fix prediction candles updates. fix for trend prediction.

NN/models/advanced_transformer_trading.py

@@ -61,7 +61,7 @@ class TradingTransformerConfig:
     use_layer_norm_variants: bool = True # Advanced normalization
     
     # Memory optimization
-    use_gradient_checkpointing: bool = True  # Trade compute for memory (saves ~30% memory)
+    use_gradient_checkpointing: bool = False  # DISABLED: Causes tensor shape mismatches during backward pass
 
 class PositionalEncoding(nn.Module):
     """Sinusoidal positional encoding for transformer"""