Fix

app.py

@@ -76,9 +76,9 @@ def refresh_overall_leaderboard():
     return format_leaderboard_table(df_results=current_dataframe)
 
 
-def fetch_latest_data():
+def fetch_latest_data(stop_event):
     import time
-    while True:
+    while not stop_event.is_set():
         try:
             fetch_hf_results()
         except Exception as e:
@@ -90,9 +90,11 @@ def fetch_latest_data():
 @contextlib.asynccontextmanager
 async def periodic_data_fetch(app):
     import threading
-    t = threading.Thread(target=fetch_latest_data, daemon=True)
+    event = threading.Event()
+    t = threading.Thread(target=fetch_latest_data, args=(event,), daemon=True)
     t.start()
     yield
+    event.set()
     t.join(3)
 
 

constants.py

@@ -68,7 +68,7 @@ API = HfApi(token=TOKEN)
 # Huggingface repos
 ORGANIZATION = "ginkgo-datapoints"
 SUBMISSIONS_REPO = f"{ORGANIZATION}/abdev-bench-submissions"
-RESULTS_REPO = f"{ORGANIZATION}/abdev-bench-results-test"
+RESULTS_REPO = f"{ORGANIZATION}/abdev-bench-results"
 
 # Leaderboard dataframes
 LEADERBOARD_RESULTS_COLUMNS = [

utils.py

@@ -30,7 +30,7 @@ def fetch_hf_results():
     # load_dataset should cache by default if not using force_redownload
     df = load_dataset(
         RESULTS_REPO,
-        data_files="data/train-00000-of-00001.parquet",
+        data_files="auto_submissions/metrics_all.csv",
     )["train"].to_pandas()
     print("fetched results from HF", df.shape)
     assert all(
@@ -56,7 +56,6 @@ def fetch_hf_results():
     # Note: Could optionally add a column "is_baseline" to the dataframe to indicate whether the model is a baseline model or not. If things get crowded.
     # Anonymize the user column at this point (so note: users can submit anonymous / non-anonymous and we'll show their latest submission regardless)
     df.loc[df["anonymous"] != False, "user"] = "anon-" + df.loc[df["anonymous"] != False, "user"].apply(readable_hash)
-    print("after filtering to latest submissions only", df.shape)
     df.to_csv("debug-current-results.csv", index=False)