Mark pipeline tests to skip them easily (#21887)

* Mark pipeline tests to skip them easily

* Mark the mixin as pipeline test

* Update src/transformers/testing_utils.py
Co-authored-by: Yih-Dar <2521628+ydshieh@users.noreply.github.com>

---------
Co-authored-by: Yih-Dar <2521628+ydshieh@users.noreply.github.com>

.circleci/create_circleci_config.py

@@ -24,7 +24,7 @@ from typing import Any, Dict, List, Optional
 import yaml
 
 
-COMMON_ENV_VARIABLES = {"OMP_NUM_THREADS": 1, "TRANSFORMERS_IS_CI": True, "PYTEST_TIMEOUT": 120}
+COMMON_ENV_VARIABLES = {"OMP_NUM_THREADS": 1, "TRANSFORMERS_IS_CI": True, "PYTEST_TIMEOUT": 120, "RUN_PIPELINE_TESTS": False}
 COMMON_PYTEST_OPTIONS = {"max-worker-restart": 0, "dist": "loadfile", "s": None}
 DEFAULT_DOCKER_IMAGE = [{"image": "cimg/python:3.7.12"}]
 
@@ -64,10 +64,12 @@ class CircleCIJob:
             self.parallelism = 1
 
     def to_dict(self):
+        env = COMMON_ENV_VARIABLES.copy()
+        env.update(self.additional_env)
         job = {
             "working_directory": self.working_directory,
             "docker": self.docker_image,
-            "environment": {**COMMON_ENV_VARIABLES, **self.additional_env},
+            "environment": env,
         }
         if self.resource_class is not None:
             job["resource_class"] = self.resource_class
@@ -239,25 +241,27 @@ flax_job = CircleCIJob(
 
 pipelines_torch_job = CircleCIJob(
     "pipelines_torch",
+    additional_env={"RUN_PIPELINE_TESTS": True},
     install_steps=[
         "sudo apt-get -y update && sudo apt-get install -y libsndfile1-dev espeak-ng",
         "pip install --upgrade pip",
         "pip install .[sklearn,torch,testing,sentencepiece,torch-speech,vision,timm,video]",
     ],
     pytest_options={"rA": None},
-    tests_to_run="tests/pipelines/"
+    marker="is_pipeline_test",
 )
 
 
 pipelines_tf_job = CircleCIJob(
     "pipelines_tf",
+    additional_env={"RUN_PIPELINE_TESTS": True},
     install_steps=[
         "pip install --upgrade pip",
         "pip install .[sklearn,tf-cpu,testing,sentencepiece,vision]",
         "pip install tensorflow_probability",
     ],
     pytest_options={"rA": None},
-    tests_to_run="tests/pipelines/"
+    marker="is_pipeline_test",
 )
 
 

conftest.py

@@ -38,6 +38,9 @@ def pytest_configure(config):
     config.addinivalue_line(
         "markers", "is_pt_flax_cross_test: mark test to run only when PT and FLAX interactions are tested"
     )
+    config.addinivalue_line(
+        "markers", "is_pipeline_test: mark test to run only when pipelines are tested"
+    )
     config.addinivalue_line("markers", "is_staging_test: mark test to run only in the staging environment")
 
 

src/transformers/testing_utils.py

@@ -145,6 +145,7 @@ _run_custom_tokenizers = parse_flag_from_env("RUN_CUSTOM_TOKENIZERS", default=Fa
 _run_staging = parse_flag_from_env("HUGGINGFACE_CO_STAGING", default=False)
 _run_git_lfs_tests = parse_flag_from_env("RUN_GIT_LFS_TESTS", default=False)
 _tf_gpu_memory_limit = parse_int_from_env("TF_GPU_MEMORY_LIMIT", default=None)
+_run_pipeline_tests = parse_flag_from_env("RUN_PIPELINE_TESTS", default=True)
 
 
 def is_pt_tf_cross_test(test_case):
@@ -202,6 +203,22 @@ def is_staging_test(test_case):
             return pytest.mark.is_staging_test()(test_case)
 
 
+def is_pipeline_test(test_case):
+    """
+    Decorator marking a test as a pipeline test. If RUN_PIPELINE_TESTS is set to a falsy value, those tests will be
+    skipped.
+    """
+    if not _run_pipeline_tests:
+        return unittest.skip("test is pipeline test")(test_case)
+    else:
+        try:
+            import pytest  # We don't need a hard dependency on pytest in the main library
+        except ImportError:
+            return test_case
+        else:
+            return pytest.mark.is_pipeline_test()(test_case)
+
+
 def slow(test_case):
     """
     Decorator marking a test as slow.

