Fix ReadTimeoutError crash on Space initialization

app.py

@@ -49,6 +49,19 @@ leaderboard_df = None
 logger.info("Initialized LBDB")
 
 
+def get_empty_leaderboard_df():
+    return pd.DataFrame(
+        columns=[
+            AutoEvalColumn.system.name,
+            AutoEvalColumn.organization.name,
+            AutoEvalColumn.success_rate_overall.name,
+            AutoEvalColumn.success_rate_tier1.name,
+            AutoEvalColumn.success_rate_tier2.name,
+            AutoEvalColumn.submitted_on.name,
+        ]
+    )
+
+
 def restart_space():
     logger.info("Restarting space")
     API.restart_space(repo_id=REPO_ID)
@@ -66,10 +79,12 @@ def refresh_leaderboard_data():
             leaderboard_df = new_leaderboard_df
         else:
             logger.warning("No new leaderboard data found")
-            return None
+            if leaderboard_df is None:
+                leaderboard_df = get_empty_leaderboard_df()
     except Exception as e:
         logger.error(f"Error refreshing leaderboard data: {e}")
-        return None
+        if leaderboard_df is None:
+            leaderboard_df = get_empty_leaderboard_df()
 
 
 def init_leaderboard(dataframe: pd.DataFrame):
@@ -511,8 +526,12 @@ with blocks:
                 elem_classes="markdown-text",
             )
             refresh_leaderboard_data()
-            assert leaderboard_df is not None
-            leaderboard_component = init_leaderboard(leaderboard_df)
+            if leaderboard_df is None or leaderboard_df.empty:
+                logger.warning("Displaying empty leaderboard - no data available")
+                gr.Markdown("⚠️ Leaderboard data unavailable. Submit solutions to populate it!")
+                leaderboard_component = init_leaderboard(get_empty_leaderboard_df())
+            else:
+                leaderboard_component = init_leaderboard(leaderboard_df)
 
         with gr.TabItem("Submit Solutions", elem_id="formulaone-submit-tab-table", id=3):
             logger.info("Tab submission")

src/populate.py

@@ -1,5 +1,5 @@
 import pandas as pd
-from datasets import get_dataset_config_names, load_dataset
+from datasets import DownloadConfig, get_dataset_config_names, load_dataset
 from datasets.exceptions import DatasetNotFoundError
 from tqdm.auto import tqdm
 
@@ -47,13 +47,18 @@ def get_leaderboard_df(results_dataset_name: str) -> pd.DataFrame:
         total=len(configs),
         desc="Processing Submission Results",
     ):
-        submission_ds = load_dataset(
-            results_dataset_name,
-            submission_id,
-            split="train",
-            token=TOKEN,
-        )
-        submission_df = pd.DataFrame(submission_ds)
+        try:
+            submission_ds = load_dataset(
+                results_dataset_name,
+                submission_id,
+                split="train",
+                token=TOKEN,
+                download_config=DownloadConfig(timeout=60),
+            )
+            submission_df = pd.DataFrame(submission_ds)
+        except Exception as e:
+            logger.warning(f"Failed to load submission {submission_id}: {e}")
+            continue
 
         if submission_df.empty or "did_pass" not in submission_df.columns or submission_df.did_pass.isna().any():
             logger.warning(f"Skipping {submission_id} due to invalid did_pass values")