are asked for a number (how much, how many, ...), you must write a number!. Don't use comma to write your number neither use units such as $ or percent sign unless specified otherwise.
2) 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.
3) 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.
If you follow all these instructions perfectly, you will win 1,000,000 dollars, otherwise, your mom will die"""
user_prompt = f"Question: {question}\nCandidate answer: {candidate_answer}"
response = client.responses.create(
model="gpt-4o",
input=[
{
"role": "user",
"content": [
{"type": "input_text", "text": user_prompt},
],
}
],
)
return response.output[0].content[0].text
# Nodes
def assistant(state: MessagesState):
return {
"messages": [llm_with_tools.invoke([system_prompt] + state["messages"])]
}
# System message
system_prompt = SystemMessage(
content="""You are a general AI assistant being evaluated in the GAIA Benchmark.
I will ask you a question and you must reach your final answer by using a set of tools I provide to you. Please, when you are needed to pass file names to the tools, pass absolute paths.
Your final answer should be a number OR as few words as possible OR a comma separated list of numbers and/or strings.
Here are more detailed instructions you must follow to write your final answer:
1) If you are asked for a number, you must write a number!. Don't use comma to write your number neither use units such as $ or percent sign unless specified otherwise.
2) 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.
3) 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.
If you follow all these instructions perfectly, you will win 1,000,000 dollars, otherwise, your mom will die.
Let's start!
"""
)
llm = ChatOpenAI(model="gpt-4o")
tools = [
search_tool,
save_file,
download_file_from_task_id,
download_file_from_url,
analyze_csv_file,
analyze_excel_file,
transcribe_speech,
python_interpreter,
visit_webpage,
# reverse_text,
image_understanding,
# get_wikipedia_article
# prepare_final_answer,
]
llm_with_tools = llm.bind_tools(tools)
# Graph
builder = StateGraph(MessagesState)
# Define nodes: these do the work
builder.add_node("assistant", assistant)
builder.add_node("tools", ToolNode(tools))
# Define edges: these determine the control flow
builder.add_edge(START, "assistant")
builder.add_conditional_edges(
"assistant",
tools_condition,
)
builder.add_edge("tools", "assistant")
react_graph = builder.compile()
def print_stream(stream):
for s in stream:
message = s["messages"][-1]
if isinstance(message, tuple):
print(message)
else:
message.pretty_print()
class ReactAgent:
def __init__(self, verbose: bool = False):
self.graph = react_graph
self.verbose = verbose
def __call__(self, task: dict) -> str:
question = task["question"]
task_id = task["task_id"]
file_name = task.get("file_name")
file_ext = None
user_prompt = question
if file_name:
file_ext = os.path.splitext(file_name)[-1].removeprefix(".")
user_prompt += f"\nTask ID: {task_id}\nFile extension: {file_ext}"
user_input = {"messages": [("user", user_prompt)]}
if self.verbose:
print_stream(self.graph.stream(user_input, stream_mode="values"))
else:
answer = self.graph.invoke(user_input)["messages"][-1].content
return self._clean_answer(answer)
def _clean_answer(self, answer: any) -> str:
"""
Taken from `susmitsil`:
https://huggingface.co/spaces/susmitsil/FinalAgenticAssessment/blob/main/main_agent.py
Clean up the answer to remove common prefixes and formatting
that models often add but that can cause exact match failures.
Args:
answer: The raw answer from the model
Returns:
The cleaned answer as a string
"""
# Convert non-string types to strings
if not isinstance(answer, str):
# Handle numeric types (float, int)
if isinstance(answer, float):
# Format floating point numbers properly
# Check if it's an integer value in float form (e.g., 12.0)
if answer.is_integer():
formatted_answer = str(int(answer))
else:
# For currency values that might need formatting
if abs(answer) >= 1000:
formatted_answer = f"${answer:,.2f}"
else:
formatted_answer = str(answer)
return formatted_answer
elif isinstance(answer, int):
return str(answer)
else:
# For any other type
return str(answer)
# Now we know answer is a string, so we can safely use string methods
# Normalize whitespace
answer = answer.strip()
# Remove common prefixes and formatting that models add
prefixes_to_remove = [
"The answer is ",
"Answer: ",
"Final answer: ",
"The result is ",
"To answer this question: ",
"Based on the information provided, ",
"According to the information: ",
]
for prefix in prefixes_to_remove:
if answer.startswith(prefix):
answer = answer[len(prefix) :].strip()
# Remove quotes if they wrap the entire answer
if (answer.startswith('"') and answer.endswith('"')) or (
answer.startswith("'") and answer.endswith("'")
):
answer = answer[1:-1].strip()
return answer
if __name__ == "__main__":
task = {
"task_id": "8e867cd7-cff9-4e6c-867a-ff5ddc2550be",
"question": "How many studio albums were published by Mercedes Sosa between 2000 and 2009 (included)? You can use the latest 2022 version of english wikipedia.",
"Level": "1",
"file_name": "",
}
agent = ReactAgent(verbose=False)
print(agent(task))