diff --git a/fastembed/common/preprocessor_utils.py b/fastembed/common/preprocessor_utils.py
index 934b234a..2fbd9986 100644
--- a/fastembed/common/preprocessor_utils.py
+++ b/fastembed/common/preprocessor_utils.py
@@ -98,6 +98,7 @@ def load_tokenizer(model_dir: Path) -> tuple[Tokenizer, dict[str, int]]:
         pad_id=padding.get("pad_id", config.get("pad_token_id", 0)),
         pad_type_id=padding.get("pad_type_id", 0),
         pad_token=pad_token,
+        pad_to_multiple_of=padding.get("pad_to_multiple_of"),
         length=None,
     )
 
diff --git a/tests/test_preprocessor_utils.py b/tests/test_preprocessor_utils.py
index 5b899529..23cfc58c 100644
--- a/tests/test_preprocessor_utils.py
+++ b/tests/test_preprocessor_utils.py
@@ -138,6 +138,21 @@ def test_serialized_left_padding_is_preserved(make_model_dir) -> None:
     assert encoded[1].attention_mask[0] == 0
 
 
+def test_serialized_pad_to_multiple_of_is_preserved(make_model_dir) -> None:
+    """Everything the tokenizer declared is kept; only the fixed length is overridden."""
+    model_dir = make_model_dir(
+        padding={"length": 128, "pad_id": 0, "pad_token": "[PAD]", "pad_to_multiple_of": 8},
+    )
+
+    tokenizer, _ = load_tokenizer(model_dir)
+
+    assert tokenizer.padding["length"] is None
+    assert tokenizer.padding["pad_to_multiple_of"] == 8
+
+    encoded = tokenizer.encode_batch(["hello world", "hello"])
+    assert len(encoded[0].ids) % 8 == 0
+
+
 def test_serialized_pad_id_takes_precedence_over_config(make_model_dir) -> None:
     mask_pad_id = 103  # [MASK] in the bert-base vocab, any id other than the config's works
     model_dir = make_model_dir(