Ubik80's picture
fixes
70b0f89 verified
raw
history blame
2.57 kB
import os
import requests
from smolagents import CodeAgent, tool, OpenAIServerModel
# ------------------------
# Constants
# ------------------------
API_URL = "https://agents-course-unit4-scoring.hf.space"
# ------------------------
# Tool definitions
# ------------------------
@tool
def fetch_questions() -> list:
"""
Fetch the full list of GAIA evaluation questions.
:return: A list of question dictionaries, each with 'task_id' and 'question'.
"""
response = requests.get(f"{API_URL}/questions", timeout=15)
response.raise_for_status()
return response.json()
@tool
def fetch_random_question() -> dict:
"""
Fetch a single random GAIA question.
:return: A dict with keys 'task_id' and 'question'.
"""
response = requests.get(f"{API_URL}/random-question", timeout=15)
response.raise_for_status()
return response.json()
@tool
def fetch_file(task_id: str) -> bytes:
"""
Download a file associated with a given task_id.
:param task_id: The ID of the GAIA task whose file you want to download.
:return: The file content as raw bytes.
"""
response = requests.get(f"{API_URL}/files/{task_id}", timeout=15)
response.raise_for_status()
return response.content
@tool
def submit_answers(
username: str,
agent_code: str,
answers: list
) -> dict:
"""
Submit the agent's answers to GAIA and return the scoring.
:param username: Your Hugging Face username (identifies the submission).
:param agent_code: URL to your HF Space code (for verification).
:param answers: List of dicts each with 'task_id' and 'submitted_answer'.
:return: A dict containing 'score', 'correct_count', 'total_attempted', 'message', etc.
"""
payload = {
"username": username,
"agent_code": agent_code,
"answers": answers
}
response = requests.post(f"{API_URL}/submit", json=payload, timeout=60)
response.raise_for_status()
return response.json()
def create_agent() -> CodeAgent:
"""
Factory that returns a configured CodeAgent instance using OpenAI.
Expects OPENAI_API_KEY in the environment.
"""
openai_model = OpenAIServerModel(model_name="gpt-3.5-turbo")
agent = CodeAgent(
tools=[fetch_questions, fetch_random_question, fetch_file, submit_answers],
model=openai_model,
prompt_template=(
"Here is a GAIA question:\n"
"{question}\n"
"Provide ONLY the exact answer (exact-match) with no extra text."
)
)
return agent