Merge branch 'main' of github.com:xlang-ai/OSWorld

desktop_env/evaluators/metrics/utils.py

@@ -459,7 +459,11 @@ def load_xlsx_styles(xlsx_file: Workbook, sheet_name: str, book_name: str, **opt
         for r in fmt.rules:
             active_cells: List[Cell] = []
             if r.type == "expression":
-                condition: Callable[[str], bool] = formula_parser.ast("=" + r.formula[0])[1].compile()
+                try:
+                    condition: Callable[[str], bool] = formula_parser.ast("=" + r.formula[0])[1].compile()
+                except:
+                    logger.exception("Formula parsing error: %s. Skipping.", repr(r.formula[0]))
+                    continue
                 logger.debug("Expression condition: %s", r.formula[0])
 
                 arguments: List[Any] = []
@@ -493,9 +497,15 @@ def load_xlsx_styles(xlsx_file: Workbook, sheet_name: str, book_name: str, **opt
                             if nb_contiguous_nothings>50:
                                 break
                             continue
-                        elif condition(cell_value, *arguments):
+                        else:
-                            logger.debug("Active Cell %s(%s) for %s", repr(cell), str(cell_value), r.formula[0])
+                            try:
-                            active_cells.append(cell)
+                                satisfies_condition: bool = condition(cell_value, *arguments)
+                            except:
+                                logger.exception("Error in formula calculation with cell value %d", repr(cell_value))
+                                satisfies_condition = False
+                            if satisfies_condition:
+                                logger.debug("Active Cell %s(%s) for %s", repr(cell), repr(cell_value), r.formula[0])
+                                active_cells.append(cell)
             else:
                 raise NotImplementedError("Not Implemented Condition Type: {:}".format(r.type))
 

evaluation_examples/examples/libreoffice_calc/01b269ae-2111-4a07-81fd-3fcd711993b0.json

@@ -82,4 +82,4 @@
   "proxy": false,
   "fixed_ip": false,
   "possibility_of_env_change": "low"
-}
+}

evaluation_examples/examples/libreoffice_calc/0326d92d-d218-48a8-9ca1-981cd6d064c7.json

@@ -91,4 +91,4 @@
   "proxy": false,
   "fixed_ip": false,
   "possibility_of_env_change": "low"
-}
+}

evaluation_examples/examples/libreoffice_calc/035f41ba-6653-43ab-aa63-c86d449d62e5.json

@@ -87,4 +87,4 @@
   "proxy": false,
   "fixed_ip": false,
   "possibility_of_env_change": "low"
-}
+}

evaluation_examples/examples/libreoffice_calc/04d9aeaf-7bed-4024-bedb-e10e6f00eb7f.json

@@ -82,4 +82,4 @@
   "proxy": false,
   "fixed_ip": false,
   "possibility_of_env_change": "low"
-}
+}

evaluation_examples/examples/libreoffice_calc/0bf05a7d-b28b-44d2-955a-50b41e24012a.json

@@ -109,4 +109,4 @@
   "proxy": false,
   "fixed_ip": false,
   "possibility_of_env_change": "low"
-}
+}

evaluation_examples/examples/libreoffice_calc/0cecd4f3-74de-457b-ba94-29ad6b5dafb6.json

@@ -95,4 +95,4 @@
   "proxy": false,
   "fixed_ip": false,
   "possibility_of_env_change": "low"
-}
+}

evaluation_examples/examples/libreoffice_calc/1d17d234-e39d-4ed7-b46f-4417922a4e7c.json

@@ -176,4 +176,4 @@
   "proxy": false,
   "fixed_ip": false,
   "possibility_of_env_change": "low"
-}
+}

evaluation_examples/examples/libreoffice_calc/26a8440e-c166-4c50-aef4-bfb77314b46b.json

@@ -82,4 +82,4 @@
   "proxy": false,
   "fixed_ip": false,
   "possibility_of_env_change": "low"
-}
+}

evaluation_examples/examples/libreoffice_calc/2bd59342-0664-4ccb-ba87-79379096cc08.json

@@ -32,4 +32,4 @@
   "proxy": false,
   "fixed_ip": false,
   "possibility_of_env_change": "low"
-}
+}

evaluation_examples/examples/libreoffice_calc/347ef137-7eeb-4c80-a3bb-0951f26a8aff.json

@@ -91,4 +91,4 @@
   "proxy": false,
   "fixed_ip": false,
   "possibility_of_env_change": "low"
-}
+}

evaluation_examples/examples/libreoffice_calc/535364ea-05bd-46ea-9937-9f55c68507e8.json

