metrics

app.py

@@ -4,6 +4,7 @@ from apscheduler.schedulers.background import BackgroundScheduler
 from huggingface_hub import snapshot_download
 
 from src.display.about import (
+    METRICS,
     CITATION_BUTTON_LABEL,
     CITATION_BUTTON_TEXT,
     EVALUATION_QUEUE_TEXT,
@@ -146,6 +147,18 @@ with demo:
                             show_label=False,
                             elem_id="search-bar",
                         )
+                    with gr.Row():
+                        shown_columns = gr.Radio(
+                            choices=[
+                                c for c in METRICS
+                            ],
+                            value=[
+                                c for c in METRICS
+                            ],
+                            label="Select metrics to show",
+                            elem_id="metrics-select",
+                            interactive=True,
+                        )
                     with gr.Row():
                         shown_columns = gr.CheckboxGroup(
                             choices=[

src/display/about.py

@@ -20,7 +20,7 @@ class Tasks(Enum):
     task6 = Task("logiqa2", "acc_base", "LogiQA2 Acc")
     task7 = Task("logiqa2", "acc_cot", "LogiQA2 AccCoT")
 
-
+METRICS = list(set([task.value.metric for task in Tasks]))