Create safe fallback for models not yet initialized with masking_type

llama_diffusion_model.py

@@ -120,6 +120,8 @@ class CustomTransformerModel(PreTrainedModel):
 
         # Build attention mask
         device = input_ids.device
+        
+        masking_type = getattr(self.config, "masking_type", "bidirectional_masked")
         if self.config.masking_type == 'bidirectional':
             base_mask = torch.ones(seq_len, seq_len, dtype=torch.bool, device=device)
         elif self.config.masking_type == 'bidirectional_masked':