@@ -88,4 +88,4 @@
   "proxy": false,
   "fixed_ip": false,
   "possibility_of_env_change": "low"
-}
+}

evaluation_examples/examples/libreoffice_calc/aa3a8974-2e85-438b-b29e-a64df44deb4b.json

@@ -61,4 +61,4 @@
   "proxy": false,
   "fixed_ip": false,
   "possibility_of_env_change": "low"
-}
+}

evaluation_examples/examples/libreoffice_calc/eb03d19a-b88d-4de4-8a64-ca0ac66f426b.json

@@ -82,4 +82,4 @@
   "proxy": false,
   "fixed_ip": false,
   "possibility_of_env_change": "low"
-}
+}

evaluation_examples/examples/libreoffice_calc/ecb0df7a-4e8d-4a03-b162-053391d3afaf.json

@@ -109,4 +109,4 @@
   "proxy": false,
   "fixed_ip": false,
   "possibility_of_env_change": "low"
-}
+}

evaluation_examples/examples/libreoffice_calc/f9584479-3d0d-4c79-affa-9ad7afdd8850.json

@@ -82,4 +82,4 @@
   "proxy": false,
   "fixed_ip": false,
   "possibility_of_env_change": "low"
-}
+}

evaluation_examples/examples/libreoffice_impress/0a211154-fda0-48d0-9274-eaac4ce5486d.json

@@ -1,7 +1,7 @@
 {
   "id": "0a211154-fda0-48d0-9274-eaac4ce5486d",
   "snapshot": "libreoffice_impress",
-  "instruction": "Set the background color of slides with a real person picture as yellow and set the title of slide 2 as \"Let's start\".",
+  "instruction": "Set the background color to yellow for any slide that contains one or more images of real people, and set the title of slide 2 as \"Let's start\".",
   "source": "https://arxiv.org/pdf/2311.01767.pdf",
   "config": [
     {

evaluation_examples/examples/libreoffice_impress/af2d657a-e6b3-4c6a-9f67-9e3ed015974c.json

@@ -83,4 +83,4 @@
   "proxy": false,
   "fixed_ip": false,
   "possibility_of_env_change": "low"
-}
+}

evaluation_examples/examples/multi_apps/82e3c869-49f6-4305-a7ce-f3e64a0618e7.json

@@ -112,4 +112,4 @@
   "proxy": false,
   "fixed_ip": false,
   "possibility_of_env_change": "low"
-}
+}

evaluation_examples/examples/multi_apps/869de13e-bef9-4b91-ba51-f6708c40b096.json

@@ -168,4 +168,4 @@
   "proxy": false,
   "fixed_ip": false,
   "possibility_of_env_change": "low"
-}
+}

monitor/templates/task_detail.html

@@ -73,7 +73,7 @@
                     <div class="video-player">
                         <video id="task-recording" controls playsinline webkit-playsinline
                                preload="metadata" width="100%" controlslist="nodownload">
-                            <source src="/task/{{ task_type }}/{{ task_id }}/recording" type="video/mp4">
+                            <source src="/task/{{ task_type }}/{{ task_id }}/recording?action_space={{ action_space }}&observation_type={{ observation_type }}&model_name={{ model_name }}" type="video/mp4">
                             Your browser does not support the video tag.
                         </video>
                         <div id="video-status" class="video-status">Loading video...</div>
@@ -103,7 +103,7 @@
                             {% endif %}</pre>
                             {% if step.screenshot_file %}
                                 <div>
-                                    <img src="/task/{{ task_type }}/{{ task_id }}/screenshot/{{ step.screenshot_file }}" 
+                                    <img src="/task/{{ task_type }}/{{ task_id }}/screenshot/{{ step.screenshot_file }}?action_space={{ action_space }}&observation_type={{ observation_type }}&model_name={{ model_name }}" 
                                             alt="Step {{ step.step_num }} Screenshot" class="step-image">
                                 </div>
                             {% endif %}
@@ -155,7 +155,7 @@
                 }, 2000);
                 
                 // Directly check video source availability with fetch API
-                fetch('/task/{{ task_type }}/{{ task_id }}/recording', {method: 'HEAD'})
+                fetch('/task/{{ task_type }}/{{ task_id }}/recording?action_space={{ action_space }}&observation_type={{ observation_type }}&model_name={{ model_name }}', {method: 'HEAD'})
                     .then(function(response) {
                         if (response.ok && videoStatus.textContent === 'Loading video...') {
                             // If HEAD request succeeds but video events haven't fired