Spaces:
Runtime error
Runtime error
Delete gaia_test.py
Browse files- gaia_test.py +0 -895
gaia_test.py
DELETED
|
@@ -1,895 +0,0 @@
|
|
| 1 |
-
"""
|
| 2 |
-
测试脚本:支持 validation 和 test 数据集,支持全量、增量、混合三种模式
|
| 3 |
-
- 全量模式(full):不管文件是否存在,都删除重新开始
|
| 4 |
-
- 增量模式(incremental):如果文件存在则增量,不存在则全量执行
|
| 5 |
-
- 混合模式(hybrid):第一次时全量(文件不存在),后面就增量(文件存在)
|
| 6 |
-
"""
|
| 7 |
-
import argparse
|
| 8 |
-
import json
|
| 9 |
-
import os
|
| 10 |
-
import re
|
| 11 |
-
import time
|
| 12 |
-
import traceback
|
| 13 |
-
from collections import OrderedDict
|
| 14 |
-
from concurrent.futures import ThreadPoolExecutor, as_completed
|
| 15 |
-
from threading import Lock
|
| 16 |
-
|
| 17 |
-
import pandas as pd
|
| 18 |
-
import requests
|
| 19 |
-
from datasets import load_dataset
|
| 20 |
-
from huggingface_hub import snapshot_download, hf_hub_download
|
| 21 |
-
from pathlib import Path
|
| 22 |
-
import shutil
|
| 23 |
-
|
| 24 |
-
# --- 1. 配置区 ---
|
| 25 |
-
BASE_URL = "http://localhost:5173/api/v1"
|
| 26 |
-
CHAT_URL = f"{BASE_URL}/sessions/chat"
|
| 27 |
-
UPLOAD_URL = f"{BASE_URL}/files"
|
| 28 |
-
HEADERS = {"Authorization": "Bearer hawk_YhCZLQYqtPOwOiEyEgeCNdfAFAbrHtTUxQvRiaOInyekgVgE"}
|
| 29 |
-
DATA_PATH = "./gaia_data"
|
| 30 |
-
REQUEST_TIMEOUT = 1800
|
| 31 |
-
MAX_CONCURRENT = 2
|
| 32 |
-
|
| 33 |
-
file_lock = Lock()
|
| 34 |
-
|
| 35 |
-
# 全局变量,根据 split 类型动态设置
|
| 36 |
-
dataset = None
|
| 37 |
-
OUTPUT_FILE = None
|
| 38 |
-
SUBMISSION_FILE = None
|
| 39 |
-
SPLIT_TYPE = None # "validation" 或 "test"
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
def check_and_download_dataset_files():
|
| 43 |
-
"""
|
| 44 |
-
检查并下载完整的 GAIA 数据集文件到本地目录(包含 validation 和 test 的所有文件)
|
| 45 |
-
|
| 46 |
-
Returns:
|
| 47 |
-
bool: 如果文件已存在或下载成功返回 True,否则返回 False
|
| 48 |
-
"""
|
| 49 |
-
base_target_dir = Path(DATA_PATH) / "2023"
|
| 50 |
-
validation_dir = base_target_dir / "validation"
|
| 51 |
-
test_dir = base_target_dir / "test"
|
| 52 |
-
|
| 53 |
-
# 检查两个目录是否都存在且有文件
|
| 54 |
-
validation_files = list(validation_dir.glob("*")) if validation_dir.exists() else []
|
| 55 |
-
test_files = list(test_dir.glob("*")) if test_dir.exists() else []
|
| 56 |
-
|
| 57 |
-
if validation_files and test_files:
|
| 58 |
-
print(f"✅ 检测到数据集文件已存在")
|
| 59 |
-
print(f" validation 文件数: {len(validation_files)}")
|
| 60 |
-
print(f" test 文件数: {len(test_files)}")
|
| 61 |
-
return True
|
| 62 |
-
|
| 63 |
-
# 需要下载数据集文件
|
| 64 |
-
print(f"📥 开始下载完整的 GAIA 数据集文件...")
|
| 65 |
-
print(f" 目标目录: {base_target_dir}")
|
| 66 |
-
|
| 67 |
-
try:
|
| 68 |
-
# 创建基础目录
|
| 69 |
-
base_target_dir.mkdir(parents=True, exist_ok=True)
|
| 70 |
-
|
| 71 |
-
# 下载完整数据集到临时目录,然后复制到目标目录
|
| 72 |
-
print(" 步骤 1/4: 正在从 Hugging Face 下载完整数据集...")
|
| 73 |
-
print(" 提示: 下载进度会显示在下方,请耐心等待...")
|
| 74 |
-
download_start = time.time()
|
| 75 |
-
|
| 76 |
-
# 使用 snapshot_download 下载完整数据集
|
| 77 |
-
cache_dir = snapshot_download(
|
| 78 |
-
repo_id="gaia-benchmark/GAIA",
|
| 79 |
-
repo_type="dataset",
|
| 80 |
-
local_dir=None, # 使用默认缓存目录
|
| 81 |
-
resume_download=True
|
| 82 |
-
)
|
| 83 |
-
|
| 84 |
-
download_duration = time.time() - download_start
|
| 85 |
-
print(f" ✅ 数据集下载完成,耗时 {download_duration:.2f} 秒")
|
| 86 |
-
|
| 87 |
-
cache_path = Path(cache_dir)
|
| 88 |
-
source_2023_dir = cache_path / "2023"
|
| 89 |
-
|
| 90 |
-
if not source_2023_dir.exists():
|
| 91 |
-
print(f" ❌ 错误: 缓存目录中未找到 2023 目录")
|
| 92 |
-
return False
|
| 93 |
-
|
| 94 |
-
# 复制 validation 和 test 目录
|
| 95 |
-
print(" 步骤 2/4: 正在复制 validation 文件...")
|
| 96 |
-
validation_source = source_2023_dir / "validation"
|
| 97 |
-
if validation_source.exists():
|
| 98 |
-
if validation_dir.exists():
|
| 99 |
-
shutil.rmtree(validation_dir)
|
| 100 |
-
shutil.copytree(validation_source, validation_dir)
|
| 101 |
-
validation_count = len(list(validation_dir.glob("*")))
|
| 102 |
-
print(f" ✅ validation 文件复制完成,共 {validation_count} 个文件")
|
| 103 |
-
else:
|
| 104 |
-
print(f" ⚠️ 警告: 未找到 validation 目录")
|
| 105 |
-
|
| 106 |
-
print(" 步骤 3/4: 正在复制 test 文件...")
