🔥Rework pipeline testing by removing `PipelineTestCaseMeta` 🚀 (#21516)

* Add PipelineTesterMixin

* remove class PipelineTestCaseMeta

* move validate_test_components

* Add for ViT

* Add to SPECIAL_MODULE_TO_TEST_MAP

* style and quality

* Add feature-extraction

* update

* raise instead of skip

* add tiny_model_summary.json

* more explicit

* skip tasks not in mapping

* add availability check

* Add Copyright

* A way to diable irrelevant tests

* update with main

* remove disable_irrelevant_tests

* skip tests

* better skip message

* better skip message

* Add all pipeline task tests

* revert

* Import PipelineTesterMixin

* subclass test classes with PipelineTesterMixin

* Add pipieline_model_mapping

* Fix import after adding pipieline_model_mapping

* Fix style and quality after adding pipieline_model_mapping

* Fix one more import after adding pipieline_model_mapping

* Fix style and quality after adding pipieline_model_mapping

* Fix test issues

* Fix import requirements

* Fix mapping for MobileViTModelTest

* Update

* Better skip message

* pipieline_model_mapping could not be None

* Remove some PipelineTesterMixin

* Fix typo

* revert tests_fetcher.py

* update

* rename

* revert

* Remove PipelineTestCaseMeta from ZeroShotAudioClassificationPipelineTests

* style and quality

* test fetcher for all pipeline/model tests

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

tests/models/transfo_xl/test_modeling_tf_transfo_xl.py

@@ -22,6 +22,7 @@ from transformers.testing_utils import require_tf, slow
 
 from ...test_configuration_common import ConfigTester
 from ...test_modeling_tf_common import TFModelTesterMixin, ids_tensor
+from ...test_pipeline_mixin import PipelineTesterMixin
 
 
 if is_tf_available():
@@ -155,11 +156,21 @@ class TFTransfoXLModelTester:
 
 
 @require_tf
-class TFTransfoXLModelTest(TFModelTesterMixin, unittest.TestCase):
+class TFTransfoXLModelTest(TFModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
     all_model_classes = (
         (TFTransfoXLModel, TFTransfoXLLMHeadModel, TFTransfoXLForSequenceClassification) if is_tf_available() else ()
     )
     all_generative_model_classes = () if is_tf_available() else ()
+    pipeline_model_mapping = (
+        {
+            "feature-extraction": TFTransfoXLModel,
+            "text-classification": TFTransfoXLForSequenceClassification,
+            "text-generation": TFTransfoXLLMHeadModel,
+            "zero-shot": TFTransfoXLForSequenceClassification,
+        }
+        if is_tf_available()
+        else {}
+    )
     # TODO: add this test when TFTransfoXLLMHead has a linear output layer implemented
     test_resize_embeddings = False
     test_head_masking = False

tests/models/transfo_xl/test_modeling_transfo_xl.py

@@ -23,6 +23,7 @@ from transformers.testing_utils import require_torch, require_torch_multi_gpu, s
 from ...generation.test_utils import GenerationTesterMixin
 from ...test_configuration_common import ConfigTester
 from ...test_modeling_common import ModelTesterMixin, ids_tensor
+from ...test_pipeline_mixin import PipelineTesterMixin
 
 
 if is_torch_available():
@@ -251,11 +252,21 @@ class TransfoXLModelTester:
 
 
 @require_torch
-class TransfoXLModelTest(ModelTesterMixin, GenerationTesterMixin, unittest.TestCase):
+class TransfoXLModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixin, unittest.TestCase):
     all_model_classes = (
         (TransfoXLModel, TransfoXLLMHeadModel, TransfoXLForSequenceClassification) if is_torch_available() else ()
     )
     all_generative_model_classes = (TransfoXLLMHeadModel,) if is_torch_available() else ()
+    pipeline_model_mapping = (
+        {
+            "feature-extraction": TransfoXLModel,
+            "text-classification": TransfoXLForSequenceClassification,
+            "text-generation": TransfoXLLMHeadModel,
+            "zero-shot": TransfoXLForSequenceClassification,
+        }
+        if is_torch_available()
+        else {}
+    )
     test_pruning = False
     test_resize_embeddings = True
     test_mismatched_shapes = False

