Set to bidirectional for debugging

llama_diffusion_model.py

@@ -122,7 +122,7 @@ class CustomTransformerModel(PreTrainedModel):
         # Build attention mask
         device = input_ids.device
         
-        masking_type = getattr(self.config, "masking_type", "unidirectional")
+        masking_type = getattr(self.config, "masking_type", "bidirectional")
         if masking_type == 'bidirectional':
             base_mask = torch.ones(seq_len, seq_len, dtype=torch.bool, device=device)
         elif masking_type == 'bidirectional_masked':