|
| 107 |
-
test_source = source_2023_dir / "test"
|
| 108 |
-
if test_source.exists():
|
| 109 |
-
if test_dir.exists():
|
| 110 |
-
shutil.rmtree(test_dir)
|
| 111 |
-
shutil.copytree(test_source, test_dir)
|
| 112 |
-
test_count = len(list(test_dir.glob("*")))
|
| 113 |
-
print(f" ✅ test 文件复制完成,共 {test_count} 个文件")
|
| 114 |
-
else:
|
| 115 |
-
print(f" ⚠️ 警告: 未找到 test 目录")
|
| 116 |
-
|
| 117 |
-
print(" 步骤 4/4: 数据集文件准备完成!")
|
| 118 |
-
print(f" 目标目录: {base_target_dir}")
|
| 119 |
-
return True
|
| 120 |
-
|
| 121 |
-
except Exception as e:
|
| 122 |
-
print(f" ❌ 下载数据集文件时出错: {e}")
|
| 123 |
-
import traceback
|
| 124 |
-
traceback.print_exc()
|
| 125 |
-
return False
|
| 126 |
-
|
| 127 |
-
|
| 128 |
-
def build_ordered_record(task_id, question, level, agent_answer, duration, has_file,
|
| 129 |
-
session_id=None, attachment_name=None, ground_truth=None, is_correct=None):
|
| 130 |
-
"""
|
| 131 |
-
按照固定顺序构建记录字典,确保字段��序一致
|
| 132 |
-
|
| 133 |
-
Args:
|
| 134 |
-
task_id: 任务ID
|
| 135 |
-
question: 问题
|
| 136 |
-
level: 难度级别
|
| 137 |
-
agent_answer: Agent答案
|
| 138 |
-
duration: 执行时长
|
| 139 |
-
has_file: 是否有文件
|
| 140 |
-
session_id: 会话 ID
|
| 141 |
-
attachment_name: 附件名称(如果有附件)
|
| 142 |
-
ground_truth: 标准答案(仅validation数据集)
|
| 143 |
-
is_correct: 是否正确(仅validation数据集)
|
| 144 |
-
|
| 145 |
-
Returns:
|
| 146 |
-
OrderedDict: 按固定顺序排列的记录
|
| 147 |
-
"""
|
| 148 |
-
record = OrderedDict()
|
| 149 |
-
record["task_id"] = task_id
|
| 150 |
-
record["question"] = question
|
| 151 |
-
record["level"] = level
|
| 152 |
-
record["duration"] = duration
|
| 153 |
-
record["has_file"] = has_file
|
| 154 |
-
# attachment_name: 如果有值就写入(即使 agent 出错也应该写入)
|
| 155 |
-
# 只要 attachment_name 不是 None 且不是空字符串,就写入
|
| 156 |
-
if attachment_name and attachment_name.strip():
|
| 157 |
-
record["attachment_name"] = attachment_name
|
| 158 |
-
# session_id: 如果有值就写入(如果 agent 出错可能为 None,不写入是合理的)
|
| 159 |
-
if session_id:
|
| 160 |
-
record["session_id"] = session_id
|
| 161 |
-
record["agent_answer"] = agent_answer
|
| 162 |
-
# validation 数据集特有字段
|
| 163 |
-
if ground_truth is not None:
|
| 164 |
-
record["ground_truth"] = ground_truth
|
| 165 |
-
if is_correct is not None:
|
| 166 |
-
record["is_correct"] = is_correct
|
| 167 |
-
|
| 168 |
-
return record
|
| 169 |
-
|
| 170 |
-
|
| 171 |
-
def load_existing_results():
|
| 172 |
-
"""
|
| 173 |
-
加载已有的测试结果文件
|
| 174 |
-
返回: dict, task_id -> 完整记录字典
|
| 175 |
-
"""
|
| 176 |
-
if not os.path.exists(OUTPUT_FILE):
|
| 177 |
-
return {}
|
| 178 |
-
|
| 179 |
-
results = {}
|
| 180 |
-
try:
|
| 181 |
-
with open(OUTPUT_FILE, "r", encoding="utf-8") as f:
|
| 182 |
-
for line in f:
|
| 183 |
-
if not line.strip():
|
| 184 |
-
continue
|
| 185 |
-
try:
|
| 186 |
-
data = json.loads(line)
|
| 187 |
-
task_id = data.get("task_id")
|
| 188 |
-
if task_id:
|
| 189 |
-
results[task_id] = data
|
| 190 |
-
except json.JSONDecodeError:
|
| 191 |
-
continue
|
| 192 |
-
print(f"✅ 已加载 {len(results)} 条历史记录")
|
| 193 |
-
except Exception as e:
|
| 194 |
-
print(f"⚠️ 加载历史记录时出错: {e}")
|
| 195 |
-
return {}
|
| 196 |
-
|
| 197 |
-
return results
|
| 198 |
-
|
| 199 |
-
|
| 200 |
-
def update_result_in_file(task_id, new_record):
|
| 201 |
-
"""
|
| 202 |
-
更新 jsonl 文件中指定 task_id 的记录
|
| 203 |
-
使用临时文件方式,确保线程安全
|
| 204 |
-
"""
|
| 205 |
-
if not os.path.exists(OUTPUT_FILE):
|
| 206 |
-
# 如果文件不存在,直接写入
|
| 207 |
-
with file_lock:
|
| 208 |
-
with open(OUTPUT_FILE, "w", encoding="utf-8") as f:
|
| 209 |
-
f.write(json.dumps(new_record, ensure_ascii=False) + "\n")
|
| 210 |
-
return
|
| 211 |
-
|
| 212 |
-
# 读取所有记录,更新指定记录,写回文件
|
| 213 |
-
with file_lock:
|
| 214 |
-
temp_file = OUTPUT_FILE + ".tmp"
|
| 215 |
-
updated = False
|
| 216 |
-
|
| 217 |
-
try:
|
| 218 |
-
with open(OUTPUT_FILE, "r", encoding="utf-8") as f_in, \
|
| 219 |
-
open(temp_file, "w", encoding="utf-8") as f_out:
|
| 220 |
-
for line in f_in:
|
| 221 |
-
if not line.strip():
|
| 222 |
-
continue
|
| 223 |
-
try:
|
| 224 |
-
data = json.loads(line)
|
| 225 |
-
if data.get("task_id") == task_id:
|
| 226 |
-
# 更新这条记录
|
| 227 |
-
f_out.write(json.dumps(new_record, ensure_ascii=False) + "\n")
|
| 228 |
-
updated = True
|
| 229 |
-
else:
|
| 230 |
-
# 保持原记录
|
| 231 |
-
f_out.write(line)
|
| 232 |
-
except json.JSONDecodeError:
|
| 233 |
-
continue
|
| 234 |
-
|
| 235 |
-
# 如果没找到要更新的记录,追加新记录
|
| 236 |
-
if not updated:
|
| 237 |
-
