Rework pipeline tests (#19366)

* Rework pipeline tests

* Try to fix Flax tests

* Try to put it before

* Use a new decorator instead

* Remove ignore marker since it doesn't work

* Filter pipeline tests

* Woopsie

* Use the fitlered list

* Clean up and fake modif

* Remove init

* Revert fake modif

.circleci/config.yml

@@ -79,10 +79,19 @@ jobs:
                 path: ~/transformers/tests_fetched_summary.txt
             - run: |
                 if [ -f test_list.txt ]; then
-                    mv test_list.txt test_preparation/test_list.txt
+                    cp test_list.txt test_preparation/test_list.txt
                 else
                     touch test_preparation/test_list.txt
                 fi
+            - run: python utils/tests_fetcher.py --filter_pipeline_tests
+            - run: |
+                if [ -f test_list.txt ]; then
+                    mv test_list.txt test_preparation/filtered_test_list.txt
+                else
+                    touch test_preparation/filtered_test_list.txt
+                fi
+            - store_artifacts:
+                  path: ~/transformers/test_preparation/filtered_test_list.txt
             - run: python utils/tests_fetcher.py --filters tests examples | tee examples_tests_fetched_summary.txt
             - store_artifacts:
                 path: ~/transformers/examples_tests_fetched_summary.txt
@@ -97,6 +106,7 @@ jobs:
                 root: test_preparation/
                 paths:
                     test_list.txt
+                    filtered_test_list.txt
                     examples_test_list.txt
 
     # To run all tests for the nightly build
@@ -110,6 +120,8 @@ jobs:
                   mkdir test_preparation
                   echo "tests" > test_preparation/test_list.txt
                   echo "tests" > test_preparation/examples_test_list.txt
+            - run: python utils/tests_fetcher.py --filter_pipeline_tests
+            - run: mv test_list.txt test_preparation/filtered_test_list.txt
 
