|
import os |
|
import io |
|
import pandas as pd |
|
import requests |
|
from openai import OpenAI |
|
|
|
class GaiaAgent: |
|
def __init__(self): |
|
self.client = OpenAI(api_key=os.getenv("OPENAI_API_KEY")) |
|
self.instructions = ( |
|
"You are an expert assistant solving GAIA benchmark questions. " |
|
"You analyze file contents (like CSV), reason step-by-step, and respond with a single factual answer." |
|
) |
|
self.api_url = "https://agents-course-unit4-scoring.hf.space" |
|
|
|
def fetch_file_context(self, task_id: str) -> str: |
|
try: |
|
url = f"{self.api_url}/files/{task_id}" |
|
response = requests.get(url, timeout=10) |
|
response.raise_for_status() |
|
content_type = response.headers.get("Content-Type", "") |
|
|
|
if "text/csv" in content_type or url.endswith(".csv"): |
|
df = pd.read_csv(io.StringIO(response.text)) |
|
if df.shape[1] <= 15 and df.shape[0] <= 30: |
|
return f"CSV table preview:\n{df.to_markdown(index=False)}" |
|
else: |
|
return f"CSV summary: {df.shape[0]} rows, {df.shape[1]} columns.\nColumns: {', '.join(df.columns[:10])}" |
|
|
|
elif "application/json" in content_type: |
|
return f"JSON content:\n{response.text[:2000]}" |
|
|
|
elif "application/pdf" in content_type: |
|
return "[PDF detected. You may need to request OCR summary.]" |
|
|
|
elif "text/plain" in content_type: |
|
return f"File preview:\n{response.text[:2000]}" |
|
|
|
else: |
|
return f"[Unsupported file type: {content_type}]" |
|
|
|
except Exception as e: |
|
return f"[Error downloading or processing file: {e}]" |
|
|
|
def __call__(self, question: str, task_id: str = None) -> str: |
|
file_context = "" |
|
if task_id: |
|
file_context = self.fetch_file_context(task_id) |
|
if file_context: |
|
file_context = f"FILE CONTEXT:\n{file_context}\n" |
|
|
|
prompt = f"{self.instructions}\n\n{file_context}QUESTION:\n{question}\nANSWER:" |
|
|
|
response = self.client.chat.completions.create( |
|
model="gpt-4-turbo", |
|
messages=[ |
|
{"role": "system", "content": self.instructions}, |
|
{"role": "user", "content": prompt} |
|
], |
|
temperature=0.0, |
|
) |
|
|
|
return response.choices[0].message.content.strip() |
|
|