simplify_ux

app.py

@@ -1,3 +1,4 @@
+import os
 import logging
 import time
 import schedule
@@ -60,18 +61,8 @@ NEW_DATA_ON_LEADERBOARD = True
 LEADERBOARD_DF = None
 
 def restart_space():
-    logging.info(f"Restarting space with repo ID: {REPO_ID}")
-    try:
-        # Check if new data is pending and download if necessary
-        if NEW_DATA_ON_LEADERBOARD:
-            logging.info("Fetching latest leaderboard data before restart.")
-            get_latest_data_leaderboard()
+    API.restart_space(repo_id=REPO_ID, token=HF_TOKEN)
 
-        # Now restart the space
-        API.restart_space(repo_id=REPO_ID, token=HF_TOKEN)
-        logging.info("Space restarted successfully.")
-    except Exception as e:
-        logging.error(f"Failed to restart space: {e}")
 
 def time_diff_wrapper(func):
     def wrapper(*args, **kwargs):
@@ -109,35 +100,29 @@ def download_dataset(repo_id, local_dir, repo_type="dataset", max_attempts=3, ba
             attempt += 1
     raise Exception(f"Failed to download {repo_id} after {max_attempts} attempts")
 
-def get_latest_data_leaderboard(leaderboard_initial_df=None):
+def get_latest_data_leaderboard(leaderboard_initial_df = None):
     global NEW_DATA_ON_LEADERBOARD
     global LEADERBOARD_DF
     if NEW_DATA_ON_LEADERBOARD:
-        logging.info("Leaderboard updated at reload!")
-        try:
-            leaderboard_dataset = datasets.load_dataset(
-                AGGREGATED_REPO,
-                "default",
-                split="train",
-                cache_dir=HF_HOME,
-                download_mode=datasets.DownloadMode.FORCE_REDOWNLOAD,  # Always download fresh data
-                verification_mode="no_checks"
-            )
-            LEADERBOARD_DF = get_leaderboard_df(
-                leaderboard_dataset=leaderboard_dataset,
-                cols=COLS,
-                benchmark_cols=BENCHMARK_COLS,
-            )
-            logging.info("Leaderboard dataset successfully downloaded.")
-        except Exception as e:
-            logging.error(f"Failed to download leaderboard dataset: {e}")
-            return
-
-        # Reset the flag after successful download
+        print("Leaderboard updated at reload!")
+        leaderboard_dataset = datasets.load_dataset(
+            AGGREGATED_REPO, 
+            "default", 
+            split="train", 
+            cache_dir=HF_HOME, 
+            download_mode=datasets.DownloadMode.REUSE_DATASET_IF_EXISTS, # Uses the cached dataset 
+            verification_mode="no_checks"
+        )
+        LEADERBOARD_DF = get_leaderboard_df(
+            leaderboard_dataset=leaderboard_dataset, 
+            cols=COLS,
+            benchmark_cols=BENCHMARK_COLS,
+        )
         NEW_DATA_ON_LEADERBOARD = False
+
     else:
         LEADERBOARD_DF = leaderboard_initial_df
-        logging.info("Using cached leaderboard dataset.")
+
     return LEADERBOARD_DF
 
 
@@ -147,9 +132,6 @@ def get_latest_data_queue():
 
 def init_space():
     """Initializes the application space, loading only necessary data."""
-    global NEW_DATA_ON_LEADERBOARD
-    NEW_DATA_ON_LEADERBOARD = True  # Ensure new data is always pulled on restart
-
     if DO_FULL_INIT:
         # These downloads only occur on full initialization
         try:
@@ -467,13 +449,18 @@ def update_leaderboard(payload: WebhookPayload) -> None:
     """Redownloads the leaderboard dataset each time it updates"""
     if payload.repo.type == "dataset" and payload.event.action == "update":
         global NEW_DATA_ON_LEADERBOARD
-        logging.info("New data detected, downloading updated leaderboard dataset.")
-        
-        # Mark the flag for new data
+        if NEW_DATA_ON_LEADERBOARD:
+            return
         NEW_DATA_ON_LEADERBOARD = True
 
-        # Now actually download the latest data immediately
-        get_latest_data_leaderboard()
+        datasets.load_dataset(
+            AGGREGATED_REPO, 
+            "default", 
+            split="train", 
+            cache_dir=HF_HOME, 
+            download_mode=datasets.DownloadMode.FORCE_REDOWNLOAD, 
+            verification_mode="no_checks"
+        )
 
 # The below code is not used at the moment, as we can manage the queue file locally
 LAST_UPDATE_QUEUE = datetime.datetime.now()
@@ -493,6 +480,5 @@ def update_queue(payload: WebhookPayload) -> None:
 webhooks_server.launch()
 
 scheduler = BackgroundScheduler()
-scheduler.add_job(restart_space, "interval", hours=1)  # Restart every 1h
-logging.info("Scheduler initialized to restart space every 1 hour.")
+scheduler.add_job(restart_space, "interval", hours=3) # restarted every 3h as backup in case automatic updates are not working
 scheduler.start()

pyproject.toml

@@ -15,9 +15,9 @@ dependencies = [
     "pandas>=2.2.2",
     "python-dateutil>=2.9.0",
     "sentencepiece>=0.2.0",
-    "transformers==4.45.1",
+    "transformers==4.44.2",
     "tokenizers>=0.19.0",
-    "gradio-space-ci",
+    "gradio-space-ci @ git+https://huggingface.co/spaces/Wauplin/[email protected]",
     "isort>=5.13.2",
     "ruff>=0.6.4",
     "gradio-leaderboard==0.0.11",
@@ -34,16 +34,16 @@ ignore=["I","EM","FBT","TRY003","S101","D101","D102","D103","D104","D105","G004"
 fixable=["ALL"]
 select=["ALL"]
 
-[tool.ruff.lint]
+ [tool.ruff.lint]
 select = ["E", "F"]
 fixable = ["ALL"]
 ignore = ["E501"] # line too long (black is taking care of this)
 
-[tool.isort]
+ [tool.isort]
 profile = "black"
 
 [tool.black]
 line-length = 119
 
-[tool.uv.sources]
-gradio-space-ci = { git = "https://huggingface.co/spaces/Wauplin/gradio-space-ci" }
+[tool.hatch.metadata]
+allow-direct-references = true

requirements.txt

@@ -6,7 +6,7 @@ huggingface-hub>=0.24.7
 pandas==2.2.2
 python-dateutil==2.9.0
 sentencepiece==0.2.0
-transformers==4.45.1
+transformers==4.44.2
 tokenizers>=0.19.0
 gradio-space-ci @ git+https://huggingface.co/spaces/Wauplin/[email protected] # CI !!!
 isort==5.13.2