docs/API.md

You can access the GAIA benchmark API provided for your agent's evaluation using the following endpoints, as described in the provided documentation:

Base URL

The API's base URL (according to the provided document) is:

https://agents-course-unit4-scoring.hf.space

API Endpoints

1. Retrieve Evaluation Questions
   Endpoint:

   GET /questions
   

   • Returns the full list of filtered evaluation questions.

2. Retrieve a Random Question
   Endpoint:

   GET /random-question
   

   • Fetches a single random question from the available set.

3. Download Associated Files
   Endpoint:

   GET /files/{task_id}
   

   • Downloads files associated with specific tasks, useful for questions that require external data or multimodal analysis.

4. Submit Agent Answers
   Endpoint:

   POST /submit
   

   • Submit your agent's answers to be evaluated against the benchmark.
   • Requires JSON payload structured as:

   {
     "username": "Your Hugging Face username",
     "agent_code": "URL to your Hugging Face Space code repository",
     "answers": [{"task_id": "task identifier", "submitted_answer": "your answer"}]
   }
   

API Usage Example:

Here's an illustrative example using Python and the requests library:

import requests

BASE_URL = "https://agents-course-unit4-scoring.hf.space"

# Retrieve all questions
response = requests.get(f"{BASE_URL}/questions")
questions = response.json()

# Fetch a random question
random_question = requests.get(f"{BASE_URL}/random-question").json()

# Download a file for a specific task_id
task_id = "example_task_id"
file_response = requests.get(f"{BASE_URL}/files/{task_id}")
with open("downloaded_file", "wb") as f:
    f.write(file_response.content)

# Submit answers
submission_payload = {
    "username": "your_username",
    "agent_code": "https://huggingface.co/spaces/your_username/your_space_name/tree/main",
    "answers": [{"task_id": "task_id", "submitted_answer": "answer_text"}]
}
submit_response = requests.post(f"{BASE_URL}/submit", json=submission_payload)
print(submit_response.json())

Ensure you have proper authentication if required, and replace placeholder texts (your_username, task_id, answer_text, etc.) with your actual values.

Let me know if you need further assistance or more detailed examples!