Graceful fallback to CPU

handler.py

@@ -19,9 +19,32 @@ from concurrent.futures import ThreadPoolExecutor, as_completed
 
 class EndpointHandler:
     def __init__(self, model_dir=None):
-        # Initialize FaceAnalysis with GPU support
-        self.app = FaceAnalysis(root="/tmp/.insightface", providers=['CUDAExecutionProvider', 'CPUExecutionProvider'])
-        self.app.prepare(ctx_id=0)  # 0 = GPU, -1 = CPU
+        # Initialize FaceAnalysis with GPU/CPU fallback support
+        print("\n" + "="*80)
+        print("INITIALIZING FACEANALYSIS")
+        print("="*80)
+        
+        try:
+            # Try GPU first
+            print("Attempting to initialize with GPU (CUDA)...")
+            self.app = FaceAnalysis(root="/tmp/.insightface", providers=['CUDAExecutionProvider', 'CPUExecutionProvider'])
+            self.app.prepare(ctx_id=0)  # 0 = GPU
+            print("✅ GPU initialization successful (ctx_id=0)")
+            self.gpu_available = True
+        except RuntimeError as e:
+            # GPU not available, fall back to CPU
+            print(f"⚠️ GPU initialization failed: {str(e)[:100]}...")
+            print("Falling back to CPU (CPUExecutionProvider)...")
+            try:
+                self.app = FaceAnalysis(root="/tmp/.insightface", providers=['CPUExecutionProvider'])
+                self.app.prepare(ctx_id=-1)  # -1 = CPU
+                print("✅ CPU initialization successful (ctx_id=-1)")
+                self.gpu_available = False
+            except Exception as cpu_error:
+                print(f"❌ CPU initialization also failed: {cpu_error}")
+                raise
+        
+        print("="*80 + "\n")
         
         print("=" * 80)
         print("InsightFace Providers:")
@@ -70,7 +93,8 @@ class EndpointHandler:
         """Pre-warm GPU and compile ONNX models on startup to eliminate cold-start latency."""
         try:
             print("\n" + "="*80)
-            print("PRE-WARMING GPU AND COMPILING MODELS")
+            mode = "GPU" if self.gpu_available else "CPU"
+            print(f"PRE-WARMING MODELS ({mode} MODE)")
             print("="*80)
             start = time.time()
             
@@ -81,7 +105,7 @@ class EndpointHandler:
             _ = self.app.get(dummy_img)
             
             elapsed = time.time() - start
-            print(f"✅ Models pre-warmed in {elapsed:.2f}s")
+            print(f"✅ Models pre-warmed in {elapsed:.2f}s ({mode})")
             print("="*80 + "\n")
         except Exception as e:
             print(f"Warning: Model pre-warming failed (non-fatal): {e}")