File size: 3,229 Bytes
b2b9de7
4be3af4
 
b2b9de7
4be3af4
 
b2b9de7
4be3af4
b2b9de7
4be3af4
b2b9de7
4be3af4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b2b9de7
4be3af4
 
 
b2b9de7
 
4be3af4
 
b2b9de7
 
4be3af4
 
 
b2b9de7
 
 
 
 
 
 
 
 
4be3af4
b2b9de7
 
4be3af4
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
# module3.py
import requests
from typing import Optional
import logging
from dotenv import load_dotenv
import os

# Set up logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

# .env 파일 λ‘œλ“œ
load_dotenv()

# Hugging Face API 정보
API_URL = "https://api-inference.huggingface.co/models/meta-llama/Meta-Llama-3-8B-Instruct"
API_KEY = os.getenv("HUGGINGFACE_API_KEY")

if not API_KEY:
    raise ValueError("API_KEYκ°€ μ„€μ •λ˜μ§€ μ•Šμ•˜μŠ΅λ‹ˆλ‹€. .env νŒŒμΌμ„ ν™•μΈν•˜μ„Έμš”.")

class AnswerVerifier:
    def verify_answer(self, question: str, choices: dict) -> Optional[str]:
        """μ£Όμ–΄μ§„ λ¬Έμ œμ™€ 보기λ₯Ό λ°”νƒ•μœΌλ‘œ 정닡을 검증"""
        try:
            prompt = self._create_prompt(question, choices)
            headers = {"Authorization": f"Bearer {API_KEY}"}
            
            response = requests.post(
                API_URL,
                headers=headers,
                json={"inputs": prompt}
            )
            response.raise_for_status()
            
            response_data = response.json()
            logger.debug(f"Raw API response: {response_data}")
            
            # API 응닡 처리
            generated_text = ""
            if isinstance(response_data, list):
                if response_data and isinstance(response_data[0], dict):
                    generated_text = response_data[0].get('generated_text', '')
                else:
                    generated_text = response_data[0] if response_data else ''
            elif isinstance(response_data, dict):
                generated_text = response_data.get('generated_text', '')
            else:
                generated_text = str(response_data)
            
            verified_answer = self._extract_answer(generated_text)
            logger.info(f"Verified answer: {verified_answer}")
            return verified_answer

        except Exception as e:
            logger.error(f"Error in verify_answer: {e}")
            return None

    def _create_prompt(self, question: str, choices: dict) -> str:
        """검증을 μœ„ν•œ ν”„λ‘¬ν”„νŠΈ 생성"""
        return f"""
        <|begin_of_text|>
        <|start_header_id|>system<|end_header_id|>
        You are an expert mathematics teacher checking student answers. 
        Please analyze the following question and select the single best answer.
        Output ONLY the letter of the correct answer (A, B, C, or D) without any explanation.
        <|eot_id|>
        <|start_header_id|>user<|end_header_id|>
        Question: {question}

        A) {choices['A']}
        B) {choices['B']}
        C) {choices['C']}
        D) {choices['D']}

        Select the correct answer letter (A, B, C, or D):
        <|eot_id|>
        <|start_header_id|>assistant<|end_header_id|>
        """.strip()

    def _extract_answer(self, response: str) -> Optional[str]:
        """μ‘λ‹΅μ—μ„œ A, B, C, D 쀑 ν•˜λ‚˜λ₯Ό μΆ”μΆœ"""
        response = response.strip().upper()
        valid_answers = {'A', 'B', 'C', 'D'}
        
        # μ‘λ‹΅μ—μ„œ μœ νš¨ν•œ λ‹΅μ•ˆ μ°ΎκΈ°
        for answer in valid_answers:
            if answer in response:
                return answer
        
        return None