f_out.write(json.dumps(new_record, ensure_ascii=False) + "\n")
|
| 238 |
-
|
| 239 |
-
# 替换原文件
|
| 240 |
-
os.replace(temp_file, OUTPUT_FILE)
|
| 241 |
-
except Exception as e:
|
| 242 |
-
# 如果出错,删除临时文件
|
| 243 |
-
if os.path.exists(temp_file):
|
| 244 |
-
os.remove(temp_file)
|
| 245 |
-
raise e
|
| 246 |
-
|
| 247 |
-
|
| 248 |
-
def upload_file(local_path):
|
| 249 |
-
"""上传文件并返回符合接口要求的 file_id 和 filename"""
|
| 250 |
-
try:
|
| 251 |
-
if not os.path.exists(local_path):
|
| 252 |
-
print(f"❌ 本地文件不存在: {local_path}")
|
| 253 |
-
return None
|
| 254 |
-
|
| 255 |
-
with open(local_path, 'rb') as f:
|
| 256 |
-
files = {'file': f}
|
| 257 |
-
response = requests.post(UPLOAD_URL, headers=HEADERS, files=files, timeout=60)
|
| 258 |
-
response.raise_for_status()
|
| 259 |
-
res_data = response.json()
|
| 260 |
-
|
| 261 |
-
if res_data.get("code") == 0:
|
| 262 |
-
file_info = res_data.get("data", {})
|
| 263 |
-
return {
|
| 264 |
-
"file_id": file_info.get("file_id"),
|
| 265 |
-
"filename": file_info.get("filename")
|
| 266 |
-
}
|
| 267 |
-
else:
|
| 268 |
-
print(f"❌ 上传接口返回错误: {res_data.get('msg')}")
|
| 269 |
-
except Exception as e:
|
| 270 |
-
print(f"❌ 文件上传异常 ({os.path.basename(local_path)}): {e}")
|
| 271 |
-
return None
|
| 272 |
-
|
| 273 |
-
|
| 274 |
-
def extract_answer(text):
|
| 275 |
-
"""从文本中提取答案"""
|
| 276 |
-
if not text:
|
| 277 |
-
return ""
|
| 278 |
-
pattern = r"(?si)<\s*answer\s*>\s*(.*?)\s*</\s*answer\s*>"
|
| 279 |
-
match = re.search(pattern, text)
|
| 280 |
-
if match:
|
| 281 |
-
ans = match.group(1).strip()
|
| 282 |
-
return re.sub(r'^["\']|["\']$', '', ans)
|
| 283 |
-
backup_pattern = r"(?i)answer\s*is[::]\s*(.*)"
|
| 284 |
-
backup_match = re.search(backup_pattern, text)
|
| 285 |
-
if backup_match:
|
| 286 |
-
return backup_match.group(1).strip().rstrip('.')
|
| 287 |
-
lines = [l.strip() for l in text.strip().split('\n') if l.strip()]
|
| 288 |
-
return lines[-1] if lines else text.strip()
|
| 289 |
-
|
| 290 |
-
|
| 291 |
-
def call_my_agent_safe(question, attachments=None, task_id=None):
|
| 292 |
-
"""
|
| 293 |
-
发送对话请求,包含附件数组
|
| 294 |
-
|
| 295 |
-
Args:
|
| 296 |
-
question: 问题内容
|
| 297 |
-
attachments: 附件列表
|
| 298 |
-
task_id: 任务ID,用于确保会话隔离
|
| 299 |
-
|
| 300 |
-
Returns:
|
| 301 |
-
tuple: (parsed_answer, session_id, raw_content)
|
| 302 |
-
"""
|
| 303 |
-
guided_prompt = (
|
| 304 |
-
f"{question}\n\n Important Requirement: \nprovide the final answer (the answer only, without explanation) inside the tags in the following format: <answer>your answer</answer>"
|
| 305 |
-
)
|
| 306 |
-
|
| 307 |
-
payload = {
|
| 308 |
-
"message": guided_prompt,
|
| 309 |
-
"streaming": False,
|
| 310 |
-
"attachments": attachments if attachments else [],
|
| 311 |
-
"recycle_sandbox": True,
|
| 312 |
-
# 明确指定创建新会话,避免会话内容混乱
|
| 313 |
-
# 如果 API 支持 session_id 参数,设置为 null 表示创建新会话
|
| 314 |
-
# 如果不支持,则不传递 session_id 参数(当前做法)
|
| 315 |
-
}
|
| 316 |
-
|
| 317 |
-
# 如果 API 支持,可以尝试以下方式之一来确保创建新会话:
|
| 318 |
-
# 1. payload["session_id"] = None # 明确创建新会话
|
| 319 |
-
# 2. payload["new_session"] = True # 如果 API 支持此参数
|
| 320 |
-
# 3. 在请求头中添加唯一标识
|
| 321 |
-
|
| 322 |
-
if task_id:
|
| 323 |
-
# 添加 task_id 作为请求标识,帮助后端区分不同请求,确保会话隔离
|
| 324 |
-
payload["task_id"] = task_id
|
| 325 |
-
|
| 326 |
-
# 在请求头中添加唯一标识,进一步确保请求隔离
|
| 327 |
-
# 如果后端支持,可以通过 X-Request-ID 或类似头部来区分请求
|
| 328 |
-
request_headers = HEADERS.copy()
|
| 329 |
-
if task_id:
|
| 330 |
-
# 添加 task_id 到请求头,帮助后端识别和隔离不同请求
|
| 331 |
-
request_headers["X-Task-ID"] = task_id
|
| 332 |
-
|
| 333 |
-
try:
|
| 334 |
-
response = requests.post(CHAT_URL, headers=request_headers, json=payload, timeout=(30, REQUEST_TIMEOUT))
|
| 335 |
-
response.raise_for_status()
|
| 336 |
-
res_data = response.json()
|
| 337 |
-
raw_content = (res_data.get("answer") or res_data.get("content") or res_data.get("response") or "").strip()
|
| 338 |
-
session_id = res_data.get("session_id")
|
| 339 |
-
parsed_answer = extract_answer(raw_content)
|
| 340 |
-
return parsed_answer, session_id, raw_content
|
| 341 |
-
except Exception as e:
|
| 342 |
-
error_traceback = traceback.format_exc()
|
| 343 |
-
return f"ERROR: {str(e)}", session_id, error_traceback
|
| 344 |
-
|
| 345 |
-
|
| 346 |
-
def process_item(item, existing_results, mode):
|
| 347 |
-
"""
|
| 348 |
-
处理单条数据:上传文件 -> 发起对话 -> 记录结果
|
| 349 |
-
hybrid + validation 模式下:如果记录已存在且 is_correct 为 true,则跳过 agent 调用,只刷新字段顺序