tests/models/trocr/test_modeling_trocr.py

@@ -22,6 +22,7 @@ from transformers.testing_utils import is_torch_available, require_torch, torch_
 from ...generation.test_utils import GenerationTesterMixin
 from ...test_configuration_common import ConfigTester
 from ...test_modeling_common import ModelTesterMixin, ids_tensor
+from ...test_pipeline_mixin import PipelineTesterMixin
 
 
 if is_torch_available():
@@ -158,9 +159,10 @@ class TrOCRStandaloneDecoderModelTester:
 
 
 @require_torch
-class TrOCRStandaloneDecoderModelTest(ModelTesterMixin, GenerationTesterMixin, unittest.TestCase):
+class TrOCRStandaloneDecoderModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixin, unittest.TestCase):
     all_model_classes = (TrOCRDecoder, TrOCRForCausalLM) if is_torch_available() else ()
     all_generative_model_classes = (TrOCRForCausalLM,) if is_torch_available() else ()
+    pipeline_model_mapping = {"text-generation": TrOCRForCausalLM} if is_torch_available() else {}
     fx_compatible = True
     test_pruning = False
 

tests/models/tvlt/test_modeling_tvlt.py

@@ -33,6 +33,7 @@ from transformers.utils import cached_property
 
 from ...test_configuration_common import ConfigTester
 from ...test_modeling_common import ModelTesterMixin, floats_tensor
+from ...test_pipeline_mixin import PipelineTesterMixin
 
 
 if is_torch_available():
@@ -322,10 +323,11 @@ class TvltModelTester:
 
 @require_torch
 @unittest.skipIf(not is_torch_greater_or_equal_than_1_10, "TVLT is only available in torch v1.10+")
-class TvltModelTest(ModelTesterMixin, unittest.TestCase):
+class TvltModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
     all_model_classes = (
         (TvltModel, TvltForPreTraining, TvltForAudioVisualClassification) if is_torch_available() else ()
     )
+    pipeline_model_mapping = {"feature-extraction": TvltModel} if is_torch_available() else {}
 
     fx_compatible = False
     test_pruning = False

tests/models/unispeech/test_modeling_unispeech.py

@@ -32,6 +32,7 @@ from ...test_modeling_common import (
     ids_tensor,
     random_attention_mask,
 )
+from ...test_pipeline_mixin import PipelineTesterMixin
 
 
 if is_torch_available():
@@ -297,12 +298,21 @@ class UniSpeechModelTester:
 
 
 @require_torch
-class UniSpeechRobustModelTest(ModelTesterMixin, unittest.TestCase):
+class UniSpeechRobustModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
     all_model_classes = (
         (UniSpeechForCTC, UniSpeechModel, UniSpeechForSequenceClassification, UniSpeechForPreTraining)
         if is_torch_available()
         else ()
     )
+    pipeline_model_mapping = (
+        {
+            "audio-classification": UniSpeechForSequenceClassification,
+            "automatic-speech-recognition": UniSpeechForCTC,
+            "feature-extraction": UniSpeechModel,
+        }
+        if is_torch_available()
+        else {}
+    )
     test_pruning = False
     test_headmasking = False
 

tests/models/unispeech_sat/test_modeling_unispeech_sat.py

@@ -32,6 +32,7 @@ from ...test_modeling_common import (
     ids_tensor,
     random_attention_mask,
 )
+from ...test_pipeline_mixin import PipelineTesterMixin
 
 
 if is_torch_available():
@@ -341,7 +342,7 @@ class UniSpeechSatModelTester:
 