tests/pipelines/test_pipelines_audio_classification.py

@@ -18,11 +18,19 @@ import numpy as np
 
 from transformers import MODEL_FOR_AUDIO_CLASSIFICATION_MAPPING
 from transformers.pipelines import AudioClassificationPipeline, pipeline
-from transformers.testing_utils import nested_simplify, require_tf, require_torch, require_torchaudio, slow
+from transformers.testing_utils import (
+    is_pipeline_test,
+    nested_simplify,
+    require_tf,
+    require_torch,
+    require_torchaudio,
+    slow,
+)
 
 from .test_pipelines_common import ANY
 
 
+@is_pipeline_test
 @require_torch
 class AudioClassificationPipelineTests(unittest.TestCase):
     model_mapping = MODEL_FOR_AUDIO_CLASSIFICATION_MAPPING

tests/pipelines/test_pipelines_automatic_speech_recognition.py

@@ -33,6 +33,7 @@ from transformers.pipelines import AutomaticSpeechRecognitionPipeline, pipeline
 from transformers.pipelines.audio_utils import chunk_bytes_iter
 from transformers.pipelines.automatic_speech_recognition import _find_timestamp_sequence, chunk_iter
 from transformers.testing_utils import (
+    is_pipeline_test,
     is_torch_available,
     nested_simplify,
     require_pyctcdecode,
@@ -53,6 +54,7 @@ if is_torch_available():
 # from .test_pipelines_common import CustomInputPipelineCommonMixin
 
 
+@is_pipeline_test
 class AutomaticSpeechRecognitionPipelineTests(unittest.TestCase):
     model_mapping = {
         k: v

tests/pipelines/test_pipelines_common.py

@@ -39,6 +39,7 @@ from transformers.testing_utils import (
     USER,
     CaptureLogger,
     RequestCounter,
+    is_pipeline_test,
     is_staging_test,
     nested_simplify,
     require_tensorflow_probability,
@@ -77,6 +78,7 @@ class ANY:
         return f"ANY({', '.join(_type.__name__ for _type in self._types)})"
 
 
+@is_pipeline_test
 class CommonPipelineTest(unittest.TestCase):
     @require_torch
     def test_pipeline_iteration(self):
@@ -194,6 +196,7 @@ class CommonPipelineTest(unittest.TestCase):
         self.assertEqual(len(outputs), 20)
 
 
+@is_pipeline_test
 class PipelineScikitCompatTest(unittest.TestCase):
     @require_torch
     def test_pipeline_predict_pt(self):
@@ -244,6 +247,7 @@ class PipelineScikitCompatTest(unittest.TestCase):
         self.assertEqual(expected_output, actual_output)
 
 
+@is_pipeline_test
 class PipelinePadTest(unittest.TestCase):
     @require_torch
     def test_pipeline_padding(self):
@@ -325,6 +329,7 @@ class PipelinePadTest(unittest.TestCase):
         )
 
 
+@is_pipeline_test
 class PipelineUtilsTest(unittest.TestCase):
     @require_torch
     def test_pipeline_dataset(self):
@@ -620,6 +625,7 @@ class CustomPipeline(Pipeline):
         return model_outputs["logits"].softmax(-1).numpy()
 
 
+@is_pipeline_test
 class CustomPipelineTest(unittest.TestCase):
     def test_warning_logs(self):
         transformers_logging.set_verbosity_debug()

tests/pipelines/test_pipelines_conversational.py

@@ -29,7 +29,7 @@ from transformers import (
     TFAutoModelForCausalLM,
     pipeline,
 )
-from transformers.testing_utils import require_tf, require_torch, slow, torch_device
+from transformers.testing_utils import is_pipeline_test, require_tf, require_torch, slow, torch_device
 
 from .test_pipelines_common import ANY
 
@@ -37,6 +37,7 @@ from .test_pipelines_common import ANY
 DEFAULT_DEVICE_NUM = -1 if torch_device == "cpu" else 0
 
 
+@is_pipeline_test
 class ConversationalPipelineTests(unittest.TestCase):
     model_mapping = dict(
         list(MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING.items())

tests/pipelines/test_pipelines_depth_estimation.py

@@ -17,7 +17,15 @@ import unittest
 
 from transformers import MODEL_FOR_DEPTH_ESTIMATION_MAPPING, is_torch_available, is_vision_available
 from transformers.pipelines import DepthEstimationPipeline, pipeline
-from transformers.testing_utils import nested_simplify, require_tf, require_timm, require_torch, require_vision, slow
+from transformers.testing_utils import (
+    is_pipeline_test,
+    nested_simplify,
+    require_tf,
+    require_timm,
+    require_torch,
+    require_vision,
+    slow,
+)
 
 from .test_pipelines_common import ANY
 
@@ -40,6 +48,7 @@ def hashimage(image: Image) -> str:
     return m.hexdigest()
 
