File size: 10,847 Bytes
2d483c2 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 | """Script to run end-to-end evaluation on the benchmark.
Utils and basic architecture credit to https://github.com/web-arena-x/webarena/blob/main/run.py.
"""
import argparse
import datetime
import json
import logging
import os
import sys
import math
import ast
import time
import backoff
import httpx
import requests
from tqdm import tqdm
from typing import Optional, Dict, Any
from multiprocessing import Pool
from openai import APIConnectionError, APIError, RateLimitError
from types import SimpleNamespace
# Add project root to path for imports
sys.path.insert(0, os.path.join(os.path.dirname(__file__), "../.."))
import lib_run_single
from run_autoglm_v import DesktopEnv, get_unfinished, get_result
from desktop_env.desktop_env import MAX_RETRIES, DesktopEnv as DesktopEnvBase
from mm_agents.autoglm_v import AutoGLMAgent
from openai import OpenAI
logger = logging.getLogger("desktopenv.experiment")
def config() -> argparse.Namespace:
parser = argparse.ArgumentParser(description="Run end-to-end evaluation on the benchmark")
# environment config
parser.add_argument("--path_to_vm", type=str)
parser.add_argument(
"--provider_name",
type=str,
default="docker",
help="Virtualization provider (vmware, docker, aws, azure, gcp, virtualbox)",
)
parser.add_argument("--headless", action="store_true", default=True, help="Run in headless machine")
parser.add_argument("--action_space", type=str, default="autoglm_computer_use", help="Action type")
parser.add_argument(
"--observation_type",
choices=["screenshot", "a11y_tree", "screenshot_a11y_tree", "som"],
default="a11y_tree",
help="Observation type",
)
parser.add_argument("--screen_width", type=int, default=1920)
parser.add_argument("--screen_height", type=int, default=1080)
parser.add_argument("--sleep_after_execution", type=float, default=1.0)
parser.add_argument("--max_steps", type=int, default=30)
# agent config
parser.add_argument("--max_trajectory_length", type=int, default=3)
parser.add_argument("--test_config_base_dir", type=str, default="evaluation_examples/examples")
# lm config
parser.add_argument("--model", type=str, default="autoglm-os")
parser.add_argument("--temperature", type=float, default=0.4)
parser.add_argument("--top_p", type=float, default=0.5)
parser.add_argument("--max_tokens", type=int, default=2048)
parser.add_argument("--stop_token", type=str, default=None)
parser.add_argument("--image_width", type=int, default=1280)
parser.add_argument("--image_height", type=int, default=720)
# example config
parser.add_argument("--domain", type=str, default="all")
parser.add_argument("--test_all_meta_path", type=str, default="evaluation_examples/test_nogdrive.json")
# aws config
parser.add_argument(
"--region", type=str, default="us-east-1", help="AWS region for the VM"
)
parser.add_argument("--client_password", type=str, default="", help="Client password")
# logging related
parser.add_argument("--result_dir", type=str, default="./results")
# parallel number
parser.add_argument("--num_workers", type=int, default=20, help="Number of parallel workers")
args = parser.parse_args()
return args
def _worker_run(task):
domain, example_id, args = task # args 为 argparse.Namespace
logger = logging.getLogger("desktopenv.experiment")
try:
config_file = os.path.join(args.test_config_base_dir, f"{domain}/{example_id}.json")
with open(config_file, "r", encoding="utf-8") as f:
example = json.load(f)
instruction = example["instruction"]
@backoff.on_exception(backoff.constant, (RateLimitError, APIConnectionError), interval=0.1)
def call_llm(messages):
logger.info("Calling LLM...")
# Prepare the request data
data = {
"model": args.model,
"messages": messages,
"max_tokens": args.max_tokens,
"temperature": args.temperature,
"top_p": args.top_p,
"skip_special_tokens": False,
"stream": False,
"include_stop_str_in_output": True,
"stop": ["<|user|>", "<|observation|>", "</answer>"]
}
# Set up proxy
# if os.environ.get('LAN_PROXY', None):
# proxies = {
# "http": os.environ.get('LAN_PROXY'),
# "https": os.environ.get('LAN_PROXY')
# }
# else:
# proxies = None
headers = {
"Content-Type": "application/json",
"Authorization": f"Bearer {os.environ.get('OPENAI_API_KEY', '')}"
}
# Get API base URL from environment or use default
base_url = os.environ.get('OPENAI_BASE_URL', 'https://api.openai.com/v1')
url = f"{base_url}/chat/completions"
response = requests.post(
url,
json=data,
headers=headers,
# proxies=proxies,
timeout=60.0
)
response.raise_for_status()
result = response.json()
logger.info("LLM called successfully.")
