Make sure dynamic objects can be saved and reloaded (#21008)

* Make sure dynamic objects can be saved and reloaded

* Remove processor test

src/transformers/models/auto/auto_factory.py

@@ -455,6 +455,7 @@ class _BaseAutoModelClass:
             model_class = get_class_from_dynamic_module(
                 pretrained_model_name_or_path, module_file + ".py", class_name, **hub_kwargs, **kwargs
             )
+            model_class.register_for_auto_class(cls.__name__)
             return model_class.from_pretrained(
                 pretrained_model_name_or_path, *model_args, config=config, **hub_kwargs, **kwargs
             )

src/transformers/models/auto/configuration_auto.py

@@ -853,6 +853,7 @@ class AutoConfig:
             config_class = get_class_from_dynamic_module(
                 pretrained_model_name_or_path, module_file + ".py", class_name, **kwargs
             )
+            config_class.register_for_auto_class()
             return config_class.from_pretrained(pretrained_model_name_or_path, **kwargs)
         elif "model_type" in config_dict:
             config_class = CONFIG_MAPPING[config_dict["model_type"]]

src/transformers/models/auto/feature_extraction_auto.py

@@ -340,6 +340,7 @@ class AutoFeatureExtractor:
                 feature_extractor_class = get_class_from_dynamic_module(
                     pretrained_model_name_or_path, module_file + ".py", class_name, **kwargs
                 )
+                feature_extractor_class.register_for_auto_class()
             else:
                 feature_extractor_class = feature_extractor_class_from_name(feature_extractor_class)
 

src/transformers/models/auto/image_processing_auto.py

@@ -352,6 +352,7 @@ class AutoImageProcessor:
                 image_processor_class = get_class_from_dynamic_module(
                     pretrained_model_name_or_path, module_file + ".py", class_name, **kwargs
                 )
+                image_processor_class.register_for_auto_class()
             else:
                 image_processor_class = image_processor_class_from_name(image_processor_class)
 

src/transformers/models/auto/processing_auto.py

@@ -256,6 +256,7 @@ class AutoProcessor:
                 processor_class = get_class_from_dynamic_module(
                     pretrained_model_name_or_path, module_file + ".py", class_name, **kwargs
                 )
+                processor_class.register_for_auto_class()
             else:
                 processor_class = processor_class_from_name(processor_class)
 

src/transformers/models/auto/tokenization_auto.py

@@ -641,6 +641,7 @@ class AutoTokenizer:
                 tokenizer_class = get_class_from_dynamic_module(
                     pretrained_model_name_or_path, module_file + ".py", class_name, **kwargs
                 )
+                tokenizer_class.register_for_auto_class()
 
             elif use_fast and not config_tokenizer_class.endswith("Fast"):
                 tokenizer_class_candidate = f"{config_tokenizer_class}Fast"

tests/models/auto/test_configuration_auto.py

@@ -110,3 +110,9 @@ class AutoConfigTest(unittest.TestCase):
     def test_from_pretrained_dynamic_config(self):
         config = AutoConfig.from_pretrained("hf-internal-testing/test_dynamic_model", trust_remote_code=True)
         self.assertEqual(config.__class__.__name__, "NewModelConfig")
+
+        # Test config can be reloaded.
+        with tempfile.TemporaryDirectory() as tmp_dir:
+            config.save_pretrained(tmp_dir)
+            reloaded_config = AutoConfig.from_pretrained(tmp_dir, trust_remote_code=True)
+        self.assertEqual(reloaded_config.__class__.__name__, "NewModelConfig")

tests/models/auto/test_feature_extraction_auto.py

@@ -96,10 +96,16 @@ class AutoFeatureExtractorTest(unittest.TestCase):
             _ = AutoFeatureExtractor.from_pretrained("hf-internal-testing/config-no-model")
 
     def test_from_pretrained_dynamic_feature_extractor(self):
-        model = AutoFeatureExtractor.from_pretrained(
+        feature_extractor = AutoFeatureExtractor.from_pretrained(
             "hf-internal-testing/test_dynamic_feature_extractor", trust_remote_code=True
         )
-        self.assertEqual(model.__class__.__name__, "NewFeatureExtractor")
+        self.assertEqual(feature_extractor.__class__.__name__, "NewFeatureExtractor")
+
+        # Test feature extractor can be reloaded.
+        with tempfile.TemporaryDirectory() as tmp_dir:
+            feature_extractor.save_pretrained(tmp_dir)
+            reloaded_feature_extractor = AutoFeatureExtractor.from_pretrained(tmp_dir, trust_remote_code=True)
+        self.assertEqual(reloaded_feature_extractor.__class__.__name__, "NewFeatureExtractor")
 
     def test_new_feature_extractor_registration(self):
         try:

tests/models/auto/test_image_processing_auto.py

@@ -130,10 +130,16 @@ class AutoImageProcessorTest(unittest.TestCase):
             _ = AutoImageProcessor.from_pretrained("hf-internal-testing/config-no-model")
 
