File size: 5,460 Bytes
b0c0df0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
import datetime
import json
import os
import re
import sys
import time
from pathlib import Path

import requests
import yaml
from loguru import logger as eval_logger

import lmms_eval.tasks._task_utils.file_utils as file_utils
from lmms_eval.filters.extraction import ExtendedRegexFilter


def clotho_aqa_doc_to_audio(doc):
    return [doc["audio"]]


def clotho_aqa_doc_to_text(doc, lmms_eval_specific_kwargs):
    question = doc["question"]
    pre_prompt = lmms_eval_specific_kwargs["pre_prompt"]
    post_prompt = lmms_eval_specific_kwargs["post_prompt"]
    return f"{pre_prompt}{question}{post_prompt}"


# functions for the clotho_asqa_v2 task, need to be tested later

with open(Path(__file__).parent / "_default_template_yaml", "r") as f:
    raw_data = f.readlines()
    safe_data = []
    for i, line in enumerate(raw_data):
        # remove function definition since yaml load cannot handle it
        if "!function" not in line:
            safe_data.append(line)

    config = yaml.safe_load("".join(safe_data))


NUM_SECONDS_TO_SLEEP = 2
GPT_EVAL_MODEL_NAME = os.getenv("MODEL_VERSION", "gpt-4o-2024-11-20")
API_TYPE = os.getenv("API_TYPE", "azure")

if API_TYPE == "openai":
    API_URL = os.getenv("OPENAI_API_URL", "https://api.openai.com/v1/chat/completions")
    API_KEY = os.getenv("OPENAI_API_KEY", "YOUR_API_KEY")
    headers = {
        "Authorization": f"Bearer {API_KEY}",
        "Content-Type": "application/json",
    }

elif API_TYPE == "azure":
    API_URL = os.getenv("AZURE_ENDPOINT", "https://api.cognitive.microsoft.com/sts/v1.0/issueToken")
    API_KEY = os.getenv("AZURE_API_KEY", "YOUR_API_KEY")
    headers = {
        "api-key": API_KEY,
        "Content-Type": "application/json",
    }

eval_prompt = """
            [Question]
            {question}

            [Reference Answer]
            {ground_truth}

            [Model Answer]
            {model_response}

            [Task]
            Rate the model's answer based on its alignment with the reference answer, focusing on accuracy and relevance to the reference provided. Please be critical on the details.
            Criteria: Assess if the model's response mirrors the reference in terms of content, accuracy, and relevance.
            Score0: The answer is completely misaligned, providing incorrect or irrelevant information compared to the reference.
            Score1: The answer shows minimal alignment, often misunderstanding or providing irrelevant details unrelated to the reference.
            Score2: The answer recognizes the topic but diverges significantly from the reference in accuracy or relevance.
            Score3: The answer aligns with the reference generally but lacks detail or precise accuracy in some aspects.
            Score4: The answer is mostly accurate and relevant, closely following the reference but could be clearer or more detailed.
            Score5: The answer is highly accurate, detailed, and matches the reference answer perfectly, capturing its essence and detail.

            Your response should be formatted as follows:
            Explanation: (Provide a concise explanation of your rating, comparing the reference answer with the model's response. "The reference answer is [XXX], while the model's answer is [YYY]. I think ...")
            Rating: (int)"""


retries = 3
NUM_SECONDS_TO_SLEEP = 5


def get_eval(max_tokens: int, content: str, retries: int = retries):
    global headers

    messages = [
        {"role": "user", "content": content},
    ]

    payload = {"model": GPT_EVAL_MODEL_NAME, "messages": messages, "temperature": 0.7, "max_tokens": max_tokens, "top_p": 0.95, "frequency_penalty": 0, "presence_penalty": 0, "stop": None}

    for attempt in range(retries):
        try:
            response = requests.post(API_URL, headers=headers, json=payload, timeout=60)
            response.raise_for_status()
            response_data = response.json()

            content = response_data["choices"][0]["message"]["content"].strip()
            if content != "":
                return content, response_data["model"]
            break  # If successful, break out of the loop

        except Exception as e:
            eval_logger.info(f"Attempt {attempt + 1} failed with error: {e}")
            if attempt < retries:  # If we have retries left, sleep and then continue to next attempt
                time.sleep(NUM_SECONDS_TO_SLEEP)
            else:  # If this was the last attempt, log and return empty
                eval_logger.error(f"All {retries} attempts failed. Last error message: {e}")
                return "", ""
    return "", ""


def clotho_aqa_v2_process_results(doc, result):
    pred = result[0]
    ground_truth_str = doc["answer"]
    content = eval_prompt.format(model_response=pred, ground_truth=ground_truth_str, question=doc["question"])
    eval_answer, model_name = get_eval(max_tokens=1024, content=content)
    return {
        "gpt_eval": {"eval_answer": eval_answer, "model_name": model_name},
    }


def clotho_aqa_v2_aggregate_results(results):
    score = 0
    for result in results:
        eval_answer = result["eval_answer"]
        eval_score = re.search(r"([0-5])", eval_answer).group(1)
        try:
            eval_score = float(eval_score)
        except Exception as e:
            eval_logger.error(f"Error parsing eval_score: {e}")
            eval_score = 0.0
        score += eval_score

    return score / len(results) * 20