rc2

Author: gasse

README.md

@@ -43,8 +43,6 @@ while not done:
     obs, reward, terminated, truncated, info = env.step(action)
 ```
 
-
-
 ### MiniWoB++ task example
 
 Boilerplate code to run an agent on a miniwob task:
@@ -133,7 +131,3 @@ You can customize your experience by changing the `model_name` to your preferred
 _multiple agent demos on the open web, WorkArena and WebArena_
 
 https://github.com/ServiceNow/BrowserGym/assets/26232819/e0bfc788-cc8e-44f1-b8c3-0d1114108b85
-
-
-
-

core/src/browsergym/core/__init__.py

@@ -1,4 +1,4 @@
-__version__ = "0.1.0rc1"
+__version__ = "0.1.0rc2"
 
 import playwright.sync_api
 

miniwob/requirements.txt

@@ -1 +1 @@
-browsergym-core==0.1.0rc1
+browsergym-core==0.1.0rc2

miniwob/src/browsergym/miniwob/__init__.py

@@ -1,4 +1,4 @@
-__version__ = "0.1.0rc1"
+__version__ = "0.1.0rc2"
 
 from browsergym.core.registration import register_task
 

pyproject.toml

@@ -25,12 +25,12 @@ classifiers = [
     "Topic :: Scientific/Engineering :: Artificial Intelligence",
     "License :: OSI Approved :: Apache Software License",
 ]
-version="0.1.0rc1"
+version="0.1.0rc2"
 dependencies = [
-    "browsergym-core==0.1.0rc1",
-    "browsergym-miniwob==0.1.0rc1",
-    "browsergym-webarena==0.1.0rc1",
-    "browsergym-workarena==0.1.0rc1",
+    "browsergym-core==0.1.0rc2",
+    "browsergym-miniwob==0.1.0rc2",
+    "browsergym-webarena==0.1.0rc2",
+    "browsergym-workarena==0.1.0rc2",
 ]
 
 [tool.setuptools]

webarena/requirements.txt

@@ -1,2 +1,2 @@
-browsergym-core==0.1.0rc1
+browsergym-core==0.1.0rc2
 libwebarena==0.0.2

webarena/src/browsergym/webarena/__init__.py

@@ -1,4 +1,4 @@
-__version__ = "0.1.0rc1"
+__version__ = "0.1.0rc2"
 
 from browsergym.core.registration import register_task
 

webarena/src/browsergym/webarena/task.py

@@ -113,13 +113,14 @@ def setup(self, seed: int, page: playwright.sync_api.Page) -> tuple[str, dict]:
 
         # This note is present in all webarena's agent prompts
         # https://github.com/web-arena-x/webarena/blob/c6475f0e9affe5252a2966e26b8cb4c834a4ae40/agent/prompts/raw/p_cot_id_actree_2s.py#L34
-        goal += f"""
+        if self.with_homepage_hint:
+            goal += f"""
 
 (Note: if you want to visit other websites, check out the homepage at {self.webarena_instance.home_url}. It has a list of websites you can visit. {self.webarena_instance.home_url}/password.html lists all the account name and password for the websites. You can use them to log in to the websites.)
 """
 
         # This note is present in some of webarena's agent prompts
-        if self.with_na:
+        if self.with_na_hint:
             goal += """\
 
 If you believe the task is impossible to complete, provide the answer "N/A".
@@ -178,23 +179,3 @@ def validate(
             return score, True, "", {}
         else:
             return score, False, "", {}
-
-    def get_goal(self) -> str:
-        goal = self.config["intent"]
-
-        if self.with_homepage_hint:
-            goal += f"""
-
-(Note: if you want to visit other websites, check out the homepage at {self.webarena_instance.home_url}. It has a list of websites you can visit. {self.webarena_instance.home_url}/password.html lists all the account name and password for the websites. You can use them to log in to the websites.)
-"""
-        # This note is present in all webarena's agent prompts
-        # https://github.com/web-arena-x/webarena/blob/c6475f0e9affe5252a2966e26b8cb4c834a4ae40/agent/prompts/raw/p_cot_id_actree_2s.py#L34
-
-        # This note is present in some of webarena's agent prompts
-        if self.with_na_hint:
-            goal += """\
-
-If you believe the task is impossible to complete, provide the answer "N/A".
-"""
-
-        return goal

webarena/tests/test_env_general.py

@@ -2,6 +2,7 @@
 import logging
 import os
 import pytest
+import random
 
 from tenacity import retry, stop_after_attempt, retry_if_exception_type
 
@@ -18,14 +19,17 @@
 
 from browsergym.webarena import ALL_WEBARENA_TASK_IDS
 
+rng = random.Random(1)
+task_ids = rng.sample(ALL_WEBARENA_TASK_IDS, 25)
+
 
 @retry(
     stop=stop_after_attempt(5),
     retry=retry_if_exception_type(TimeoutError),
     reraise=True,
     before_sleep=lambda _: logging.info("Retrying due to a TimeoutError..."),
 )
-@pytest.mark.parametrize("task_id", ALL_WEBARENA_TASK_IDS)
+@pytest.mark.parametrize("task_id", task_ids)
 @pytest.mark.slow
 def test_env_generic(task_id):
     env = gym.make(