     def test_from_pretrained_dynamic_image_processor(self):
-        model = AutoImageProcessor.from_pretrained(
+        image_processor = AutoImageProcessor.from_pretrained(
             "hf-internal-testing/test_dynamic_image_processor", trust_remote_code=True
         )
-        self.assertEqual(model.__class__.__name__, "NewImageProcessor")
+        self.assertEqual(image_processor.__class__.__name__, "NewImageProcessor")
+
+        # Test image processor can be reloaded.
+        with tempfile.TemporaryDirectory() as tmp_dir:
+            image_processor.save_pretrained(tmp_dir)
+            reloaded_image_processor = AutoImageProcessor.from_pretrained(tmp_dir, trust_remote_code=True)
+        self.assertEqual(reloaded_image_processor.__class__.__name__, "NewImageProcessor")
 
     def test_new_image_processor_registration(self):
         try:

tests/models/auto/test_modeling_auto.py

@@ -276,10 +276,28 @@ class AutoModelTest(unittest.TestCase):
         model = AutoModel.from_pretrained("hf-internal-testing/test_dynamic_model", trust_remote_code=True)
         self.assertEqual(model.__class__.__name__, "NewModel")
 
+        # Test model can be reloaded.
+        with tempfile.TemporaryDirectory() as tmp_dir:
+            model.save_pretrained(tmp_dir)
+            reloaded_model = AutoModel.from_pretrained(tmp_dir, trust_remote_code=True)
+
+        self.assertEqual(reloaded_model.__class__.__name__, "NewModel")
+        for p1, p2 in zip(model.parameters(), reloaded_model.parameters()):
+            self.assertTrue(torch.equal(p1, p2))
+
         # This one uses a relative import to a util file, this checks it is downloaded and used properly.
         model = AutoModel.from_pretrained("hf-internal-testing/test_dynamic_model_with_util", trust_remote_code=True)
         self.assertEqual(model.__class__.__name__, "NewModel")
 
+        # Test model can be reloaded.
+        with tempfile.TemporaryDirectory() as tmp_dir:
+            model.save_pretrained(tmp_dir)
+            reloaded_model = AutoModel.from_pretrained(tmp_dir, trust_remote_code=True)
+
+        self.assertEqual(reloaded_model.__class__.__name__, "NewModel")
+        for p1, p2 in zip(model.parameters(), reloaded_model.parameters()):
+            self.assertTrue(torch.equal(p1, p2))
+
     def test_new_model_registration(self):
         AutoConfig.register("custom", CustomConfig)
 

tests/models/auto/test_processor_auto.py

@@ -157,12 +157,12 @@ class AutoFeatureExtractorTest(unittest.TestCase):
             self.assertEqual(tokenizer.__class__.__name__, "NewTokenizerFast")
 
             # Test we can also load the slow version
-            processor = AutoProcessor.from_pretrained(
+            new_processor = AutoProcessor.from_pretrained(
                 "hf-internal-testing/test_dynamic_processor", trust_remote_code=True, use_fast=False
             )
-            tokenizer = processor.tokenizer
+            new_tokenizer = new_processor.tokenizer
-            self.assertTrue(tokenizer.special_attribute_present)
+            self.assertTrue(new_tokenizer.special_attribute_present)
-            self.assertEqual(tokenizer.__class__.__name__, "NewTokenizer")
+            self.assertEqual(new_tokenizer.__class__.__name__, "NewTokenizer")
         else:
             self.assertEqual(tokenizer.__class__.__name__, "NewTokenizer")
 

tests/models/auto/test_tokenization_auto.py

@@ -302,8 +302,15 @@ class AutoTokenizerTest(unittest.TestCase):
     def test_from_pretrained_dynamic_tokenizer(self):
         tokenizer = AutoTokenizer.from_pretrained("hf-internal-testing/test_dynamic_tokenizer", trust_remote_code=True)
         self.assertTrue(tokenizer.special_attribute_present)
+        # Test tokenizer can be reloaded.
+        with tempfile.TemporaryDirectory() as tmp_dir:
+            tokenizer.save_pretrained(tmp_dir)
+            reloaded_tokenizer = AutoTokenizer.from_pretrained(tmp_dir, trust_remote_code=True)
+        self.assertTrue(reloaded_tokenizer.special_attribute_present)
+
         if is_tokenizers_available():
             self.assertEqual(tokenizer.__class__.__name__, "NewTokenizerFast")
+            self.assertEqual(reloaded_tokenizer.__class__.__name__, "NewTokenizerFast")
 
             # Test we can also load the slow version
             tokenizer = AutoTokenizer.from_pretrained(
@@ -311,8 +318,15 @@ class AutoTokenizerTest(unittest.TestCase):
             )
             self.assertTrue(tokenizer.special_attribute_present)
             self.assertEqual(tokenizer.__class__.__name__, "NewTokenizer")
+            # Test tokenizer can be reloaded.
+            with tempfile.TemporaryDirectory() as tmp_dir:
+                tokenizer.save_pretrained(tmp_dir)
+                reloaded_tokenizer = AutoTokenizer.from_pretrained(tmp_dir, trust_remote_code=True, use_fast=False)
+            self.assertEqual(reloaded_tokenizer.__class__.__name__, "NewTokenizer")
+            self.assertTrue(reloaded_tokenizer.special_attribute_present)
         else:
             self.assertEqual(tokenizer.__class__.__name__, "NewTokenizer")
+            self.assertEqual(reloaded_tokenizer.__class__.__name__, "NewTokenizer")
 
     def test_from_pretrained_dynamic_tokenizer_legacy_format(self):
         tokenizer = AutoTokenizer.from_pretrained(