feat: add toggle checkbox for qa

app.py

@@ -11,7 +11,7 @@ from src.about import (
 from src.benchmarks import DOMAIN_COLS_QA, LANG_COLS_QA, DOMAIN_COLS_LONG_DOC, LANG_COLS_LONG_DOC, METRIC_LIST, \
     DEFAULT_METRIC
 from src.display.css_html_js import custom_css
-from src.display.utils import COL_NAME_IS_ANONYMOUS
+from src.display.utils import COL_NAME_IS_ANONYMOUS, COL_NAME_REVISION, COL_NAME_TIMESTAMP
 from src.envs import API, EVAL_RESULTS_PATH, REPO_ID, RESULTS_REPO, TOKEN
 from src.read_evals import get_raw_eval_results, get_leaderboard_df
 from src.utils import update_table, update_metric, update_table_long_doc, upload_file, get_default_cols, submit_results
@@ -45,6 +45,7 @@ leaderboard_df_qa = original_df_qa.copy()
 shown_columns_qa, types_qa = get_default_cols(
     'qa', leaderboard_df_qa.columns, add_fix_cols=True)
 leaderboard_df_qa = leaderboard_df_qa[~leaderboard_df_qa[COL_NAME_IS_ANONYMOUS]][shown_columns_qa]
+leaderboard_df_qa.drop([COL_NAME_REVISION, COL_NAME_TIMESTAMP], axis=1, inplace=True)
 
 leaderboard_df_long_doc = original_df_long_doc.copy()
 shown_columns_long_doc, types_long_doc = get_default_cols(
@@ -80,9 +81,10 @@ def update_table_without_ranking(
         langs,
         reranking_query,
         query,
-        show_anonymous
+        show_anonymous,
+        show_revision_and_timestamp,
 ):
-    return update_table(hidden_df, domains, langs, reranking_query, query, show_anonymous, reset_ranking=False)
+    return update_table(hidden_df, domains, langs, reranking_query, query, show_anonymous, reset_ranking=False, show_revision_and_timestamp=show_revision_and_timestamp)
 
 
 def update_table_without_ranking_long_doc(
@@ -193,6 +195,20 @@ with demo:
                 leaderboard_table,
             )
 
+            show_revision_and_timestamp.change(
+                update_table_without_ranking,
+                [
+                    hidden_leaderboard_table_for_search,
+                    selected_domains,
+                    selected_langs,
+                    selected_rerankings,
+                    search_bar,
+                    show_anonymous,
+                    show_revision_and_timestamp
+                ],
+                leaderboard_table
+            )
+
             selected_rerankings.change(
                 update_table_without_ranking,
                 [
@@ -202,6 +218,7 @@ with demo:
                     selected_rerankings,
                     search_bar,
                     show_anonymous,
+                    show_revision_and_timestamp
                 ],
                 leaderboard_table,
             )

src/utils.py

@@ -9,7 +9,7 @@ import pandas as pd
 from src.benchmarks import BENCHMARK_COLS_QA, BENCHMARK_COLS_LONG_DOC, BenchmarksQA, BenchmarksLongDoc
 from src.display.formatting import styled_message, styled_error
 from src.display.utils import COLS_QA, TYPES_QA, COLS_LONG_DOC, TYPES_LONG_DOC, COL_NAME_RANK, COL_NAME_AVG, \
-    COL_NAME_RERANKING_MODEL, COL_NAME_RETRIEVAL_MODEL, COL_NAME_IS_ANONYMOUS, get_default_auto_eval_column_dict
+    COL_NAME_RERANKING_MODEL, COL_NAME_RETRIEVAL_MODEL, COL_NAME_IS_ANONYMOUS, COL_NAME_TIMESTAMP, COL_NAME_REVISION, get_default_auto_eval_column_dict
 from src.envs import API, SEARCH_RESULTS_REPO
 from src.read_evals import FullEvalResult, get_leaderboard_df, calculate_mean
 
@@ -130,14 +130,18 @@ def _update_table(
         reranking_query: list,
         query: str,
         show_anonymous: bool,
-        reset_ranking: bool = True
+        reset_ranking: bool = True,
+        show_revision_and_timestamp: bool = False
 ):
     filtered_df = hidden_df.copy()
     if not show_anonymous:
         filtered_df = filtered_df[~filtered_df[COL_NAME_IS_ANONYMOUS]]
     filtered_df = filter_models(filtered_df, reranking_query)
     filtered_df = filter_queries(query, filtered_df)
-    return select_columns(filtered_df, domains, langs, task, reset_ranking)
+    filtered_df = select_columns(filtered_df, domains, langs, task, reset_ranking)
+    if not show_revision_and_timestamp:
+        filtered_df.drop([COL_NAME_REVISION, COL_NAME_TIMESTAMP], axis=1, inplace=True)
+    return filtered_df
 
 
 def update_table(
@@ -147,10 +151,11 @@ def update_table(
         reranking_query: list,
         query: str,
         show_anonymous: bool,
-        reset_ranking: bool = True
+        reset_ranking: bool = True,
+        show_revision_and_timestamp: bool = False
 ):
     return _update_table(
-        "qa", hidden_df, domains, langs, reranking_query, query, show_anonymous, reset_ranking)
+        "qa", hidden_df, domains, langs, reranking_query, query, show_anonymous, reset_ranking, show_revision_and_timestamp)
 
 
 def update_table_long_doc(