Implement login for GPQA Details

README.md

@@ -6,6 +6,7 @@ colorTo: green
 sdk: gradio
 sdk_version: 4.44.1
 app_file: app.py
+hf_oauth: true
 pinned: false
 short_description: Compare Open LLM Leaderboard results
 datasets:

app.py

@@ -82,7 +82,7 @@ with gr.Blocks(fill_height=True, fill_width=True) as demo:
             configs = gr.HTML()
         with gr.Tab("Details"):
             details_task = gr.Radio(
-                [value for value in constants.TASKS.values() if value[1] != "leaderboard_gpqa"],
+                list(constants.TASKS.values()),
                 label="Tasks",
                 info="Evaluation tasks to be loaded",
                 interactive=True,
@@ -91,11 +91,13 @@ with gr.Blocks(fill_height=True, fill_width=True) as demo:
                 label="Task Description",
                 lines=3,
             )
-            subtask = gr.Radio(
-                # constants.SUBTASKS.get(details_task.value),
-                label="Subtasks",
-                info="Evaluation subtasks to be loaded (choose one of the Tasks above)",
-            )
+            with gr.Row():
+                login_btn = gr.LoginButton(size="sm", visible=False)
+                subtask = gr.Radio(
+                    choices=None,  # constants.SUBTASKS.get(details_task.value),
+                    label="Subtasks",
+                    info="Evaluation subtasks to be loaded (choose one of the Tasks above)",
+                )
             load_details_btn = gr.Button("Load Details", interactive=False)
             clear_details_btn = gr.Button("Clear Details")
             sample_idx = gr.Number(
@@ -166,7 +168,7 @@ with gr.Blocks(fill_height=True, fill_width=True) as demo:
     ).then(
         fn=update_subtasks_component,
         inputs=details_task,
-        outputs=subtask,
+        outputs=[login_btn, subtask],
     )
     gr.on(
         triggers=[model_id_1.input, model_id_2.input, subtask.input, details_task.input],

src/constants.py

@@ -39,9 +39,9 @@ SUBTASKS = {
         ("Web of Lies", "leaderboard_bbh_web_of_lies"),
     ],
     "leaderboard_gpqa": [
-        "leaderboard_gpqa_extended",
-        "leaderboard_gpqa_diamond",
-        "leaderboard_gpqa_main",
+        ("Extended", "leaderboard_gpqa_extended"),
+        ("Diamond", "leaderboard_gpqa_diamond"),
+        ("Main", "leaderboard_gpqa_main"),
     ],
     "leaderboard_ifeval": [("IFEval", "leaderboard_ifeval")],
     # "leaderboard_math_hard": [

src/details.py

@@ -20,11 +20,16 @@ def update_task_description_component(task):
     )
 
 
-def update_subtasks_component(task):
-    return gr.Radio(
-        constants.SUBTASKS.get(task),
-        info="Evaluation subtasks to be loaded",
-        value=None,
+def update_subtasks_component(task, profile: gr.OAuthProfile | None):
+    visible_login_btn = True if task == "leaderboard_gpqa" else False
+    subtasks = None if task == "leaderboard_gpqa" and not profile else constants.SUBTASKS.get(task)
+    return (
+        gr.LoginButton(size="sm", visible=visible_login_btn),
+        gr.Radio(
+            choices=subtasks,
+            info="Evaluation subtasks to be loaded",
+            value=None,
+        ),
     )