Upload gpt_judge.py
Browse files- metric/gpt_judge.py +163 -0
metric/gpt_judge.py
ADDED
|
@@ -0,0 +1,163 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
import openai
|
| 3 |
+
from openai import OpenAI
|
| 4 |
+
import base64
|
| 5 |
+
from tqdm import tqdm
|
| 6 |
+
import json
|
| 7 |
+
import os
|
| 8 |
+
from typing import Any
|
| 9 |
+
from threading import Lock
|
| 10 |
+
from pathlib import Path
|
| 11 |
+
import time
|
| 12 |
+
|
| 13 |
+
import json
|
| 14 |
+
import os
|
| 15 |
+
from pathlib import Path
|
| 16 |
+
from threading import Lock
|
| 17 |
+
from typing import Any
|
| 18 |
+
import re
|
| 19 |
+
|
| 20 |
+
_json_write_lock = Lock()
|
| 21 |
+
|
| 22 |
+
def save_json_file(
|
| 23 |
+
data: Any,
|
| 24 |
+
file_path: str,
|
| 25 |
+
indent: int = 4,
|
| 26 |
+
temp_suffix: str = ".tmp"
|
| 27 |
+
) -> None:
|
| 28 |
+
|
| 29 |
+
path = Path(file_path)
|
| 30 |
+
path.parent.mkdir(parents=True, exist_ok=True)
|
| 31 |
+
|
| 32 |
+
temp_path = f"{file_path}{temp_suffix}"
|
| 33 |
+
|
| 34 |
+
with _json_write_lock:
|
| 35 |
+
try:
|
| 36 |
+
|
| 37 |
+
with open(temp_path, "w", encoding="utf-8") as f:
|
| 38 |
+
json.dump(data, f, ensure_ascii=False, indent=indent)
|
| 39 |
+
|
| 40 |
+
f.flush()
|
| 41 |
+
os.fsync(f.fileno())
|
| 42 |
+
|
| 43 |
+
os.replace(temp_path, file_path)
|
| 44 |
+
|
| 45 |
+
except Exception as e:
|
| 46 |
+
try:
|
| 47 |
+
if os.path.exists(temp_path):
|
| 48 |
+
os.remove(temp_path)
|
| 49 |
+
except OSError:
|
| 50 |
+
pass
|
| 51 |
+
raise RuntimeError(f"save json failed: {e}") from e
|
| 52 |
+
|
| 53 |
+
def read_json_file(file_path):
|
| 54 |
+
"""
|
| 55 |
+
Reads a JSON file and returns the parsed data as a Python object.
|
| 56 |
+
|
| 57 |
+
:param file_path: The path to the JSON file
|
| 58 |
+
:return: The data parsed from the JSON file
|
| 59 |
+
"""
|
| 60 |
+
with open(file_path, 'r', encoding='utf-8') as f:
|
| 61 |
+
data = json.load(f)
|
| 62 |
+
return data
|
| 63 |
+
|
| 64 |
+
def openai_api( prompt = None):
|
| 65 |
+
if prompt == None:
|
| 66 |
+
prompt = "What's in this image?"
