Add timeout to weight download and .dockerignore to prevent repeated slow downloads

.dockerignore

@@ -0,0 +1,65 @@
+# Git and version control
+.git
+.gitignore
+.gitattributes
+
+# Python cache
+__pycache__
+*.pyc
+*.pyo
+*.egg-info
+.pytest_cache
+.tox
+
+# Virtual environments
+venv
+env
+.venv
+
+# IDE and editors
+.vscode
+.idea
+*.swp
+*.swo
+*~
+
+# OS files
+.DS_Store
+Thumbs.db
+.directory
+
+# Large files that shouldn't be in Docker
+*.pth
+*.pt
+*.whl
+*.zip
+*.tar.gz
+
+# Test and log files
+*.log
+test_output
+*.tmp
+
+# Frontend dev
+node_modules
+npm-debug.log
+
+# Model outputs (will be generated at runtime)
+deployment/backend/outputs/
+model/work_dirs/
+model/result/
+
+# Documentation
+docs/
+*.md
+
+# Training data and models
+data/
+checkpoints/
+weights/
+
+# Assets
+assets/
+*.png
+*.jpg
+*.jpeg

Dockerfile

@@ -26,14 +26,16 @@ COPY requirements.txt /app/
 RUN pip install --no-cache-dir --upgrade pip setuptools wheel && \
     pip install --no-cache-dir -r requirements.txt
 
-# Copy application code
+# Copy application code (excluding large files via .dockerignore)
 COPY . /app/
 
 # Create checkpoint directory
 RUN mkdir -p /app/finetuning_rodla/finetuning_rodla/checkpoints
 
-# Download model weights during build
-RUN python /app/deployment/backend/download_weights.py || echo "⚠️  Weight download script completed (may need manual download)"
+# Optional: Attempt to download weights at build time (non-blocking)
+# If download fails or times out, the backend will fall back to heuristic detection
+# Weights can be manually placed in: /app/finetuning_rodla/finetuning_rodla/checkpoints/rodla_internimage_xl_publaynet.pth
+RUN timeout 300 python /app/deployment/backend/download_weights.py 2>&1 || echo "⚠️  Weight download timed out or failed - backend will use heuristic detection. To enable full detection, manually place weights file in checkpoints directory."
 
 # Run backend.py on port 7860 (HuggingFace standard)
 CMD ["python", "deployment/backend/backend.py"]

deployment/backend/download_weights.py

@@ -2,56 +2,95 @@
 """
 Download model weights from Google Drive
 This script is used during Docker build or startup to download required weights
+Includes timeout to prevent hanging on HuggingFace Spaces
 """
 import os
-import urllib.request
 import sys
+import signal
 from pathlib import Path
 import gdown
 
-def download_weights():
-    """Download weights from Google Drive"""
-    
-    # Google Drive file ID from the shared link
-    # https://drive.google.com/file/d/1BHyz2jH52Irt6izCeTRb4g2J5lXsA9cz/view?usp=sharing
-    DRIVE_FILE_ID = "1BHyz2jH52Irt6izCeTRb4g2J5lXsA9cz"
-    
-    # Destination path
-    REPO_ROOT = Path("/home/admin/CV/rodla-academic")
-    WEIGHTS_DIR = REPO_ROOT / "finetuning_rodla/finetuning_rodla/checkpoints"
-    WEIGHTS_DIR.mkdir(parents=True, exist_ok=True)
-    
-    WEIGHTS_PATH = WEIGHTS_DIR / "rodla_internimage_xl_publaynet.pth"
-    
-    # Check if weights already exist and are not empty
-    if WEIGHTS_PATH.exists() and WEIGHTS_PATH.stat().st_size > 1000000:  # > 1MB
-        print(f"✅ Weights already exist at {WEIGHTS_PATH}")
-        print(f"   Size: {WEIGHTS_PATH.stat().st_size / (1024**3):.2f} GB")
-        return True
+# Timeout handler
+class TimeoutError(Exception):
+    pass
+
+def timeout_handler(signum, frame):
+    raise TimeoutError("Download timed out")
+
+def download_weights(timeout_seconds=300):
+    """Download weights from Google Drive with timeout"""
     
