Avoid using uncessary get_values(MODEL_MAPPING) (#29362)

* more fixes

* more fixes

---------

Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>

tests/models/beit/test_modeling_beit.py

@@ -21,7 +21,6 @@ from datasets import load_dataset
 from packaging import version
 
 from transformers import BeitConfig
-from transformers.models.auto import get_values
 from transformers.testing_utils import require_torch, require_torch_multi_gpu, require_vision, slow, torch_device
 from transformers.utils import cached_property, is_torch_available, is_vision_available
 
@@ -36,14 +35,13 @@ if is_torch_available():
     from torch import nn
 
     from transformers import (
-        MODEL_FOR_BACKBONE_MAPPING,
-        MODEL_MAPPING,
         BeitBackbone,
         BeitForImageClassification,
         BeitForMaskedImageModeling,
         BeitForSemanticSegmentation,
         BeitModel,
     )
+    from transformers.models.auto.modeling_auto import MODEL_FOR_BACKBONE_MAPPING_NAMES, MODEL_MAPPING_NAMES
     from transformers.models.beit.modeling_beit import BEIT_PRETRAINED_MODEL_ARCHIVE_LIST
 
 
@@ -312,10 +310,10 @@ class BeitModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
 
         for model_class in self.all_model_classes:
             # we don't test BeitForMaskedImageModeling
-            if model_class in [
-                *get_values(MODEL_MAPPING),
-                *get_values(MODEL_FOR_BACKBONE_MAPPING),
-                BeitForMaskedImageModeling,
+            if model_class.__name__ in [
+                *MODEL_MAPPING_NAMES.values(),
+                *MODEL_FOR_BACKBONE_MAPPING_NAMES.values(),
+                "BeitForMaskedImageModeling",
             ]:
                 continue
 
@@ -337,8 +335,12 @@ class BeitModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
         for model_class in self.all_model_classes:
             # we don't test BeitForMaskedImageModeling
             if (
-                model_class
-                in [*get_values(MODEL_MAPPING), *get_values(MODEL_FOR_BACKBONE_MAPPING), BeitForMaskedImageModeling]
+                model_class.__name__
+                in [
+                    *MODEL_MAPPING_NAMES.values(),
+                    *MODEL_FOR_BACKBONE_MAPPING_NAMES.values(),
+                    "BeitForMaskedImageModeling",
+                ]
                 or not model_class.supports_gradient_checkpointing
             ):
                 continue