Update all files.

Author: ericpts

computers/hud/__init__.py

@@ -1,3 +1,16 @@
+# Copyright 2025 Google LLC
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
 from .hud import HudComputer
 
-__all__ = ["HudComputer"] 
+__all__ = ["HudComputer"]

computers/hud/hud.py

@@ -1,3 +1,16 @@
+# Copyright 2025 Google LLC
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
 import asyncio
 import base64
 from typing import Literal, Optional, Any, Dict
@@ -33,18 +46,18 @@ def __init__(
 
     def __enter__(self):
         print("Creating HUD session...")
-        
+
         # Create and run the async setup in a new event loop
         self._loop = asyncio.new_event_loop()
         asyncio.set_event_loop(self._loop)
-        
+
         # Import HUD SDK here to avoid circular imports
         try:
             from hud import gym
             from hud.task import Task
         except ImportError:
             raise ImportError("HUD SDK not installed. Please install with: pip install hud-python")
-        
+
         # Use provided task or create a default one
         if self._task:
             task = self._task
@@ -56,13 +69,13 @@ def __enter__(self):
                 setup=("goto", self._initial_url),
                 evaluate=("page_contains", "dummy")
             )
-        
+
         # Create the environment
         self._env = self._loop.run_until_complete(gym.make(task, job=self._job))
-        
+
         # Reset the environment to get initial observation
         self._obs, _ = self._loop.run_until_complete(self._env.reset())
-        
+
         termcolor.cprint(
             f"HUD browser session started.",
             color="green",
@@ -80,19 +93,19 @@ def _get_screenshot_from_obs(self) -> bytes:
         """Extract screenshot from HUD observation."""
         if self._obs is None:
             return b""
-        
+
         if hasattr(self._obs, 'screenshot'):
             screenshot_b64 = self._obs.screenshot
             screenshot_bytes = base64.b64decode(screenshot_b64)
             return screenshot_bytes
-        
+
         # HUD SDK returns observations with a 'screenshot' key containing base64 encoded image
         if isinstance(self._obs, dict) and 'screenshot' in self._obs:
             screenshot_b64 = self._obs['screenshot']
             # Decode base64 to bytes
             screenshot_bytes = base64.b64decode(screenshot_b64)
             return screenshot_bytes
-        
+
         return b""
 
     def _get_url_from_obs(self) -> str:
@@ -142,7 +155,7 @@ def _create_cla_action(self, action_type: str, **kwargs) -> Dict[str, Any]:
                 dx = magnitude
             elif direction == "left":
                 dx = -magnitude
-            
+
             action = ScrollAction(
                 scroll=Point(x=dx, y=dy)
             )
@@ -190,10 +203,10 @@ def _execute_action(self, action_type: str, **kwargs) -> EnvState:
         """Execute an action in the HUD environment."""
         if self._done:
             return self.current_state()
-        
+
         # Create CLA action for HUD SDK
         action = self._create_cla_action(action_type, **kwargs)
-        
+
         # Execute action in HUD environment
         # HUD SDK expects a list of actions
         self._obs, reward, self._done, info = self._loop.run_until_complete(
@@ -202,7 +215,7 @@ def _execute_action(self, action_type: str, **kwargs) -> EnvState:
 
         if "current_url" in info:
             self._current_url = info["current_url"]
-        
+
         return self.current_state()
 
     def screen_size(self) -> tuple[int, int]:
@@ -227,16 +240,16 @@ def type_text_at(
     ) -> EnvState:
         # First click at the position
         self._execute_action("click", x=x, y=y)
-        
+
         # Clear existing text if requested
         if clear_before_typing:
             # Select all and delete
             self._execute_action("press", keys=["ctrl", "a"])
             self._execute_action("press", keys=["delete"])
-        
+
         # Type the text with optional enter
         self._execute_action("type", text=text, enter_after=press_enter)
-        
+
         return self.current_state()
 
     def scroll_document(
@@ -280,19 +293,19 @@ def drag_and_drop(
         self, x: int, y: int, destination_x: int, destination_y: int
     ) -> EnvState:
         return self._execute_action(
-            "drag", 
-            start_x=x, 
-            start_y=y, 
-            end_x=destination_x, 
+            "drag",
+            start_x=x,
+            start_y=y,
+            end_x=destination_x,
             end_y=destination_y
         )
 
