增加环境变量指向g14部署webarena环境；

map网站使用socks5连接到官网openstreetmap；
模型改成gpt-4o; llm fuzzy math也使用4o；
增加generate_test_data.py脚本

evaluation_harness/evaluators.py

@@ -11,7 +11,7 @@ from typing import Any, Tuple, Union, Optional
 
 from beartype import beartype
 import nltk
-nltk.download('punkt')
+# nltk.download('punkt') # NOTE: you need to download the punkt model first
 from nltk.tokenize import word_tokenize  # type: ignore
 
 from playwright.sync_api import CDPSession, Page

evaluation_harness/helper_functions.py

@@ -158,7 +158,7 @@ def llm_fuzzy_match(pred: str, reference: str, question: str) -> float:
         {"role": "user", "content": message},
     ]
     response = generate_from_openai_chat_completion(
-        model="gpt-4-1106-preview",
+        model="gpt-4o",
         messages=messages,
         temperature=0,
         max_tokens=768,
@@ -193,7 +193,7 @@ def llm_ua_match(pred: str, reference: str, question: str) -> float:
     ]
 
     response = generate_from_openai_chat_completion(
-        model="gpt-4-1106-preview",
+        model="gpt-4o",
         messages=messages,
         temperature=0,
         max_tokens=768,