return result['choices'][0]['message']['content']
env = DesktopEnv(
provider_name=args.provider_name,
region=args.region,
client_password=args.client_password,
path_to_vm=args.path_to_vm,
action_space=args.action_space,
screen_size=(args.screen_width, args.screen_height),
headless=args.headless,
os_type="Ubuntu",
require_a11y_tree=args.observation_type in ["a11y_tree", "screenshot_a11y_tree", "som"],
)
agent = AutoGLMAgent(
action_space=args.action_space,
observation_type=args.observation_type,
screen_size=(args.screen_width, args.screen_height),
image_size=(args.image_width, args.image_height),
max_trajectory_length=args.max_trajectory_length,
client_password=args.client_password,
gen_func=call_llm,
)
example_result_dir = os.path.join(
args.result_dir,
args.action_space,
args.observation_type,
args.model,
domain,
example_id,
)
os.makedirs(example_result_dir, exist_ok=True)
local_scores = []
try:
lib_run_single.run_single_example_autoglm(
agent,
env,
example,
args.max_steps,
instruction,
args,
example_result_dir,
local_scores,
)
except Exception as e:
logger.error(f"[并发任务异常] {domain}/{example_id}: {e}")
if hasattr(env, "controller") and env.controller is not None:
try:
env.controller.end_recording(os.path.join(example_result_dir, "recording.mp4"))
except Exception:
pass
with open(os.path.join(example_result_dir, "traj.jsonl"), "a") as f:
f.write(json.dumps({"Error": f"Exception in {domain}/{example_id}: {str(e)}"}) + "\n")
finally:
try:
env.close()
except Exception:
pass
score = None
result_path = os.path.join(example_result_dir, "result.txt")
if os.path.exists(result_path):
try:
with open(result_path, "r") as rf:
res = rf.read().strip()
if res.lower() == "true":
score = 1.0
else:
score = float(res)
except Exception:
score = 0.0
else:
score = 0.0
logger.info(f"[Finish] {domain}/{example_id} score={score}")
return (domain, example_id, score)
except Exception as e:
logger = logging.getLogger("desktopenv.experiment")
logger.error(f"[Initializing Fail] {domain}/{example_id}: {e}")
return (domain, example_id, 0.0)
def test_parallel(args: argparse.Namespace, test_all_meta: dict):
tasks = []
for domain in test_all_meta:
for example_id in test_all_meta[domain]:
tasks.append((domain, example_id, args))
if not tasks:
logger.info("No pending tasks")
return
logger.info(f"Starting parallel execution: {args.num_workers} processes, {len(tasks)} tasks total")
results = []
with Pool(processes=args.num_workers) as pool:
for res in tqdm(pool.imap_unordered(_worker_run, tasks), total=len(tasks), desc="Parallel execution"):
results.append(res)
scores = [s for (_, _, s) in results if s is not None]
if scores:
avg = sum(scores) / len(scores)
logger.info(f"Parallel execution completed. Average score: {avg}")
else:
logger.info("No scores obtained.")
if __name__ == "__main__":
####### The complete version of the list of examples #######
os.environ["TOKENIZERS_PARALLELISM"] = "false"
args = config()
if args.client_password == "":
if args.provider_name == "aws":
args.client_password = "osworld-public-evaluation"
else:
args.client_password = "password"
else:
args.client_password = args.client_password
# save args to json in result_dir/action_space/observation_type/model/args.json
path_to_args = os.path.join(
args.result_dir,
args.action_space,
args.observation_type,
args.model,
"args.json",
)
os.makedirs(os.path.dirname(path_to_args), exist_ok=True)
with open(path_to_args, "w", encoding="utf-8") as f:
json.dump(vars(args), f, indent=4)
with open(args.test_all_meta_path, "r", encoding="utf-8") as f:
test_all_meta = json.load(f)
if args.domain != "all":
test_all_meta = {args.domain: test_all_meta[args.domain]}
test_file_list = get_unfinished(
args.action_space,
args.model,
args.observation_type,
args.result_dir,
test_all_meta,
)
left_info = ""
for domain in test_file_list:
left_info += f"{domain}: {len(test_file_list[domain])}\n"
logger.info(f"Left tasks:\n{left_info}")
get_result(
args.action_space,
args.model,
args.observation_type,
args.result_dir,
test_all_meta,
)
test_parallel(args, test_file_list)
|