|
| 350 |
-
其他情况:所有记录都重新执行并刷新,确保字段顺序一致
|
| 351 |
-
(test 数据集没有 is_correct 字段,无法判断是否正确,所以总是重新执行)
|
| 352 |
-
|
| 353 |
-
Args:
|
| 354 |
-
item: 数据集项
|
| 355 |
-
existing_results: 已有结果字典
|
| 356 |
-
mode: 执行模式 ("full"、"incremental" 或 "hybrid")
|
| 357 |
-
"""
|
| 358 |
-
task_id = item['task_id']
|
| 359 |
-
level = item.get('Level', 'Unknown')
|
| 360 |
-
question = item['Question']
|
| 361 |
-
file_name = item.get('file_name', "")
|
| 362 |
-
|
| 363 |
-
# hybrid + validation 模式下:如果记录已存在且成功,只刷新字段顺序,不调用 agent
|
| 364 |
-
# 只有 validation 数据集有 is_correct 字段,可以判断是否正确
|
| 365 |
-
if mode == "hybrid" and SPLIT_TYPE == "validation" and task_id in existing_results:
|
| 366 |
-
existing_record = existing_results[task_id]
|
| 367 |
-
if existing_record.get("is_correct", False):
|
| 368 |
-
# 已成功,只刷新字段顺序,不调用 agent
|
| 369 |
-
# 使用当前的 file_name 更新 attachment_name,确保数据一致性
|
| 370 |
-
current_has_file = bool(file_name)
|
| 371 |
-
current_attachment_name = file_name if file_name else None
|
| 372 |
-
record = build_ordered_record(
|
| 373 |
-
task_id=task_id,
|
| 374 |
-
question=existing_record.get("question", question),
|
| 375 |
-
level=existing_record.get("level", level),
|
| 376 |
-
agent_answer=existing_record.get("agent_answer", ""),
|
| 377 |
-
duration=existing_record.get("duration", 0),
|
| 378 |
-
has_file=current_has_file,
|
| 379 |
-
session_id=existing_record.get("session_id"),
|
| 380 |
-
attachment_name=current_attachment_name,
|
| 381 |
-
ground_truth=existing_record.get("ground_truth", ""),
|
| 382 |
-
is_correct=True
|
| 383 |
-
)
|
| 384 |
-
# 更新已有记录(刷新字段顺序)
|
| 385 |
-
update_result_in_file(task_id, record)
|
| 386 |
-
return task_id, True, "refreshed"
|
| 387 |
-
|
| 388 |
-
# 需要调用 agent 的情况(新记录、错误记录、或非 hybrid 模式)
|
| 389 |
-
attachments = []
|
| 390 |
-
|
| 391 |
-
# 1. 如果有文件,先执行上传
|
| 392 |
-
if file_name:
|
| 393 |
-
# 根据 split 类型选择不同的文件夹
|
| 394 |
-
folder = "validation" if SPLIT_TYPE == "validation" else "test"
|
| 395 |
-
local_file_path = os.path.abspath(os.path.join(DATA_PATH, "2023", folder, file_name))
|
| 396 |
-
upload_data = upload_file(local_file_path)
|
| 397 |
-
if upload_data:
|
| 398 |
-
attachments.append(upload_data)
|
| 399 |
-
|
| 400 |
-
# 2. 调用 Agent(传递 task_id 确保会话隔离)
|
| 401 |
-
start_time = time.time()
|
| 402 |
-
agent_answer, session_id, _ = call_my_agent_safe(question, attachments, task_id=task_id)
|
| 403 |
-
duration = time.time() - start_time
|
| 404 |
-
|
| 405 |
-
# 3. 构建记录(使用固定顺序)
|
| 406 |
-
if SPLIT_TYPE == "validation":
|
| 407 |
-
# validation 数据集:添加标准答案和正确性判断
|
| 408 |
-
ground_truth = str(item['Final answer']).strip()
|
| 409 |
-
clean_agent = str(agent_answer).lower().rstrip('.')
|
| 410 |
-
clean_gt = ground_truth.lower().rstrip('.')
|
| 411 |
-
is_correct = (clean_agent == clean_gt)
|
| 412 |
-
record = build_ordered_record(
|
| 413 |
-
task_id=task_id,
|
| 414 |
-
question=question,
|
| 415 |
-
level=level,
|
| 416 |
-
duration=round(duration, 2),
|
| 417 |
-
has_file=bool(file_name),
|
| 418 |
-
session_id=session_id,
|
| 419 |
-
attachment_name=file_name if file_name else None,
|
| 420 |
-
agent_answer=agent_answer,
|
| 421 |
-
ground_truth=ground_truth,
|
| 422 |
-
is_correct=is_correct
|
| 423 |
-
)
|
| 424 |
-
result_correct = is_correct
|
| 425 |
-
else: # test 数据集:没有标准答案
|
| 426 |
-
record = build_ordered_record(
|
| 427 |
-
task_id=task_id,
|
| 428 |
-
question=question,
|
| 429 |
-
level=level,
|
| 430 |
-
duration=round(duration, 2),
|
| 431 |
-
has_file=bool(file_name),
|
| 432 |
-
session_id=session_id,
|
| 433 |
-
attachment_name=file_name if file_name else None,
|
| 434 |
-
agent_answer=agent_answer
|
| 435 |
-
)
|
| 436 |
-
result_correct = None
|
| 437 |
-
|
| 438 |
-
# 4. 更新或追加记录
|
| 439 |
-
if task_id in existing_results:
|
| 440 |
-
# 更新已有记录
|
| 441 |
-
update_result_in_file(task_id, record)
|
| 442 |
-
return task_id, result_correct, "updated"
|
| 443 |
-
else:
|
| 444 |
-
# 追加新记录
|
| 445 |
-
with file_lock:
|
| 446 |
-
with open(OUTPUT_FILE, "a", encoding="utf-8") as f:
|
| 447 |
-
f.write(json.dumps(record, ensure_ascii=False) + "\n")
|
| 448 |
-
return task_id, result_correct, "new"
|
| 449 |
-
|
| 450 |
-
|
| 451 |
-
def generate_submission():
|
| 452 |
-
"""
|
| 453 |
-
生成官网提交格式文件
|
| 454 |
-
GAIA 提交格式要求:
|
| 455 |
-
- 文件格式:JSONL(每行一个 JSON 对象)
|
| 456 |
-
- 必需字段:task_id, model_answer
|
| 457 |
-
- 编码:UTF-8
|