 
+@is_pipeline_test
 @require_vision
 @require_timm
 @require_torch

tests/pipelines/test_pipelines_document_question_answering.py

@@ -18,6 +18,7 @@ from transformers import MODEL_FOR_DOCUMENT_QUESTION_ANSWERING_MAPPING, AutoToke
 from transformers.pipelines import pipeline
 from transformers.pipelines.document_question_answering import apply_tesseract
 from transformers.testing_utils import (
+    is_pipeline_test,
     nested_simplify,
     require_detectron2,
     require_pytesseract,
@@ -52,6 +53,7 @@ INVOICE_URL = (
 )
 
 
+@is_pipeline_test
 @require_torch
 @require_vision
 class DocumentQuestionAnsweringPipelineTests(unittest.TestCase):

tests/pipelines/test_pipelines_feature_extraction.py

@@ -27,7 +27,7 @@ from transformers import (
     is_torch_available,
     pipeline,
 )
-from transformers.testing_utils import nested_simplify, require_tf, require_torch
+from transformers.testing_utils import is_pipeline_test, nested_simplify, require_tf, require_torch
 
 
 if is_torch_available():
@@ -37,6 +37,7 @@ if is_tf_available():
     import tensorflow as tf
 
 
+@is_pipeline_test
 class FeatureExtractionPipelineTests(unittest.TestCase):
     model_mapping = MODEL_MAPPING
     tf_model_mapping = TF_MODEL_MAPPING

tests/pipelines/test_pipelines_fill_mask.py

@@ -16,11 +16,19 @@ import unittest
 
 from transformers import MODEL_FOR_MASKED_LM_MAPPING, TF_MODEL_FOR_MASKED_LM_MAPPING, FillMaskPipeline, pipeline
 from transformers.pipelines import PipelineException
-from transformers.testing_utils import nested_simplify, require_tf, require_torch, require_torch_gpu, slow
+from transformers.testing_utils import (
+    is_pipeline_test,
+    nested_simplify,
+    require_tf,
+    require_torch,
+    require_torch_gpu,
+    slow,
+)
 
 from .test_pipelines_common import ANY
 
 
+@is_pipeline_test
 class FillMaskPipelineTests(unittest.TestCase):
     model_mapping = MODEL_FOR_MASKED_LM_MAPPING
     tf_model_mapping = TF_MODEL_FOR_MASKED_LM_MAPPING

tests/pipelines/test_pipelines_image_classification.py

@@ -22,6 +22,7 @@ from transformers import (
 )
 from transformers.pipelines import ImageClassificationPipeline, pipeline
 from transformers.testing_utils import (
+    is_pipeline_test,
     nested_simplify,
     require_tf,
     require_torch,
@@ -43,6 +44,7 @@ else:
             pass
 
 
+@is_pipeline_test
 @require_torch_or_tf
 @require_vision
 class ImageClassificationPipelineTests(unittest.TestCase):

tests/pipelines/test_pipelines_image_segmentation.py

@@ -34,7 +34,15 @@ from transformers import (
     is_vision_available,
     pipeline,
 )
-from transformers.testing_utils import nested_simplify, require_tf, require_timm, require_torch, require_vision, slow
+from transformers.testing_utils import (
+    is_pipeline_test,
+    nested_simplify,
+    require_tf,
+    require_timm,
+    require_torch,
+    require_vision,
+    slow,
+)
 
 from .test_pipelines_common import ANY
 
@@ -67,6 +75,7 @@ def mask_to_test_readable_only_shape(mask: Image) -> Dict:
     return {"shape": shape}
 
