Refactor init_leaderboard function to handle multiple subsets, improve column selection and hiding, and include Dataset Version in filter_columns

app.py

@@ -65,40 +65,80 @@ except Exception:
     restart_space()
 
 
-LEADERBOARD_DF, SUBSETS = get_leaderboard_df(RESULTS_REPO)
+LEADERBOARD_DF = get_leaderboard_df(RESULTS_REPO)
 
 
-def init_leaderboard(dataframes, subsets):
-    subsets = list(subsets)
+def init_leaderboard(dataframes):
+    subsets = list(dataframes.keys())
     
     with gr.Row():
         selected_subset = gr.Dropdown(choices=subsets, label="Select Dataset Subset", value=subsets[-1])
+        research_textbox = gr.Textbox(placeholder="🔍 Search Models... [press enter]", label="Filter Models by Name")
+        selected_columns = gr.CheckboxGroup(choices=[c.name for c in fields(AutoEvalColumn) if not c.hidden], label="Select Columns to Display", value=[c.name for c in fields(AutoEvalColumn) if c.displayed_by_default])
+
     
-    # with gr.Row():
-    #     datatype = [c.type for c in fields(AutoEvalColumn)]
-    #     dataframe = gr.Dataframe(dataframes, datatype=datatype, type="pandas")
-
-
-    return Leaderboard(
-        value=dataframes,
-        datatype=[c.type for c in fields(AutoEvalColumn)],
-        select_columns=SelectColumns(
-            default_selection=[c.name for c in fields(AutoEvalColumn) if c.displayed_by_default],
-            cant_deselect=[c.name for c in fields(AutoEvalColumn) if c.never_hidden],
-            label="Select Columns to Display:",
-        ),
-        search_columns=[AutoEvalColumn.model.name],
-        hide_columns=[c.name for c in fields(AutoEvalColumn) if c.hidden],
-        filter_columns=[
-            ColumnFilter(
-                column=AutoEvalColumn.dataset_version.name,
-                choices=subsets,
-                default=subsets[-1],
-            )
-            # gr.Dropdown(choices=subsets, label="Select Dataset Subset", value=subsets[-1])
-        ],
-        interactive=False,
+    data = dataframes[subsets[-1]]
+    
+    with gr.Row():
+        datatype = [c.type for c in fields(AutoEvalColumn)]
+        df = gr.Dataframe(data, datatype=datatype, type="pandas")
+
+    def refresh(subset):
+        global LEADERBOARD_DF
+        LEADERBOARD_DF = get_leaderboard_df(RESULTS_REPO)
+        research_textbox.value = ""
+        selected_subset.choices = subsets
+        update_data(subset, research_textbox, selected_columns)
+
+        
+
+    def update_data(subset, search_term, selected_columns):
+        return dataframes[subset][dataframes[subset].model.str.contains(search_term, case=False)][selected_columns]
+
+    with gr.Row():
+        refresh_button = gr.Button("Refresh")
+        refresh_button.click(refresh, inputs=[
+            selected_subset, 
+        ], outputs=data, concurrency_limit=20)
+
+
+    
+    selected_subset.change(update_data, inputs=[
+        selected_subset, research_textbox, selected_columns
+        ], outputs=data)
+    research_textbox.submit(
+        update_data, 
+        inputs=[selected_subset, research_textbox, selected_columns], 
+        outputs=data
+    )
+    selected_columns.change(
+        update_data, 
+        inputs=[selected_subset, research_textbox, selected_columns],
+        outputs=data
     )
+
+
+
+    # return Leaderboard(
+    #     value=dataframes,
+    #     datatype=[c.type for c in fields(AutoEvalColumn)],
+    #     select_columns=SelectColumns(
+    #         default_selection=[c.name for c in fields(AutoEvalColumn) if c.displayed_by_default],
+    #         cant_deselect=[c.name for c in fields(AutoEvalColumn) if c.never_hidden],
+    #         label="Select Columns to Display:",
+    #     ),
+    #     search_columns=[AutoEvalColumn.model.name],
+    #     hide_columns=[c.name for c in fields(AutoEvalColumn) if c.hidden],
+    #     filter_columns=[
+    #         ColumnFilter(
+    #             column=AutoEvalColumn.dataset_version.name,
+    #             choices=subsets,
+    #             default=subsets[-1],
+    #         )
+    #         # gr.Dropdown(choices=subsets, label="Select Dataset Subset", value=subsets[-1])
+    #     ],
+    #     interactive=False,
+    # )
     
 demo = gr.Blocks(css=custom_css)
 with demo:
@@ -107,7 +147,7 @@ with demo:
 
     with gr.Tabs(elem_classes="tab-buttons") as tabs:
         with gr.TabItem("🏅 LiveBench Results", elem_id="llm-benchmark-tab-table", id=0):
-            init_leaderboard(LEADERBOARD_DF, SUBSETS)
+            init_leaderboard(LEADERBOARD_DF)
 
         with gr.TabItem("📝 About", elem_id="llm-benchmark-tab-table", id=2):
             gr.Markdown(LLM_BENCHMARKS_TEXT, elem_classes="markdown-text")

src/populate.py

@@ -17,8 +17,9 @@ def get_leaderboard_df(results_repo):
         subset_df = subset_data.to_pandas()
         subset_df = subset_df.sort_values(by="Total", ascending=False)
         subset_df = subset_df.round(2)  # Round all numeric columns to two decimal places
-        subset_df["Dataset Version"] = [subset] * len(subset_df)
+        # subset_df["Dataset Version"] = [subset] * len(subset_df)
         print(subset_df)
         subset_dfs[subset] = subset_df
-    df = pd.concat(subset_dfs.values())
-    return df, subset_dfs.keys()
+    # df = pd.concat(subset_dfs.values())
+    # return df, subset_dfs.keys()
+    return subset_dfs