| 458 |
-
- test 数据集需要包含所有 285 个用例的答案
|
| 459 |
-
"""
|
| 460 |
-
if not os.path.exists(OUTPUT_FILE):
|
| 461 |
-
print(f"⚠️ 警告:结果文件 {OUTPUT_FILE} 不存在,无法生成提交文件")
|
| 462 |
-
return
|
| 463 |
-
|
| 464 |
-
# 读取所有结果并按 task_id 排序(确保顺序一致)
|
| 465 |
-
results = []
|
| 466 |
-
with open(OUTPUT_FILE, "r", encoding="utf-8") as f:
|
| 467 |
-
for line in f:
|
| 468 |
-
if not line.strip():
|
| 469 |
-
continue
|
| 470 |
-
try:
|
| 471 |
-
data = json.loads(line)
|
| 472 |
-
if "task_id" in data and "agent_answer" in data:
|
| 473 |
-
results.append(data)
|
| 474 |
-
except json.JSONDecodeError:
|
| 475 |
-
continue
|
| 476 |
-
|
| 477 |
-
if not results:
|
| 478 |
-
print(f"⚠️ 警告:结果文件 {OUTPUT_FILE} 中没有有效数据")
|
| 479 |
-
return
|
| 480 |
-
|
| 481 |
-
# 按 task_id 排序,确保顺序一致
|
| 482 |
-
results.sort(key=lambda x: x.get("task_id", ""))
|
| 483 |
-
|
| 484 |
-
# 生成提交文件
|
| 485 |
-
with open(SUBMISSION_FILE, "w", encoding="utf-8") as f_out:
|
| 486 |
-
for data in results:
|
| 487 |
-
submission_data = {
|
| 488 |
-
"task_id": data["task_id"],
|
| 489 |
-
"model_answer": str(data["agent_answer"])
|
| 490 |
-
}
|
| 491 |
-
f_out.write(json.dumps(submission_data, ensure_ascii=False) + "\n")
|
| 492 |
-
|
| 493 |
-
print(f"✅ 提交文件已生成: {SUBMISSION_FILE} (共 {len(results)} 条记录)")
|
| 494 |
-
|
| 495 |
-
# test 数据集验证:检查是否包含所有用例
|
| 496 |
-
if SPLIT_TYPE == "test":
|
| 497 |
-
expected_count = 285
|
| 498 |
-
if len(results) < expected_count:
|
| 499 |
-
print(f"⚠️ 警告:test 数据集应该有 {expected_count} 个用例,当前只有 {len(results)} 个")
|
| 500 |
-
else:
|
| 501 |
-
print(f"✅ test 数据集已包含 {len(results)} 个用例,符合提交要求")
|
| 502 |
-
|
| 503 |
-
|
| 504 |
-
def get_current_accuracy():
|
| 505 |
-
"""
|
| 506 |
-
获取当前的整体正确率(仅 validation 数据集)
|
| 507 |
-
|
| 508 |
-
Returns:
|
| 509 |
-
float or None: 正确率(百分比),如果不是 validation 数据集或文件不存在则返回 None
|
| 510 |
-
"""
|
| 511 |
-
# test 数据集没有标准答案,无法计算正确率
|
| 512 |
-
if SPLIT_TYPE != "validation":
|
| 513 |
-
return None
|
| 514 |
-
|
| 515 |
-
if not os.path.exists(OUTPUT_FILE):
|
| 516 |
-
return None
|
| 517 |
-
|
| 518 |
-
try:
|
| 519 |
-
results = []
|
| 520 |
-
with open(OUTPUT_FILE, "r", encoding="utf-8") as f:
|
| 521 |
-
for line in f:
|
| 522 |
-
if not line.strip():
|
| 523 |
-
continue
|
| 524 |
-
try:
|
| 525 |
-
data = json.loads(line)
|
| 526 |
-
results.append(data)
|
| 527 |
-
except json.JSONDecodeError:
|
| 528 |
-
continue
|
| 529 |
-
|
| 530 |
-
if not results:
|
| 531 |
-
return None
|
| 532 |
-
|
| 533 |
-
total = len(results)
|
| 534 |
-
correct = sum(1 for r in results if r.get("is_correct", False))
|
| 535 |
-
accuracy = (correct / total * 100) if total > 0 else 0.0
|
| 536 |
-
return accuracy
|
| 537 |
-
except Exception:
|
| 538 |
-
return None
|
| 539 |
-
|
| 540 |
-
|
| 541 |
-
def generate_report():
|
| 542 |
-
"""生成统计成绩单(仅 validation 数据集有标准答案,才生成成绩单)"""
|
| 543 |
-
# test 数据集没有标准答案,不生成成绩单
|
| 544 |
-
if SPLIT_TYPE != "validation":
|
| 545 |
-
return
|
| 546 |
-
|
| 547 |
-
if not os.path.exists(OUTPUT_FILE):
|
| 548 |
-
return
|
| 549 |
-
|
| 550 |
-
results = [json.loads(line) for line in open(OUTPUT_FILE, "r", encoding="utf-8")]
|
| 551 |
-
df = pd.DataFrame(results)
|
| 552 |
-
total = len(df)
|
| 553 |
-
acc = (df['is_correct'].sum() / total) * 100
|
| 554 |
-
|
| 555 |
-
print("\n" + "=" * 50)
|
| 556 |
-
print(f"测试完成! 总数: {total} | 总准确率: {acc:.2f}%")
|
| 557 |
-
print("=" * 50)
|
| 558 |
-
|
| 559 |
-
|
| 560 |
-
def run_test_concurrent(num_questions=200, mode="hybrid", split="validation", threads=MAX_CONCURRENT, target_task_id=None):
|
| 561 |
-
"""
|
| 562 |
-
测试主函数
|
| 563 |
-
|
| 564 |
-
Args:
|
| 565 |
-
num_questions: 要执行的用例数量
|
| 566 |
-
mode: 执行模式
|
| 567 |
-
- "full": 全量模式,不管文件是否存在,都删除重新开始
|
| 568 |
-
- "incremental": 增量模式,如果文件存在则增量,不存在则全量执行
|
| 569 |
-
- "hybrid": 混合模式,第一次时全量(文件不存在),后面就增量(文件存在)
|
| 570 |
-
- "error": 错误模式,只重新执行 agent_answer 包含 ERROR 的记录
|
| 571 |
-
split: 数据集类型,"validation" 或 "test"
|
| 572 |
-
threads: 并发线程数
|
| 573 |
-
target_task_id: 可选,指定要运行的 task_id,如果指定则只运行该用例
|
| 574 |
-
"""
|
| 575 |
-
global dataset, OUTPUT_FILE, SUBMISSION_FILE, SPLIT_TYPE
|
| 576 |
-
|
| 577 |
-
# 设置全局变量
|
| 578 |
-
SPLIT_TYPE = split
|
| 579 |
-
|
| 580 |
-
# 根据 split 类型设置输出文件名
|
| 581 |
-
if split == "validation":
|
| 582 |
-
OUTPUT_FILE = "validation_results.jsonl"
|
| 583 |
-
SUBMISSION_FILE = "validation_submission.jsonl"
|
| 584 |
-
print("📥 正在检查 GAIA 验证集数据...")