 
 @require_torch
-class UniSpeechSatModelTest(ModelTesterMixin, unittest.TestCase):
+class UniSpeechSatModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
     all_model_classes = (
         (
             UniSpeechSatForCTC,
@@ -354,6 +355,15 @@ class UniSpeechSatModelTest(ModelTesterMixin, unittest.TestCase):
         if is_torch_available()
         else ()
     )
+    pipeline_model_mapping = (
+        {
+            "audio-classification": UniSpeechSatForSequenceClassification,
+            "automatic-speech-recognition": UniSpeechSatForCTC,
+            "feature-extraction": UniSpeechSatModel,
+        }
+        if is_torch_available()
+        else {}
+    )
     test_pruning = False
     test_headmasking = False
     test_torchscript = False

tests/models/upernet/test_modeling_upernet.py

@@ -26,6 +26,7 @@ from transformers.utils import is_torch_available, is_vision_available
 
 from ...test_configuration_common import ConfigTester
 from ...test_modeling_common import ModelTesterMixin, _config_zero_init, floats_tensor, ids_tensor
+from ...test_pipeline_mixin import PipelineTesterMixin
 
 
 if is_torch_available():
@@ -138,13 +139,14 @@ class UperNetModelTester:
 
 
 @require_torch
-class UperNetModelTest(ModelTesterMixin, unittest.TestCase):
+class UperNetModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
     """
     Here we also overwrite some of the tests of test_modeling_common.py, as UperNet does not use input_ids, inputs_embeds,
     attention_mask and seq_length.
     """
 
     all_model_classes = (UperNetForSemanticSegmentation,) if is_torch_available() else ()
+    pipeline_model_mapping = {"image-segmentation": UperNetForSemanticSegmentation} if is_torch_available() else {}
     fx_compatible = False
     test_pruning = False
     test_resize_embeddings = False

tests/models/van/test_modeling_van.py

@@ -25,6 +25,7 @@ from transformers.utils import cached_property, is_scipy_available, is_torch_ava
 
 from ...test_configuration_common import ConfigTester
 from ...test_modeling_common import ModelTesterMixin, _config_zero_init, floats_tensor, ids_tensor
+from ...test_pipeline_mixin import PipelineTesterMixin
 
 
 if is_scipy_available():
@@ -115,13 +116,18 @@ class VanModelTester:
 
 
 @require_torch
-class VanModelTest(ModelTesterMixin, unittest.TestCase):
+class VanModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
     """
     Here we also overwrite some of the tests of test_modeling_common.py, as Van does not use input_ids, inputs_embeds,
     attention_mask and seq_length.
     """
 
     all_model_classes = (VanModel, VanForImageClassification) if is_torch_available() else ()
+    pipeline_model_mapping = (
+        {"feature-extraction": VanModel, "image-classification": VanForImageClassification}
+        if is_torch_available()
+        else {}
+    )
 
     test_pruning = False
     test_resize_embeddings = False

tests/models/videomae/test_modeling_videomae.py

@@ -29,6 +29,7 @@ from transformers.utils import cached_property, is_torch_available, is_vision_av
 
 from ...test_configuration_common import ConfigTester
 from ...test_modeling_common import ModelTesterMixin, floats_tensor, ids_tensor
+from ...test_pipeline_mixin import PipelineTesterMixin
 
 
 if is_torch_available():
@@ -162,7 +163,7 @@ class VideoMAEModelTester:
 
 
 @require_torch
-class VideoMAEModelTest(ModelTesterMixin, unittest.TestCase):
+class VideoMAEModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
     """
     Here we also overwrite some of the tests of test_modeling_common.py, as VideoMAE does not use input_ids, inputs_embeds,
     attention_mask and seq_length.
@@ -171,6 +172,11 @@ class VideoMAEModelTest(ModelTesterMixin, unittest.TestCase):
     all_model_classes = (
         (VideoMAEModel, VideoMAEForPreTraining, VideoMAEForVideoClassification) if is_torch_available() else ()
     )
+    pipeline_model_mapping = (
+        {"feature-extraction": VideoMAEModel, "video-classification": VideoMAEForVideoClassification}
+        if is_torch_available()
+        else {}
+    )
 
     test_pruning = False
     test_torchscript = False

tests/models/vilt/test_modeling_vilt.py

@@ -26,6 +26,7 @@ from transformers.utils import cached_property
 
 from ...test_configuration_common import ConfigTester
 from ...test_modeling_common import ModelTesterMixin, floats_tensor, ids_tensor, random_attention_mask
+from ...test_pipeline_mixin import PipelineTesterMixin
 
 
 if is_torch_available():
@@ -218,7 +219,7 @@ class ViltModelTester:
 
