Fix marian tokenizer save pretrained (#5043)

src/transformers/tokenization_marian.py

@@ -40,9 +40,9 @@ class MarianTokenizer(PreTrainedTokenizer):
 
     def __init__(
         self,
-        vocab=None,
-        source_spm=None,
-        target_spm=None,
+        vocab,
+        source_spm,
+        target_spm,
         source_lang=None,
         target_lang=None,
         unk_token="<unk>",
@@ -59,6 +59,7 @@ class MarianTokenizer(PreTrainedTokenizer):
             pad_token=pad_token,
             **kwargs,
         )
+        assert Path(source_spm).exists(), f"cannot find spm source {source_spm}"
         self.encoder = load_json(vocab)
         if self.unk_token not in self.encoder:
             raise KeyError("<unk> token must be in vocab")
@@ -179,10 +180,11 @@ class MarianTokenizer(PreTrainedTokenizer):
         assert save_dir.is_dir(), f"{save_directory} should be a directory"
         save_json(self.encoder, save_dir / self.vocab_files_names["vocab"])
 
-        for f in self.spm_files:
+        for orig, f in zip(["source.spm", "target.spm"], self.spm_files):
             dest_path = save_dir / Path(f).name
             if not dest_path.exists():
-                copyfile(f, save_dir / Path(f).name)
+                copyfile(f, save_dir / orig)
+
         return tuple(save_dir / f for f in self.vocab_files_names)
 
     def get_vocab(self) -> Dict:

tests/test_tokenization_marian.py

@@ -15,6 +15,7 @@
 
 
 import os
+import tempfile
 import unittest
 from pathlib import Path
 from shutil import copyfile
@@ -23,7 +24,6 @@ from transformers.tokenization_marian import MarianTokenizer, save_json, vocab_f
 from transformers.tokenization_utils import BatchEncoding
 
 from .test_tokenization_common import TokenizerTesterMixin
-from .utils import slow
 
 
 SAMPLE_SP = os.path.join(os.path.dirname(os.path.abspath(__file__)), "fixtures/test_sentencepiece.model")
@@ -60,10 +60,15 @@ class MarianTokenizationTest(TokenizerTesterMixin, unittest.TestCase):
             "This is a test",
         )
 
-    @slow
     def test_tokenizer_equivalence_en_de(self):
         en_de_tokenizer = MarianTokenizer.from_pretrained(f"{ORG_NAME}opus-mt-en-de")
         batch = en_de_tokenizer.prepare_translation_batch(["I am a small frog"], return_tensors=None)
         self.assertIsInstance(batch, BatchEncoding)
         expected = [38, 121, 14, 697, 38848, 0]
         self.assertListEqual(expected, batch.input_ids[0])
+
+        save_dir = tempfile.mkdtemp()
+        en_de_tokenizer.save_pretrained(save_dir)
+        contents = [x.name for x in Path(save_dir).glob("*")]
+        self.assertIn("source.spm", contents)
+        MarianTokenizer.from_pretrained(save_dir)