|
| 585 |
-
else: # test
|
| 586 |
-
OUTPUT_FILE = "test_results.jsonl"
|
| 587 |
-
SUBMISSION_FILE = "test_submission.jsonl"
|
| 588 |
-
print("📥 正在检查 GAIA 测试集数据...")
|
| 589 |
-
|
| 590 |
-
# 1. 检查并下载完整数据集文件(如果需要,包含 validation 和 test 的所有文件)
|
| 591 |
-
print("\n【步骤 1/2】检查数据集文件...")
|
| 592 |
-
check_and_download_dataset_files()
|
| 593 |
-
|
| 594 |
-
# 2. 加载数据集元数据(如果首次下载会显示下载进度)
|
| 595 |
-
print(f"\n【步骤 2/2】加载数据集元数据...")
|
| 596 |
-
print(f" 数据集: gaia-benchmark/GAIA (2023_all, split={split})")
|
| 597 |
-
print(" 提示: 如果是首次下载,请耐心等待,下载进度会显示在下方...")
|
| 598 |
-
print(" 如果已下载过,会直接从缓存加载,速度较快")
|
| 599 |
-
start_time = time.time()
|
| 600 |
-
dataset = load_dataset("gaia-benchmark/GAIA", "2023_all", split=split)
|
| 601 |
-
load_duration = time.time() - start_time
|
| 602 |
-
print(f"✅ 数据集元数据加载完成!共 {len(dataset)} 条记录,耗时 {load_duration:.2f} 秒\n")
|
| 603 |
-
|
| 604 |
-
# 1. 根据模式处理已有结果
|
| 605 |
-
file_exists = os.path.exists(OUTPUT_FILE)
|
| 606 |
-
|
| 607 |
-
if mode == "full":
|
| 608 |
-
# 全量模式:删除旧文件,从头开始
|
| 609 |
-
if file_exists:
|
| 610 |
-
os.remove(OUTPUT_FILE)
|
| 611 |
-
print("🔄 全量模式:已删除旧结果文件,从头开始执行")
|
| 612 |
-
existing_results = {}
|
| 613 |
-
elif mode == "incremental":
|
| 614 |
-
# 增量模式:如果文件存在则增量,不存在则全量执行
|
| 615 |
-
if file_exists:
|
| 616 |
-
existing_results = load_existing_results()
|
| 617 |
-
print(f"📋 增量模式:已加载 {len(existing_results)} 条历史记录")
|
| 618 |
-
else:
|
| 619 |
-
existing_results = {}
|
| 620 |
-
print("📋 增量模式:未找到历史记录,将全量执行")
|
| 621 |
-
elif mode == "error":
|
| 622 |
-
# 错误模式:只重新执行 agent_answer 包含 ERROR 的记录
|
| 623 |
-
if file_exists:
|
| 624 |
-
existing_results = load_existing_results()
|
| 625 |
-
print(f"📋 错误模式:已加载 {len(existing_results)} 条历史记录,将重新执行包含 ERROR 的记录")
|
| 626 |
-
else:
|
| 627 |
-
existing_results = {}
|
| 628 |
-
print("📋 错误模式:未找到历史记录,无法执行错误重试")
|
| 629 |
-
else: # hybrid
|
| 630 |
-
# 混合模式:第一次时全量(文件不存在),后面就增量(文件存在)
|
| 631 |
-
if file_exists:
|
| 632 |
-
existing_results = load_existing_results()
|
| 633 |
-
print(f"📋 混合模式:检测到已有文件,进入增量模式(已加载 {len(existing_results)} 条历史记录)")
|
| 634 |
-
else:
|
| 635 |
-
existing_results = {}
|
| 636 |
-
print("📋 混合模式:首次执行,进入全量模式")
|
| 637 |
-
|
| 638 |
-
# 2. 筛选需要执行的用例
|
| 639 |
-
if target_task_id:
|
| 640 |
-
# 如果指定了 task_id,只运行该用例
|
| 641 |
-
print(f"🎯 指定运行 task_id: {target_task_id}")
|
| 642 |
-
tasks_to_run = []
|
| 643 |
-
found = False
|
| 644 |
-
for item in dataset:
|
| 645 |
-
if item['task_id'] == target_task_id:
|
| 646 |
-
tasks_to_run = [item]
|
| 647 |
-
found = True
|
| 648 |
-
break
|
| 649 |
-
if not found:
|
| 650 |
-
print(f"❌ 错误: 在 {split} 数据集中未找到 task_id: {target_task_id}")
|
| 651 |
-
return
|
| 652 |
-
num_to_run = 1
|
| 653 |
-
else:
|
| 654 |
-
# 正常模式,根据 num_questions 筛选
|
| 655 |
-
num_to_run = min(num_questions, len(dataset))
|
| 656 |
-
tasks_to_run = dataset.select(range(num_to_run))
|
| 657 |
-
|
| 658 |
-
# 统计需要执行的用例
|
| 659 |
-
tasks_to_execute = []
|
| 660 |
-
refresh_count = 0 # hybrid 模式下只刷新字段顺序的记录数
|
| 661 |
-
update_count = 0 # 需要重新调用 agent 的记录数
|
| 662 |
-
new_count = 0 # 新记录数
|
| 663 |
-
error_count = 0 # error 模式下包含 ERROR 的记录数
|
| 664 |
-
|
| 665 |
-
for item in tasks_to_run:
|
| 666 |
-
task_id = item['task_id']
|
| 667 |
-
|
| 668 |
-
if mode == "error":
|
| 669 |
-
# error 模式:只重新执行 agent_answer 包含 ERROR 的记录
|
| 670 |
-
if task_id in existing_results:
|
| 671 |
-
agent_answer = existing_results[task_id].get("agent_answer", "")
|
| 672 |
-
if agent_answer and "ERROR" in str(agent_answer):
|
| 673 |
-
error_count += 1
|
| 674 |
-
tasks_to_execute.append(item)
|
| 675 |
-
# 如果记录不存在或 agent_answer 不包含 ERROR,则跳过
|
| 676 |
-
else:
|
| 677 |
-
# 其他模式:正常处理
|
| 678 |
-
if task_id in existing_results:
|
| 679 |
-
# hybrid + validation 模式下:如果已成功,只刷新字段顺序
|
| 680 |
-
# test 数据集没有 is_correct 字段,无法判断是否正确,所以总是重新执行
|
| 681 |
-
if mode == "hybrid" and split == "validation":
|
| 682 |
-
if existing_results[task_id].get("is_correct", False):
|
| 683 |