 @require_torch
 @unittest.skipIf(not is_torch_greater_or_equal_than_1_10, "Vilt is only available in torch v1.10+")
-class ViltModelTest(ModelTesterMixin, unittest.TestCase):
+class ViltModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
     all_model_classes = (
         (
             ViltModel,
@@ -230,6 +231,11 @@ class ViltModelTest(ModelTesterMixin, unittest.TestCase):
         if is_torch_available()
         else ()
     )
+    pipeline_model_mapping = (
+        {"feature-extraction": ViltModel, "visual-question-answering": ViltForQuestionAnswering}
+        if is_torch_available()
+        else {}
+    )
     test_pruning = False
     test_headmasking = False
     test_torchscript = False

tests/models/visual_bert/test_modeling_visual_bert.py

@@ -22,6 +22,7 @@ from transformers.testing_utils import require_torch, slow, torch_device
 
 from ...test_configuration_common import ConfigTester
 from ...test_modeling_common import ModelTesterMixin, floats_tensor, ids_tensor
+from ...test_pipeline_mixin import PipelineTesterMixin
 
 
 if is_torch_available():
@@ -299,7 +300,7 @@ class VisualBertModelTester:
 
 
 @require_torch
-class VisualBertModelTest(ModelTesterMixin, unittest.TestCase):
+class VisualBertModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
     all_model_classes = (
         (
             VisualBertModel,
@@ -312,6 +313,7 @@ class VisualBertModelTest(ModelTesterMixin, unittest.TestCase):
         if is_torch_available()
         else ()
     )
+    pipeline_model_mapping = {"feature-extraction": VisualBertModel} if is_torch_available() else {}
     test_torchscript = False
     test_pruning = False
 

tests/models/vit/test_modeling_tf_vit.py

@@ -24,6 +24,7 @@ from transformers.utils import cached_property, is_tf_available, is_vision_avail
 
 from ...test_configuration_common import ConfigTester
 from ...test_modeling_tf_common import TFModelTesterMixin, floats_tensor, ids_tensor
+from ...test_pipeline_mixin import PipelineTesterMixin
 
 
 if is_tf_available():
@@ -148,13 +149,18 @@ class TFViTModelTester:
 
 
 @require_tf
-class TFViTModelTest(TFModelTesterMixin, unittest.TestCase):
+class TFViTModelTest(TFModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
     """
     Here we also overwrite some of the tests of test_modeling_tf_common.py, as ViT does not use input_ids, inputs_embeds,
     attention_mask and seq_length.
     """
 
     all_model_classes = (TFViTModel, TFViTForImageClassification) if is_tf_available() else ()
+    pipeline_model_mapping = (
+        {"feature-extraction": TFViTModel, "image-classification": TFViTForImageClassification}
+        if is_tf_available()
+        else {}
+    )
 
     test_resize_embeddings = False
     test_head_masking = False

tests/models/vit/test_modeling_vit.py

@@ -31,6 +31,7 @@ from transformers.utils import cached_property, is_torch_available, is_vision_av
 
 from ...test_configuration_common import ConfigTester
 from ...test_modeling_common import ModelTesterMixin, floats_tensor, ids_tensor
+from ...test_pipeline_mixin import PipelineTesterMixin
 
 
 if is_torch_available():
@@ -176,7 +177,7 @@ class ViTModelTester:
 
