diff --git a/optimum/intel/ipex/modeling_base.py b/optimum/intel/ipex/modeling_base.py
index 8611bddd2..d8f830e51 100644
--- a/optimum/intel/ipex/modeling_base.py
+++ b/optimum/intel/ipex/modeling_base.py
@@ -276,6 +276,8 @@ def forward(
         attention_mask: Optional[torch.FloatTensor] = None,
         **kwargs,
     ) -> CausalLMOutputWithPast:
+        if self.add_patch and input_ids is not None and attention_mask is None:
+            attention_mask = torch.ones_like(input_ids)
         return self.model(input_ids=input_ids, attention_mask=attention_mask, **kwargs)
 
     def _prepare_generation_config(