-
refresh_count += 1
|
| 684 |
-
else:
|
| 685 |
-
update_count += 1
|
| 686 |
-
else:
|
| 687 |
-
# 非 hybrid 模式,或 test 数据集:所有已有记录都需要重新执行
|
| 688 |
-
update_count += 1
|
| 689 |
-
else:
|
| 690 |
-
new_count += 1
|
| 691 |
-
tasks_to_execute.append(item)
|
| 692 |
-
|
| 693 |
-
total_to_execute = len(tasks_to_execute)
|
| 694 |
-
|
| 695 |
-
print(f"\n📊 统计信息:")
|
| 696 |
-
print(f" 数据集: {split}")
|
| 697 |
-
print(f" 执行模式: {mode}")
|
| 698 |
-
if target_task_id:
|
| 699 |
-
print(f" 指定 task_id: {target_task_id}")
|
| 700 |
-
print(f" 总用例数: {num_to_run}")
|
| 701 |
-
if existing_results:
|
| 702 |
-
if mode == "error":
|
| 703 |
-
print(f" 需要执行: {total_to_execute} (包含 ERROR 的记录: {error_count})")
|
| 704 |
-
elif mode == "hybrid":
|
| 705 |
-
print(
|
| 706 |
-
f" 需要执行: {total_to_execute} (新用例: {new_count}, 刷新字段顺序: {refresh_count}, 重新测试: {update_count})")
|
| 707 |
-
else:
|
| 708 |
-
print(
|
| 709 |
-
f" 需要执行: {total_to_execute} (新用例: {new_count}, 刷新已有记录: {refresh_count + update_count})")
|
| 710 |
-
else:
|
| 711 |
-
print(f" 需要执行: {total_to_execute} (全量执行)")
|
| 712 |
-
print(f"🚀 开始测试 | 并发数: {threads} | 待执行: {total_to_execute}")
|
| 713 |
-
|
| 714 |
-
if total_to_execute == 0:
|
| 715 |
-
if mode == "error":
|
| 716 |
-
print("✅ 没有包含 ERROR 的记录,无需执行")
|
| 717 |
-
elif split == "validation":
|
| 718 |
-
print("✅ 所有用例已完成且正确,无需执行")
|
| 719 |
-
else:
|
| 720 |
-
print("✅ 所有用例已完成,无需执行")
|
| 721 |
-
generate_report()
|
| 722 |
-
generate_submission()
|
| 723 |
-
return
|
| 724 |
-
|
| 725 |
-
# 3. 并发执行
|
| 726 |
-
with ThreadPoolExecutor(max_workers=threads) as executor:
|
| 727 |
-
future_to_item = {executor.submit(process_item, item, existing_results, mode): item for item in
|
| 728 |
-
tasks_to_execute}
|
| 729 |
-
|
| 730 |
-
done = 0
|
| 731 |
-
for future in as_completed(future_to_item):
|
| 732 |
-
done += 1
|
| 733 |
-
item = future_to_item[future]
|
| 734 |
-
tid = item['task_id']
|
| 735 |
-
try:
|
| 736 |
-
_, is_ok, status = future.result()
|
| 737 |
-
if status == "refreshed":
|
| 738 |
-
status_icon = "🔄"
|
| 739 |
-
elif split == "validation":
|
| 740 |
-
status_icon = "✅" if is_ok else "❌"
|
| 741 |
-
else: # test
|
| 742 |
-
status_icon = "✅"
|
| 743 |
-
|
| 744 |
-
# 计算并显示当前整体正确率
|
| 745 |
-
accuracy_info = ""
|
| 746 |
-
if split == "validation":
|
| 747 |
-
current_accuracy = get_current_accuracy()
|
| 748 |
-
if current_accuracy is not None:
|
| 749 |
-
accuracy_info = f" | 当前正确率: {current_accuracy:.2f}%"
|
| 750 |
-
|
| 751 |
-
print(f"[{done}/{total_to_execute}] ID: {tid} | 状态: {status_icon} ({status}){accuracy_info}")
|
| 752 |
-
except Exception as e:
|
| 753 |
-
error_traceback = traceback.format_exc()
|
| 754 |
-
print(f"[{done}/{total_to_execute}] ID: {tid} 运行异常: {e}")
|
| 755 |
-
print(f"异常堆栈:\n{error_traceback}")
|
| 756 |
-
|
| 757 |
-
# 4. 生成报表
|
| 758 |
-
generate_report()
|
| 759 |
-
generate_submission()
|
| 760 |
-
|
| 761 |
-
|
| 762 |
-
def print_help():
|
| 763 |
-
"""打印详细的帮助信息"""
|
| 764 |
-
print("=" * 70)
|
| 765 |
-
print("GAIA 测试脚本 - 参数说明")
|
| 766 |
-
print("=" * 70)
|
| 767 |
-
print()
|
| 768 |
-
print("用法:")
|
| 769 |
-
print(" python gaia_test.py [参数]")
|
| 770 |
-
print()
|
| 771 |
-
print("参数说明:")
|
| 772 |
-
print()
|
| 773 |
-
print(" --split <类型>")
|
| 774 |
-
print(" 数据集类型")
|
| 775 |
-
print(" 可选值: validation, test")
|
| 776 |
-
print(" 默认值: validation")
|
| 777 |
-
print(" 说明:")
|
| 778 |
-
print(" - validation: 验证集,有标准答案,可以计算正确率")
|
| 779 |
-
print(" - test: 测试集,无标准答案,用于最终提交")
|
| 780 |
-
print()
|
| 781 |
-
print(" --mode <模式>")
|
| 782 |
-
print(" 执行模式")
|
| 783 |
-
print(" 可选值: full, incremental, hybrid, error")
|
| 784 |
-
print(" 默认值: hybrid")
|
| 785 |
-
print(" 说明:")
|
| 786 |
-
print(" - full: 全量模式,删除旧结果文件,从头开始执行")
|
| 787 |
-
print(" - incremental: 增量模式,如果文件存在则增量,不存在则全量执行")
|
| 788 |
-
print(" - hybrid: 混合模式(推荐),首次全量,后续增量")
|
| 789 |
-
print(" 在 hybrid 模式下,validation 数据集中已正确的记录")
|
| 790 |
-