|
| 67 |
+
|
| 68 |
+
client = OpenAI(
|
| 69 |
+
base_url='your_url',
|
| 70 |
+
|
| 71 |
+
api_key='your_key'
|
| 72 |
+
)
|
| 73 |
+
response = client.chat.completions.create(
|
| 74 |
+
model="gpt-4o",
|
| 75 |
+
messages=[
|
| 76 |
+
{
|
| 77 |
+
"role": "user",
|
| 78 |
+
"content": [
|
| 79 |
+
{"type": "text", "text": prompt}
|
| 80 |
+
],
|
| 81 |
+
}
|
| 82 |
+
],
|
| 83 |
+
max_tokens=5000,
|
| 84 |
+
)
|
| 85 |
+
|
| 86 |
+
return response
|
| 87 |
+
|
| 88 |
+
|
| 89 |
+
def extract_last_bracket_list(s: str) -> list:
|
| 90 |
+
"""
|
| 91 |
+
"""
|
| 92 |
+
last_open = s.rfind('[')
|
| 93 |
+
last_close = s.rfind(']')
|
| 94 |
+
content = s[last_open:last_close+1]
|
| 95 |
+
|
| 96 |
+
try:
|
| 97 |
+
|
| 98 |
+
result = json.loads(content)
|
| 99 |
+
except:
|
| 100 |
+
content = re.sub(r'[\n\t\r]', '', content)
|
| 101 |
+
content = ' '.join(content.split())
|
| 102 |
+
result = []
|
| 103 |
+
content = content.split("\",")
|
| 104 |
+
for item in content:
|
| 105 |
+
for item2 in item.split("\',"):
|
| 106 |
+
result.append(item2)
|
| 107 |
+
return result
|
| 108 |
+
|
| 109 |
+
|
| 110 |
+
def ads_task_llm_judge(jsonlist):
|
| 111 |
+
for json_file in jsonlist:
|
| 112 |
+
print(json_file)
|
| 113 |
+
max_retries = 4
|
| 114 |
+
retry_wait = 10
|
| 115 |
+
tasks = read_json_file(json_file)
|
| 116 |
+
for item in tqdm(tasks):
|
| 117 |
+
if "judge" in item: continue
|
| 118 |
+
|
| 119 |
+
prompt_templet = 'Please help me determine if the content in the Description contains Key Information. If it does, answer directly with "Yes"; if it does not, answer directly with "No". Please respond only with "Yes" or "No", without any additional output.'
|
| 120 |
+
|
| 121 |
+
judge_list = []
|
| 122 |
+
for keypoint in item["gt"]:
|
| 123 |
+
response = item["response"]
|
| 124 |
+
if isinstance(response, list):
|
| 125 |
+
content = ""
|
| 126 |
+
for text in response:
|
| 127 |
+
content += text
|
| 128 |
+
if isinstance(response, str):
|
| 129 |
+
content = response
|
| 130 |
+
prompt = prompt_templet +"\n"+ "Description: " + content +"\n"+ "Key Information: " + keypoint
|
| 131 |
+
|
| 132 |
+
for attempt in range(max_retries):
|
| 133 |
+
try:
|
| 134 |
+
response = openai_api(prompt)
|
| 135 |
+
current_judge = response.choices[0].message.content
|
| 136 |
+
judge_list.append(current_judge)
|
| 137 |
+
break
|
| 138 |
+
except Exception as e:
|
| 139 |
+
print(f"[Warning] Request failed: {e}")
|
| 140 |
+
if attempt < max_retries - 1:
|
| 141 |
+
print(f"Retrying in {retry_wait} seconds... (attempt {attempt + 1})")
|
| 142 |
+
time.sleep(retry_wait)
|
| 143 |
+
else:
|
| 144 |
+
print("[Error] Reached max retries. Skipping this item.")
|
| 145 |
+
item["error"] = str(e)
|
| 146 |
+
item["judge"] = judge_list
|
| 147 |
+
print(item["judge"])
|
| 148 |
+
for attempt in range(max_retries):
|
| 149 |
+
try:
|
| 150 |
+
save_json_file(tasks, json_file)
|
| 151 |
+
break
|
| 152 |
+
except Exception as e:
|
| 153 |
+
if attempt < max_retries - 1:
|
| 154 |
+
print(f"save error)")
|
| 155 |
+
time.sleep(2)
|
| 156 |
+
else:
|
| 157 |
+
print("[Error] Reached SAVE max retries. Skipping this item.")
|
| 158 |
+
|
| 159 |
+
if __name__ == "__main__":
|
| 160 |
+
jsonlist = [
|
| 161 |
+
r"your_path/gpt4o_bench.json",
|
| 162 |
+
]
|
| 163 |
+
ads_task_llm_judge(jsonlist)
|