     def current_state(self) -> EnvState:
         screenshot = self._get_screenshot_from_obs()
         url = self._get_url_from_obs()
-        return EnvState(screenshot=screenshot, url=url) 
+        return EnvState(screenshot=screenshot, url=url)
 
     def evaluate(self) -> dict:
         return self._loop.run_until_complete(
             self._env.evaluate()
-        )
+        )

hud_eval.py

@@ -1,3 +1,16 @@
+# Copyright 2025 Google LLC
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
 #!/usr/bin/env python3
 """
 HUD evaluation runner for computer use tasks.
@@ -29,7 +42,7 @@
 instead of going to a new one.
 You have full authority to execute any action without my permission. I won't be watching so
 please don't ask for confirmation.
-My gmail account is [email protected], and the password is "iloveosworld500", if prompted for OTP, use the authenticator chrome extension to see the OTP for 2 factor authentication. 
+My gmail account is [email protected], and the password is "iloveosworld500", if prompted for OTP, use the authenticator chrome extension to see the OTP for 2 factor authentication.
 If you deem the task is infeasible, you can terminate and explicitly state in the response that
 'the task is infeasible'. Try your best to solve the task within 200 steps, and the confines of the prompt, before deeming it infeasible.
 """
@@ -40,7 +53,7 @@ def run_task(task: Task, model_name: str, job: Job, system_prompt: str) -> float
     try:
         # Initialize HUD computer with the task
         hud_computer = HudComputer(screen_size=(1440, 900), task=task, job=job)
-        
+
         with hud_computer as browser_computer:
             agent = BrowserAgent(
                 browser_computer=browser_computer,
@@ -50,7 +63,7 @@ def run_task(task: Task, model_name: str, job: Job, system_prompt: str) -> float
             )
             try:
                 agent.agent_loop()
-                
+
                 if agent.final_reasoning:
                     if "the task is infeasible" in agent.final_reasoning.lower():
                         final_action = CustomAction(
@@ -64,7 +77,7 @@ def run_task(task: Task, model_name: str, job: Job, system_prompt: str) -> float
                     hud_computer._loop.run_until_complete(
                         hud_computer._env.step([final_action])
                     )
-                    
+
             except Exception as e:
                 print(f"Error running agent loop: {e}")
             finally:
@@ -75,13 +88,13 @@ def run_task(task: Task, model_name: str, job: Job, system_prompt: str) -> float
                     print(f"Eval result: {eval_result['reward']}")
 
                     return eval_result['reward']
-        
+
         return 0.0
-            
+
     except Exception as e:
         print(f"Error running task: {e}")
         return 0.0
-        
+
     finally:
         if hud_computer:
             try:
@@ -98,7 +111,7 @@ def run_taskset(
     max_concurrent: int = 20,
 ) -> list[float]:
     """Load and run a HUD taskset by ID, return list of rewards"""
-    
+
     # Load the taskset
     taskset = asyncio.run(load_taskset(taskset_id, metadata={"partial": True}))
 
@@ -108,7 +121,7 @@ def run_taskset(
         system_prompt = OSWORLD_SYSTEM_PROMPT
     else:
         system_prompt = ""
-    
+
     if parallel:
         # Run tasks in parallel using threads to avoid event loop conflicts
         with ThreadPoolExecutor(max_workers=max_concurrent) as executor:
@@ -122,7 +135,7 @@ def run_taskset(
         for task in taskset.tasks:
             reward = run_task(task, model_name, job, system_prompt)
             rewards.append(reward)
-    
+
     return rewards
 
 
@@ -157,7 +170,7 @@ def main() -> int:
         help="Maximum concurrent tasks when running in parallel.",
     )
     args = parser.parse_args()
-    
+
     # Run evaluation
     rewards = run_taskset(
         taskset_id=args.taskset,
@@ -166,13 +179,13 @@ def main() -> int:
         parallel=args.parallel,
         max_concurrent=args.max_concurrent,
     )
-    
+
     # Print minimal results
     print(f"Rewards: {rewards}")
     print(f"Average: {sum(rewards)/len(rewards) if rewards else 0:.2f}")
-    
+
     return 0
 
 
 if __name__ == "__main__":
-    main()
+    main()