print(" 只刷新字段顺序,不重新调用 agent")
|
| 791 |
-
print(" - error: 错误模式,只重新执行 agent_answer 包含 ERROR 的记录")
|
| 792 |
-
print()
|
| 793 |
-
print(" --num <数量>")
|
| 794 |
-
print(" 要执行的用例数量")
|
| 795 |
-
print(" 类型: 整数")
|
| 796 |
-
print(" 默认值: 200")
|
| 797 |
-
print(" 说明:")
|
| 798 |
-
print(" - test 数据集共 285 题,可以设置 --num 285 执行全部")
|
| 799 |
-
print(" - validation 数据集可以根据需要设置数量")
|
| 800 |
-
print()
|
| 801 |
-
print(" --threads <数量>")
|
| 802 |
-
print(" 并发执行的线程数")
|
| 803 |
-
print(" 类型: 整数")
|
| 804 |
-
print(" 默认值: 2")
|
| 805 |
-
print(" 说明:")
|
| 806 |
-
print(" - 根据服务器性能调整,过高可能导致服务器压力过大")
|
| 807 |
-
print(" - 建议范围: 1-4")
|
| 808 |
-
print()
|
| 809 |
-
print(" --task-id <task_id>")
|
| 810 |
-
print(" 指定要运行的 task_id")
|
| 811 |
-
print(" 类型: 字符串")
|
| 812 |
-
print(" 默认值: 无(运行多个用例)")
|
| 813 |
-
print(" 说明:")
|
| 814 |
-
print(" - 如果指定此参数,则只运行该 task_id 对应的用例")
|
| 815 |
-
print(" - 指定此参数时,--num 参数会被忽略")
|
| 816 |
-
print(" - 如果指定的 task_id 不存在,脚本会报错并退出")
|
| 817 |
-
print()
|
| 818 |
-
print(" -h, --help")
|
| 819 |
-
print(" 显示此帮助信息并退出")
|
| 820 |
-
print()
|
| 821 |
-
print("示例:")
|
| 822 |
-
print(" # 使用默认参数(validation 数据集,hybrid 模式,200 题)")
|
| 823 |
-
print(" python gaia_test.py")
|
| 824 |
-
print()
|
| 825 |
-
print(" # 测试 test 数据集,执行全部 285 题")
|
| 826 |
-
print(" python gaia_test.py --split test --num 285")
|
| 827 |
-
print()
|
| 828 |
-
print(" # 使用 error 模式重新执行错误记录")
|
| 829 |
-
print(" python gaia_test.py --mode error")
|
| 830 |
-
print()
|
| 831 |
-
print(" # 使用全量模式,4 个并发线程")
|
| 832 |
-
print(" python gaia_test.py --mode full --threads 4")
|
| 833 |
-
print()
|
| 834 |
-
print(" # 运行指定的 task_id")
|
| 835 |
-
print(" python gaia_test.py --task-id c61d22de-5f6c-4958-a7f6-5e9707bd3466")
|
| 836 |
-
print()
|
| 837 |
-
print("=" * 70)
|
| 838 |
-
print("配置文件:")
|
| 839 |
-
print(" 运行前请确保已正确配置 gaia_test.py 中的以下参数:")
|
| 840 |
-
print(" - BASE_URL: API 服务地址")
|
| 841 |
-
print(" - HEADERS: 认证 Token(必须修改)")
|
| 842 |
-
print(" - DATA_PATH: 数据文件路径")
|
| 843 |
-
print(" - REQUEST_TIMEOUT: 请求超时时间")
|
| 844 |
-
print(" - MAX_CONCURRENT: 最大并发数")
|
| 845 |
-
print("=" * 70)
|
| 846 |
-
|
| 847 |
-
|
| 848 |
-
if __name__ == "__main__":
|
| 849 |
-
import sys
|
| 850 |
-
|
| 851 |
-
# 检查是否有 -h 或 --help 参数
|
| 852 |
-
if "-h" in sys.argv or "--help" in sys.argv:
|
| 853 |
-
print_help()
|
| 854 |
-
sys.exit(0)
|
| 855 |
-
|
| 856 |
-
parser = argparse.ArgumentParser(
|
| 857 |
-
description="GAIA 测试脚本(支持 validation 和 test 数据集,支持全量、增量、混合三种模式)",
|
| 858 |
-
formatter_class=argparse.RawDescriptionHelpFormatter
|
| 859 |
-
)
|
| 860 |
-
parser.add_argument(
|
| 861 |
-
"--split",
|
| 862 |
-
type=str,
|
| 863 |
-
choices=["validation", "test"],
|
| 864 |
-
default="validation",
|
| 865 |
-
help="数据集类型: 'validation' 验证集(有标准答案)、'test' 测试集(无标准答案,默认: validation)"
|
| 866 |
-
)
|
| 867 |
-
parser.add_argument(
|
| 868 |
-
"--mode",
|
| 869 |
-
type=str,
|
| 870 |
-
choices=["full", "incremental", "hybrid", "error"],
|
| 871 |
-
default="hybrid",
|
| 872 |
-
help="执行模式: 'full' 全量模式(删除旧文件重新执行)、'incremental' 增量模式(文件存在则增量,不存在则全量)、'hybrid' 混合模式(首次全量,后续增量,默认)、'error' 错误模式(只重新执行 agent_answer 包含 ERROR 的记录)"
|
| 873 |
-
)
|
| 874 |
-
parser.add_argument(
|
| 875 |
-
"--num",
|
| 876 |
-
type=int,
|
| 877 |
-
default=200,
|
| 878 |
-
help="要执行的用例数量(默认: 200,test 集共 285 题)"
|
| 879 |
-
)
|
| 880 |
-
parser.add_argument(
|
| 881 |
-
"--threads",
|
| 882 |
-
type=int,
|
| 883 |
-
default=MAX_CONCURRENT,
|
| 884 |
-
help="执行的并发数(默认: 2)"
|
| 885 |
-
)
|
| 886 |
-
parser.add_argument(
|
| 887 |
-
"--task-id",
|
| 888 |
-
type=str,
|
| 889 |
-
default=None,
|
| 890 |
-
help="指定要运行的 task_id,如果指定则只运行该用例(忽略 --num 参数)"
|
| 891 |
-
)
|
| 892 |
-
|
| 893 |
-
args = parser.parse_args()
|
| 894 |
-
|
| 895 |
-
run_test_concurrent(num_questions=args.num, mode=args.mode, split=args.split, threads=args.threads, target_task_id=args.task_id)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|