fixed model training

NN/models/advanced_transformer_trading.py

@@ -1229,7 +1229,7 @@ class TradingTransformerTrainer:
                     for k, v in batch.items()}
             
             # Use automatic mixed precision (FP16) for memory efficiency
-            with torch.cuda.amp.autocast(enabled=self.use_amp):
+            with torch.amp.autocast('cuda', enabled=self.use_amp):
                 # Forward pass with multi-timeframe data
                 outputs = self.model(
                     price_data_1s=batch.get('price_data_1s'),

NN/models/cob_rl_model.py

@@ -323,7 +323,7 @@ class COBRLModelInterface(ModelInterface):
         self.optimizer.zero_grad()
         
         if self.scaler:
-            with torch.cuda.amp.autocast():
+            with torch.amp.autocast('cuda'):
                 outputs = self.model(features)
                 loss = self._calculate_loss(outputs, targets)
             

NN/models/dqn_agent.py

@@ -1436,7 +1436,7 @@ class DQNAgent:
             import warnings
             with warnings.catch_warnings():
                 warnings.simplefilter("ignore", FutureWarning)
-                with torch.cuda.amp.autocast():
+                with torch.amp.autocast('cuda'):
                     # Get current Q values and predictions
                     current_q_values, current_extrema_pred, current_price_pred, hidden_features, current_advanced_pred = self._safe_cnn_forward(self.policy_net, states)
                     current_q_values = current_q_values.gather(1, actions.unsqueeze(1)).squeeze(1)