-    print(f"📥 Downloading model weights from Google Drive...")
-    print(f"   File ID: {DRIVE_FILE_ID}")
-    print(f"   Destination: {WEIGHTS_PATH}")
+    # Set timeout (5 minutes max for HuggingFace Spaces)
+    signal.signal(signal.SIGALRM, timeout_handler)
+    signal.alarm(timeout_seconds)
     
     try:
-        # Download using gdown
-        url = f"https://drive.google.com/uc?id={DRIVE_FILE_ID}"
-        gdown.download(url, str(WEIGHTS_PATH), quiet=False)
+        # Google Drive file ID from the shared link
+        # https://drive.google.com/file/d/1BHyz2jH52Irt6izCeTRb4g2J5lXsA9cz/view?usp=sharing
+        DRIVE_FILE_ID = "1BHyz2jH52Irt6izCeTRb4g2J5lXsA9cz"
+        
+        # Destination path - works in Docker (/app) and local dev
+        if os.path.exists("/app"):
+            REPO_ROOT = Path("/app")
+        else:
+            REPO_ROOT = Path(__file__).parent.parent.parent.parent
+        
+        WEIGHTS_DIR = REPO_ROOT / "finetuning_rodla/finetuning_rodla/checkpoints"
+        WEIGHTS_DIR.mkdir(parents=True, exist_ok=True)
         
-        # Verify download
-        if WEIGHTS_PATH.exists() and WEIGHTS_PATH.stat().st_size > 1000000:
-            print(f"✅ Download successful!")
+        WEIGHTS_PATH = WEIGHTS_DIR / "rodla_internimage_xl_publaynet.pth"
+        
+        # Check if weights already exist and are not empty
+        if WEIGHTS_PATH.exists() and WEIGHTS_PATH.stat().st_size > 1000000:  # > 1MB
+            print(f"✅ Weights already exist at {WEIGHTS_PATH}")
             print(f"   Size: {WEIGHTS_PATH.stat().st_size / (1024**3):.2f} GB")
+            signal.alarm(0)  # Cancel alarm
             return True
-        else:
-            print(f"❌ Download failed or file is too small")
-            return False
+        
+        print(f"📥 Downloading model weights from Google Drive...")
+        print(f"   File ID: {DRIVE_FILE_ID}")
+        print(f"   Destination: {WEIGHTS_PATH}")
+        print(f"   Timeout: {timeout_seconds} seconds")
+        
+        try:
+            # Download using gdown
+            url = f"https://drive.google.com/uc?id={DRIVE_FILE_ID}"
+            gdown.download(url, str(WEIGHTS_PATH), quiet=False)
             
+            # Verify download
+            if WEIGHTS_PATH.exists() and WEIGHTS_PATH.stat().st_size > 1000000:
+                print(f"✅ Download successful!")
+                print(f"   Size: {WEIGHTS_PATH.stat().st_size / (1024**3):.2f} GB")
+                signal.alarm(0)  # Cancel alarm
+                return True
+            else:
+                print(f"❌ Download failed or file is too small")
+                signal.alarm(0)  # Cancel alarm
+                return False
+                
+        except TimeoutError:
+            print(f"⏱️  Download timed out after {timeout_seconds} seconds")
+            print(f"   This is expected on HuggingFace Spaces. Backend will use heuristic detection.")
+            signal.alarm(0)  # Cancel alarm
+            return False
+        except Exception as e:
+            print(f"❌ Error downloading weights: {e}")
+            signal.alarm(0)  # Cancel alarm
+            return False
+    
+    except TimeoutError:
+        print(f"⏱️  Download timed out after {timeout_seconds} seconds")
+        print(f"   This is expected on HuggingFace Spaces. Backend will use heuristic detection.")
+        return False
     except Exception as e:
-        print(f"❌ Error downloading weights: {e}")
+        print(f"❌ Unexpected error: {e}")
         return False
+    finally:
+        signal.alarm(0)  # Make sure alarm is cancelled
 
 
 if __name__ == "__main__":
-    success = download_weights()
+    # Allow timeout to be passed as argument, default to 5 minutes
+    timeout = int(sys.argv[1]) if len(sys.argv) > 1 else 300
+    success = download_weights(timeout_seconds=timeout)
     sys.exit(0 if success else 1)