Improve GPU detection and fallback logic

app.py

@@ -617,16 +617,31 @@ def create_interface():
         - **Example**: Use from your frontend with `fetch()` or `axios()`
         """)
         
-        if HAS_SPACES and torch.cuda.is_available():
-            try:
-                predict_fn = spaces.GPU(predict_volume)
-                print("Using GPU acceleration")
-            except Exception as e:
-                print(f"GPU decorator failed: {e}. Using CPU fallback.")
+        use_gpu = False
+        if HAS_SPACES:
+            if torch.cuda.is_available():
+                try:
+                    test_tensor = torch.zeros(1).cuda()
+                    del test_tensor
+                    torch.cuda.empty_cache()
+                    use_gpu = True
+                    predict_fn = spaces.GPU(predict_volume)
+                    print("✓ Using GPU acceleration (ZeroGPU/T4)")
+                except Exception as e:
+                    print(f"⚠ GPU available but failed to initialize: {e}")
+                    print("Falling back to CPU mode")
+                    use_gpu = False
+                    predict_fn = predict_volume
+            else:
+                print("ℹ No GPU available. Running on CPU (slower but free)")
                 predict_fn = predict_volume
         else:
+            if torch.cuda.is_available():
+                print("✓ CUDA available, using GPU")
+                use_gpu = True
+            else:
+                print("ℹ Running on CPU")
             predict_fn = predict_volume
-            print("Running on CPU (GPU not available or quota exhausted)")
         
         predict_btn.click(
             fn=predict_fn,