try to fix input dimentions

NN/models/enhanced_cnn.py

@@ -498,10 +498,20 @@ class EnhancedCNN(nn.Module):
         """Enhanced action selection with ultra massive model predictions"""
         if explore and np.random.random() < 0.1:  # 10% random exploration
             return np.random.choice(self.n_actions)
-        
+ 
         self.eval()
-        state_tensor = torch.FloatTensor(state).unsqueeze(0).to(self.device)
-        
+
+        # Accept both NumPy arrays and already-built torch tensors
+        if isinstance(state, torch.Tensor):
+            state_tensor = state.detach().to(self.device)
+            if state_tensor.dim() == 1:
+                state_tensor = state_tensor.unsqueeze(0)
+        else:
+            # Convert to tensor **directly on the target device** to avoid intermediate CPU copies
+            state_tensor = torch.as_tensor(state, dtype=torch.float32, device=self.device)
+            if state_tensor.dim() == 1:
+                state_tensor = state_tensor.unsqueeze(0)
+ 
         with torch.no_grad():
             q_values, extrema_pred, price_predictions, features, advanced_predictions = self(state_tensor)