 
 @require_torch
-class ViTModelTest(ModelTesterMixin, unittest.TestCase):
+class ViTModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
     """
     Here we also overwrite some of the tests of test_modeling_common.py, as ViT does not use input_ids, inputs_embeds,
     attention_mask and seq_length.
@@ -191,6 +192,11 @@ class ViTModelTest(ModelTesterMixin, unittest.TestCase):
         if is_torch_available()
         else ()
     )
+    pipeline_model_mapping = (
+        {"feature-extraction": ViTModel, "image-classification": ViTForImageClassification}
+        if is_torch_available()
+        else {}
+    )
     fx_compatible = True
 
     test_pruning = False

tests/models/vit_hybrid/test_modeling_vit_hybrid.py

@@ -24,6 +24,7 @@ from transformers.utils import cached_property, is_torch_available, is_vision_av
 
 from ...test_configuration_common import ConfigTester
 from ...test_modeling_common import ModelTesterMixin, _config_zero_init, floats_tensor, ids_tensor
+from ...test_pipeline_mixin import PipelineTesterMixin
 
 
 if is_torch_available():
@@ -147,13 +148,18 @@ class ViTHybridModelTester:
 
 
 @require_torch
-class ViTHybridModelTest(ModelTesterMixin, unittest.TestCase):
+class ViTHybridModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
     """
     Here we also overwrite some of the tests of test_modeling_common.py, as ViT does not use input_ids, inputs_embeds,
     attention_mask and seq_length.
     """
 
     all_model_classes = (ViTHybridModel, ViTHybridForImageClassification) if is_torch_available() else ()
+    pipeline_model_mapping = (
+        {"feature-extraction": ViTHybridModel, "image-classification": ViTHybridForImageClassification}
+        if is_torch_available()
+        else {}
+    )
     test_pruning = False
     test_resize_embeddings = False
     test_head_masking = False

tests/models/vit_mae/test_modeling_tf_vit_mae.py

@@ -32,6 +32,7 @@ from transformers.testing_utils import require_tf, require_vision, slow
 
 from ...test_configuration_common import ConfigTester
 from ...test_modeling_tf_common import TFModelTesterMixin, floats_tensor, ids_tensor
+from ...test_pipeline_mixin import PipelineTesterMixin
 
 
 if is_tf_available():
@@ -155,13 +156,14 @@ class TFViTMAEModelTester:
 
 
 @require_tf
-class TFViTMAEModelTest(TFModelTesterMixin, unittest.TestCase):
+class TFViTMAEModelTest(TFModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
     """
     Here we also overwrite some of the tests of test_modeling_common.py, as ViTMAE does not use input_ids, inputs_embeds,
     attention_mask and seq_length.
     """
 
     all_model_classes = (TFViTMAEModel, TFViTMAEForPreTraining) if is_tf_available() else ()
+    pipeline_model_mapping = {"feature-extraction": TFViTMAEModel} if is_tf_available() else {}
 
     test_pruning = False
     test_onnx = False

tests/models/vit_mae/test_modeling_vit_mae.py

@@ -28,6 +28,7 @@ from transformers.utils import cached_property, is_torch_available, is_vision_av
 
 from ...test_configuration_common import ConfigTester
 from ...test_modeling_common import ModelTesterMixin, floats_tensor, ids_tensor
+from ...test_pipeline_mixin import PipelineTesterMixin
 
 
 if is_torch_available():
@@ -153,13 +154,14 @@ class ViTMAEModelTester:
 
 
 @require_torch
-class ViTMAEModelTest(ModelTesterMixin, unittest.TestCase):
+class ViTMAEModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
     """
     Here we also overwrite some of the tests of test_modeling_common.py, as ViTMAE does not use input_ids, inputs_embeds,
     attention_mask and seq_length.
     """
 
     all_model_classes = (ViTMAEModel, ViTMAEForPreTraining) if is_torch_available() else ()
+    pipeline_model_mapping = {"feature-extraction": ViTMAEModel} if is_torch_available() else {}
 
     test_pruning = False
     test_torchscript = False

tests/models/vit_msn/test_modeling_vit_msn.py

@@ -24,6 +24,7 @@ from transformers.utils import cached_property, is_torch_available, is_vision_av
 
 from ...test_configuration_common import ConfigTester
 from ...test_modeling_common import ModelTesterMixin, floats_tensor, ids_tensor
+from ...test_pipeline_mixin import PipelineTesterMixin
 
 
 if is_torch_available():
@@ -144,13 +145,18 @@ class ViTMSNModelTester:
 
