maybe fix storage

app.py

@@ -3,6 +3,9 @@ from gradio_leaderboard import Leaderboard, ColumnFilter, SelectColumns
 import pandas as pd
 from apscheduler.schedulers.background import BackgroundScheduler
 from huggingface_hub import snapshot_download
+import os
+import shutil
+import time
 
 from src.about import (
     CITATION_BUTTON_LABEL,
@@ -41,20 +44,50 @@ def restart_space():
     API.restart_space(repo_id=REPO_ID)
 
 ### Space initialisation
-try:
-    print(EVAL_REQUESTS_PATH)
-    snapshot_download(
-        repo_id=QUEUE_REPO, local_dir=EVAL_REQUESTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30, token=TOKEN
-    )
-except Exception:
-    restart_space()
-try:
-    print(EVAL_RESULTS_PATH)
-    snapshot_download(
-        repo_id=RESULTS_REPO, local_dir=EVAL_RESULTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30, token=TOKEN
-    )
-except Exception:
-    restart_space()
+
+def cleanup_old_cache():
+    """Remove old cache directories to free up space"""
+    cache_dirs = [EVAL_REQUESTS_PATH, EVAL_RESULTS_PATH]
+    for cache_dir in cache_dirs:
+        if os.path.exists(cache_dir):
+            # Check if cache is older than 6 hours
+            cache_age = time.time() - os.path.getctime(cache_dir)
+            if cache_age > 21600:  # 6 hours in seconds
+                print(f"Removing old cache: {cache_dir}")
+                shutil.rmtree(cache_dir, ignore_errors=True)
+
+def safe_download_dataset(repo_id, local_dir, repo_type="dataset"):
+    """Download dataset only if not already cached or cache is stale"""
+    try:
+        # Check if directory exists and has recent data
+        if os.path.exists(local_dir) and os.listdir(local_dir):
+            cache_age = time.time() - os.path.getctime(local_dir)
+            if cache_age < 3600:  # Less than 1 hour old
+                print(f"Using cached data: {local_dir}")
+                return
+
+        print(f"Downloading: {repo_id} to {local_dir}")
+        snapshot_download(
+            repo_id=repo_id,
+            local_dir=local_dir,
+            repo_type=repo_type,
+            tqdm_class=None,
+            etag_timeout=30,
+            token=TOKEN,
+            resume_download=True,
+            force_download=False
+        )
+    except Exception as e:
+        print(f"Download failed for {repo_id}: {e}")
+        if not os.path.exists(local_dir) or not os.listdir(local_dir):
+            restart_space()
+
+# Clean up old cache to free space
+cleanup_old_cache()
+
+# Download datasets with caching
+safe_download_dataset(QUEUE_REPO, EVAL_REQUESTS_PATH, "dataset")
+safe_download_dataset(RESULTS_REPO, EVAL_RESULTS_PATH, "dataset")
 
 
 LEADERBOARD_DF = get_leaderboard_df(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, COLS, BENCHMARK_COLS)

src/envs.py

@@ -14,12 +14,15 @@ QUEUE_REPO = f"{OWNER}/requests"
 RESULTS_REPO = f"{OWNER}/results"
 
 # If you setup a cache later, just change HF_HOME
-CACHE_PATH=os.getenv("HF_HOME", ".")
+CACHE_PATH=os.getenv("HF_HOME", "/tmp/hf_cache")
 
-# Local caches
+# Local caches - use /tmp for better cleanup
 EVAL_REQUESTS_PATH = os.path.join(CACHE_PATH, "eval-queue")
 EVAL_RESULTS_PATH = os.path.join(CACHE_PATH, "eval-results")
 EVAL_REQUESTS_PATH_BACKEND = os.path.join(CACHE_PATH, "eval-queue-bk")
 EVAL_RESULTS_PATH_BACKEND = os.path.join(CACHE_PATH, "eval-results-bk")
 
+# Ensure cache directory exists
+os.makedirs(CACHE_PATH, exist_ok=True)
+
 API = HfApi(token=TOKEN)