diff --git a/model/sequencer.py b/model/sequencer.py
index a70fcb6..a04e5fc 100644
--- a/model/sequencer.py
+++ b/model/sequencer.py
@@ -158,8 +158,12 @@ def pad_token_ids(self, token_ids: Tensor, pad_id: int) -> Tensor:
         """
 
         pad_size = self.window_size - token_ids.size(1)
-        padding = full((1, pad_size), pad_id, dtype=long_)
-        return cat([token_ids, padding.to(device=self.device)], dim=1)
+        if pad_size > 0:
+            padding = full((1, pad_size), pad_id, dtype=long_, device=self.device)
+            token_ids = cat([token_ids, padding], dim=1)
+        elif pad_size < 0:
+            token_ids = token_ids[:, -self.window_size:]
+        return token_ids
 
 
     def generate_text(self, tokens: List[str]) -> str: