Spaces:
Running
Running
File size: 5,650 Bytes
92a4906 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 |
#!/usr/bin/python3
# -*- coding: utf-8 -*-
"""
https://cloud.google.com/vertex-ai/generative-ai/docs/partner-models/claude/use-claude?hl=zh-cn
"""
import argparse
from datetime import datetime
import json
import os
from pathlib import Path
import sys
import time
import tempfile
from zoneinfo import ZoneInfo # Python 3.9+ 自带，无需安装
pwd = os.path.abspath(os.path.dirname(__file__))
sys.path.append(os.path.join(pwd, "../"))
from google import genai
from google.genai import types
from anthropic import AnthropicVertex
from project_settings import environment, project_path
def get_args():
parser = argparse.ArgumentParser()
parser.add_argument(
"--model_name",
default="claude-opus-4@20250514",
# default="claude-sonnet-4@20250514",
type=str
)
parser.add_argument(
"--eval_dataset_name",
# default="agent-bingoplus-ph-90-choice.jsonl",
default="agent-lingoace-zh-400-choice.jsonl",
# default="arc-easy-1000-choice.jsonl",
type=str
)
parser.add_argument(
"--eval_dataset_dir",
default=(project_path / "data/dataset").as_posix(),
type=str
)
parser.add_argument(
"--eval_data_dir",
default=(project_path / "data/eval_data").as_posix(),
type=str
)
parser.add_argument(
"--client",
default="shenzhen_sase",
type=str
)
parser.add_argument(
"--service",
# default="google_potent_veld_462405_t3",
default="google_nxcloud_312303",
type=str
)
parser.add_argument(
"--create_time_str",
default="null",
# default="20250731_162116",
type=str
)
parser.add_argument(
"--interval",
default=1,
type=int
)
args = parser.parse_args()
return args
def main():
args = get_args()
service = environment.get(args.service, dtype=json.loads)
project_id = service["project_id"]
google_application_credentials = Path(tempfile.gettempdir()) / f"llm_eval_system/{project_id}.json"
google_application_credentials.parent.mkdir(parents=True, exist_ok=True)
with open(google_application_credentials.as_posix(), "w", encoding="utf-8") as f:
content = json.dumps(service, ensure_ascii=False, indent=4)
f.write(f"{content}\n")
os.environ["GOOGLE_APPLICATION_CREDENTIALS"] = google_application_credentials.as_posix()
eval_dataset_dir = Path(args.eval_dataset_dir)
eval_dataset_dir.mkdir(parents=True, exist_ok=True)
eval_data_dir = Path(args.eval_data_dir)
eval_data_dir.mkdir(parents=True, exist_ok=True)
if args.create_time_str == "null":
tz = ZoneInfo("Asia/Shanghai")
now = datetime.now(tz)
create_time_str = now.strftime("%Y%m%d_%H%M%S")
# create_time_str = "20250729-interval-5"
else:
create_time_str = args.create_time_str
eval_dataset = eval_dataset_dir / args.eval_dataset_name
output_file = eval_data_dir / f"google_anthropic/anthropic/{args.model_name}/{args.client}/{args.service}/{create_time_str}/{args.eval_dataset_name}"
output_file.parent.mkdir(parents=True, exist_ok=True)
client = AnthropicVertex(project_id=project_id, region="us-east5")
total = 0
total_correct = 0
# finished
finished_idx_set = set()
if os.path.exists(output_file.as_posix()):
with open(output_file.as_posix(), "r", encoding="utf-8") as f:
for row in f:
row = json.loads(row)
idx = row["idx"]
total = row["total"]
total_correct = row["total_correct"]
finished_idx_set.add(idx)
print(f"finished count: {len(finished_idx_set)}")
with open(eval_dataset.as_posix(), "r", encoding="utf-8") as fin, open(output_file.as_posix(), "a+", encoding="utf-8") as fout:
for row in fin:
row = json.loads(row)
idx = row["idx"]
prompt = row["prompt"]
response = row["response"]
if idx in finished_idx_set:
continue
finished_idx_set.add(idx)
try:
time.sleep(args.interval)
print(f"sleep: {args.interval}")
time_begin = time.time()
message = client.messages.create(
model=args.model_name,
max_tokens=1024,
messages=[
{
"role": "user",
"content": prompt,
}
],
)
time_cost = time.time() - time_begin
print(f"time_cost: {time_cost}")
except Exception as e:
print(f"request failed, error type: {type(e)}, error text: {str(e)}")
continue
prediction = message.content[0].text
correct = 1 if prediction == response else 0
total += 1
total_correct += correct
score = total_correct / total
row_ = {
"idx": idx,
"prompt": prompt,
"response": response,
"prediction": prediction,
"correct": correct,
"total": total,
"total_correct": total_correct,
"score": score,
"time_cost": time_cost,
}
row_ = json.dumps(row_, ensure_ascii=False)
fout.write(f"{row_}\n")
fout.flush()
return
if __name__ == "__main__":
main()
|