update text-only input bugs

modeling_imp.py

@@ -1237,6 +1237,8 @@ class LlavaMetaForCausalLM(ABC):
     ):
         vision_tower = self.get_vision_tower()
         # if vision_tower is None or images is None or past_key_values.seqlen_offset != 0:
+        if images is None:
+            return input_ids, position_ids, attention_mask, past_key_values, None, labels
         if past_key_values is not None:
             target_shape = past_key_values[0][0].shape[2] + 1
             attention_mask = torch.ones(