 
+@is_pipeline_test
 @require_vision
 @require_timm
 @require_torch

tests/pipelines/test_pipelines_image_to_text.py

@@ -16,7 +16,7 @@ import unittest
 
 from transformers import MODEL_FOR_VISION_2_SEQ_MAPPING, TF_MODEL_FOR_VISION_2_SEQ_MAPPING, is_vision_available
 from transformers.pipelines import pipeline
-from transformers.testing_utils import require_tf, require_torch, require_vision, slow
+from transformers.testing_utils import is_pipeline_test, require_tf, require_torch, require_vision, slow
 
 from .test_pipelines_common import ANY
 
@@ -31,6 +31,7 @@ else:
             pass
 
 
+@is_pipeline_test
 @require_vision
 class ImageToTextPipelineTests(unittest.TestCase):
     model_mapping = MODEL_FOR_VISION_2_SEQ_MAPPING

tests/pipelines/test_pipelines_object_detection.py

@@ -23,6 +23,7 @@ from transformers import (
     pipeline,
 )
 from transformers.testing_utils import (
+    is_pipeline_test,
     nested_simplify,
     require_pytesseract,
     require_tf,
@@ -45,6 +46,7 @@ else:
             pass
 
 
+@is_pipeline_test
 @require_vision
 @require_timm
 @require_torch

tests/pipelines/test_pipelines_question_answering.py

@@ -22,11 +22,19 @@ from transformers import (
 )
 from transformers.data.processors.squad import SquadExample
 from transformers.pipelines import QuestionAnsweringArgumentHandler, pipeline
-from transformers.testing_utils import nested_simplify, require_tf, require_torch, require_torch_or_tf, slow
+from transformers.testing_utils import (
+    is_pipeline_test,
+    nested_simplify,
+    require_tf,
+    require_torch,
+    require_torch_or_tf,
+    slow,
+)
 
 from .test_pipelines_common import ANY
 
 
+@is_pipeline_test
 class QAPipelineTests(unittest.TestCase):
     model_mapping = MODEL_FOR_QUESTION_ANSWERING_MAPPING
     tf_model_mapping = TF_MODEL_FOR_QUESTION_ANSWERING_MAPPING

tests/pipelines/test_pipelines_summarization.py

@@ -21,7 +21,7 @@ from transformers import (
     TFPreTrainedModel,
     pipeline,
 )
-from transformers.testing_utils import get_gpu_count, require_tf, require_torch, slow, torch_device
+from transformers.testing_utils import get_gpu_count, is_pipeline_test, require_tf, require_torch, slow, torch_device
 from transformers.tokenization_utils import TruncationStrategy
 
 from .test_pipelines_common import ANY
@@ -30,6 +30,7 @@ from .test_pipelines_common import ANY
 DEFAULT_DEVICE_NUM = -1 if torch_device == "cpu" else 0
 
 
+@is_pipeline_test
 class SummarizationPipelineTests(unittest.TestCase):
     model_mapping = MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING
     tf_model_mapping = TF_MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING

tests/pipelines/test_pipelines_table_question_answering.py

@@ -22,9 +22,17 @@ from transformers import (
     TFAutoModelForTableQuestionAnswering,
     pipeline,
 )
-from transformers.testing_utils import require_pandas, require_tensorflow_probability, require_tf, require_torch, slow
+from transformers.testing_utils import (
+    is_pipeline_test,
+    require_pandas,
+    require_tensorflow_probability,
+    require_tf,
+    require_torch,
+    slow,
+)
 
 
+@is_pipeline_test
 class TQAPipelineTests(unittest.TestCase):
     # Putting it there for consistency, but TQA do not have fast tokenizer
     # which are needed to generate automatic tests

tests/pipelines/test_pipelines_text2text_generation.py

@@ -20,7 +20,7 @@ from transformers import (
     Text2TextGenerationPipeline,
     pipeline,
 )
-from transformers.testing_utils import require_tf, require_torch
+from transformers.testing_utils import is_pipeline_test, require_tf, require_torch
 from transformers.utils import is_torch_available
 
 from .test_pipelines_common import ANY
@@ -30,6 +30,7 @@ if is_torch_available():
     import torch
 
 
+@is_pipeline_test
 class Text2TextGenerationPipelineTests(unittest.TestCase):
     model_mapping = MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING
     tf_model_mapping = TF_MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING

tests/pipelines/test_pipelines_text_classification.py

@@ -20,11 +20,12 @@ from transformers import (
     TextClassificationPipeline,
     pipeline,
 )
-from transformers.testing_utils import nested_simplify, require_tf, require_torch, slow
+from transformers.testing_utils import is_pipeline_test, nested_simplify, require_tf, require_torch, slow
 
 from .test_pipelines_common import ANY
 
 
+@is_pipeline_test
 class TextClassificationPipelineTests(unittest.TestCase):
     model_mapping = MODEL_FOR_SEQUENCE_CLASSIFICATION_MAPPING
     tf_model_mapping = TF_MODEL_FOR_SEQUENCE_CLASSIFICATION_MAPPING