 
 @require_torch
-class ViTMSNModelTest(ModelTesterMixin, unittest.TestCase):
+class ViTMSNModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
     """
     Here we also overwrite some of the tests of test_modeling_common.py, as ViTMSN does not use input_ids, inputs_embeds,
     attention_mask and seq_length.
     """
 
     all_model_classes = (ViTMSNModel, ViTMSNForImageClassification) if is_torch_available() else ()
+    pipeline_model_mapping = (
+        {"feature-extraction": ViTMSNModel, "image-classification": ViTMSNForImageClassification}
+        if is_torch_available()
+        else {}
+    )
 
     test_pruning = False
     test_torchscript = False

tests/models/wav2vec2/test_modeling_tf_wav2vec2.py

@@ -41,6 +41,7 @@ from transformers.utils import is_librosa_available, is_pyctcdecode_available
 
 from ...test_configuration_common import ConfigTester
 from ...test_modeling_tf_common import TFModelTesterMixin, ids_tensor
+from ...test_pipeline_mixin import PipelineTesterMixin
 
 
 if is_tf_available():
@@ -281,8 +282,9 @@ class TFWav2Vec2ModelTester:
 
 
 @require_tf
-class TFWav2Vec2ModelTest(TFModelTesterMixin, unittest.TestCase):
+class TFWav2Vec2ModelTest(TFModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
     all_model_classes = (TFWav2Vec2Model, TFWav2Vec2ForCTC) if is_tf_available() else ()
+    pipeline_model_mapping = {"feature-extraction": TFWav2Vec2Model} if is_tf_available() else {}
     test_resize_embeddings = False
     test_head_masking = False
     test_onnx = False

tests/models/wav2vec2/test_modeling_wav2vec2.py

@@ -49,6 +49,7 @@ from ...test_modeling_common import (
     ids_tensor,
     random_attention_mask,
 )
+from ...test_pipeline_mixin import PipelineTesterMixin
 
 
 if is_torch_available():
@@ -467,12 +468,22 @@ class Wav2Vec2ModelTester:
 
 
 @require_torch
-class Wav2Vec2ModelTest(ModelTesterMixin, unittest.TestCase):
+class Wav2Vec2ModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
     all_model_classes = (
         (Wav2Vec2ForCTC, Wav2Vec2Model, Wav2Vec2ForMaskedLM, Wav2Vec2ForSequenceClassification, Wav2Vec2ForPreTraining)
         if is_torch_available()
         else ()
     )
+    pipeline_model_mapping = (
+        {
+            "audio-classification": Wav2Vec2ForSequenceClassification,
+            "automatic-speech-recognition": Wav2Vec2ForCTC,
+            "feature-extraction": Wav2Vec2Model,
+            "fill-mask": Wav2Vec2ForMaskedLM,
+        }
+        if is_torch_available()
+        else {}
+    )
     fx_compatible = True
     test_pruning = False
     test_headmasking = False

tests/models/wav2vec2_conformer/test_modeling_wav2vec2_conformer.py

@@ -31,6 +31,7 @@ from ...test_modeling_common import (
     ids_tensor,
     random_attention_mask,
 )
+from ...test_pipeline_mixin import PipelineTesterMixin
 
 
 if is_torch_available():
@@ -389,7 +390,7 @@ class Wav2Vec2ConformerModelTester:
 
 
 @require_torch
-class Wav2Vec2ConformerModelTest(ModelTesterMixin, unittest.TestCase):
+class Wav2Vec2ConformerModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
     all_model_classes = (
         (
             Wav2Vec2ConformerForCTC,
@@ -402,6 +403,15 @@ class Wav2Vec2ConformerModelTest(ModelTesterMixin, unittest.TestCase):
         if is_torch_available()
         else ()
     )
+    pipeline_model_mapping = (
+        {
+            "audio-classification": Wav2Vec2ConformerForSequenceClassification,
+            "automatic-speech-recognition": Wav2Vec2ConformerForCTC,
+            "feature-extraction": Wav2Vec2ConformerModel,
+        }
+        if is_torch_available()
+        else {}
+    )
     test_pruning = False
     test_headmasking = False
     test_torchscript = False