             - persist_to_workspace:
                   root: test_preparation/
@@ -132,7 +144,7 @@ jobs:
             - attach_workspace:
                 at: ~/transformers/test_preparation
             - run: |
-                if [ ! -s test_preparation/test_list.txt ]; then
+                if [ ! -s test_preparation/filtered_test_list.txt ]; then
                     echo "No tests to run, exiting early!"
                     circleci-agent step halt
                 fi
@@ -152,7 +164,7 @@ jobs:
                 key: v0.5-torch_and_tf-{{ checksum "setup.py" }}
                 paths:
                     - '~/.cache/pip'
-            - run: python -m pytest -n 8 --max-worker-restart=0 --dist=loadfile -rA -s --make-reports=tests_torch_and_tf $(cat test_preparation/test_list.txt) -m is_pt_tf_cross_test --durations=0 | tee tests_output.txt
+            - run: python -m pytest -n 8 --max-worker-restart=0 --dist=loadfile -rA -s --make-reports=tests_torch_and_tf $(cat test_preparation/filtered_test_list.txt) -m is_pt_tf_cross_test --durations=0 | tee tests_output.txt
             - store_artifacts:
                   path: ~/transformers/tests_output.txt
             - store_artifacts:
@@ -174,7 +186,7 @@ jobs:
             - attach_workspace:
                 at: ~/transformers/test_preparation
             - run: |
-                if [ ! -s test_preparation/test_list.txt ]; then
+                if [ ! -s test_preparation/filtered_test_list.txt ]; then
                     echo "No tests to run, exiting early!"
                     circleci-agent step halt
                 fi
@@ -192,7 +204,7 @@ jobs:
                 key: v0.5-torch_and_flax-{{ checksum "setup.py" }}
                 paths:
                     - '~/.cache/pip'
-            - run: python -m pytest -n 8 --max-worker-restart=0 --dist=loadfile -rA -s --make-reports=tests_torch_and_flax $(cat test_preparation/test_list.txt) -m is_pt_flax_cross_test --durations=0 | tee tests_output.txt
+            - run: python -m pytest -n 8 --max-worker-restart=0 --dist=loadfile -rA -s --make-reports=tests_torch_and_flax $(cat test_preparation/filtered_test_list.txt) -m is_pt_flax_cross_test --durations=0 | tee tests_output.txt
             - store_artifacts:
                   path: ~/transformers/tests_output.txt
             - store_artifacts:
@@ -213,7 +225,7 @@ jobs:
             - attach_workspace:
                 at: ~/transformers/test_preparation
             - run: |
-                if [ ! -s test_preparation/test_list.txt ]; then
+                if [ ! -s test_preparation/filtered_test_list.txt ]; then
                     echo "No tests to run, exiting early!"
                     circleci-agent step halt
                 fi
@@ -231,7 +243,7 @@ jobs:
                   key: v0.5-torch-{{ checksum "setup.py" }}
                   paths:
                       - '~/.cache/pip'
-            - run: python -m pytest -n 3 --max-worker-restart=0 --dist=loadfile -s --make-reports=tests_torch $(cat test_preparation/test_list.txt) | tee tests_output.txt
+            - run: python -m pytest -n 3 --max-worker-restart=0 --dist=loadfile -s --make-reports=tests_torch $(cat test_preparation/filtered_test_list.txt) | tee tests_output.txt
             - store_artifacts:
                   path: ~/transformers/tests_output.txt
             - store_artifacts:
@@ -252,7 +264,7 @@ jobs:
             - attach_workspace:
                 at: ~/transformers/test_preparation
             - run: |
-                if [ ! -s test_preparation/test_list.txt ]; then
+                if [ ! -s test_preparation/filtered_test_list.txt ]; then
                     echo "No tests to run, exiting early!"
                     circleci-agent step halt
                 fi
@@ -269,7 +281,7 @@ jobs:
                   key: v0.5-tf-{{ checksum "setup.py" }}
                   paths:
                       - '~/.cache/pip'
-            - run: python -m pytest -n 8 --max-worker-restart=0 --dist=loadfile -rA -s --make-reports=tests_tf $(cat test_preparation/test_list.txt) | tee tests_output.txt
+            - run: python -m pytest -n 8 --max-worker-restart=0 --dist=loadfile -rA -s --make-reports=tests_tf $(cat test_preparation/filtered_test_list.txt) | tee tests_output.txt
             - store_artifacts:
                   path: ~/transformers/tests_output.txt
             - store_artifacts:
@@ -290,7 +302,7 @@ jobs:
             - attach_workspace:
                 at: ~/transformers/test_preparation
             - run: |
-                if [ ! -s test_preparation/test_list.txt ]; then
+                if [ ! -s test_preparation/filtered_test_list.txt ]; then
                     echo "No tests to run, exiting early!"
                     circleci-agent step halt
                 fi
@@ -306,7 +318,7 @@ jobs:
                   key: v0.5-flax-{{ checksum "setup.py" }}
                   paths:
                       - '~/.cache/pip'
-            - run: python -m pytest -n 8 --max-worker-restart=0 --dist=loadfile -rA -s --make-reports=tests_flax $(cat test_preparation/test_list.txt) | tee tests_output.txt
+            - run: python -m pytest -n 8 --max-worker-restart=0 --dist=loadfile -rA -s --make-reports=tests_flax $(cat test_preparation/filtered_test_list.txt) | tee tests_output.txt
             - store_artifacts:
                   path: ~/transformers/tests_output.txt
             - store_artifacts:
@@ -318,7 +330,6 @@ jobs:
             - image: cimg/python:3.7.12
         environment:
             OMP_NUM_THREADS: 1
-            RUN_PIPELINE_TESTS: yes
             TRANSFORMERS_IS_CI: yes
             PYTEST_TIMEOUT: 120
         resource_class: xlarge
@@ -345,7 +356,7 @@ jobs:
                   key: v0.5-torch-{{ checksum "setup.py" }}
                   paths:
                       - '~/.cache/pip'
-            - run: python -m pytest -n 8 --max-worker-restart=0 --dist=loadfile -rA -s --make-reports=tests_pipelines_torch -m is_pipeline_test $(cat test_preparation/test_list.txt) | tee tests_output.txt
+            - run: python -m pytest -n 8 --max-worker-restart=0 --dist=loadfile -rA -s --make-reports=tests_pipelines_torch tests/pipelines | tee tests_output.txt
             - store_artifacts:
                   path: ~/transformers/tests_output.txt
             - store_artifacts:
@@ -357,7 +368,6 @@ jobs:
             - image: cimg/python:3.7.12
         environment:
             OMP_NUM_THREADS: 1
-            RUN_PIPELINE_TESTS: yes
             TRANSFORMERS_IS_CI: yes
             PYTEST_TIMEOUT: 120
         resource_class: xlarge
@@ -382,7 +392,7 @@ jobs:
                   key: v0.5-tf-{{ checksum "setup.py" }}
                   paths:
                       - '~/.cache/pip'
-            - run: python -m pytest -n 8 --max-worker-restart=0 --dist=loadfile -rA -s --make-reports=tests_pipelines_tf $(cat test_preparation/test_list.txt) -m is_pipeline_test | tee tests_output.txt
+            - run: python -m pytest -n 8 --max-worker-restart=0 --dist=loadfile -rA -s --make-reports=tests_pipelines_tf tests/pipelines | tee tests_output.txt
             - store_artifacts:
                   path: ~/transformers/tests_output.txt
             - store_artifacts:
@@ -401,7 +411,7 @@ jobs:
             - attach_workspace:
                 at: ~/transformers/test_preparation
             - run: |
-                if [ ! -s test_preparation/test_list.txt ]; then
+                if [ ! -s test_preparation/filtered_test_list.txt ]; then
                     echo "No tests to run, exiting early!"
                     circleci-agent step halt
                 fi
@@ -557,7 +567,7 @@ jobs:
             - attach_workspace:
                 at: ~/transformers/test_preparation
             - run: |
-                if [ ! -s test_preparation/test_list.txt ]; then
+                if [ ! -s test_preparation/filtered_test_list.txt ]; then
                     echo "No tests to run, exiting early!"
                     circleci-agent step halt
                 fi
@@ -575,7 +585,7 @@ jobs:
                   key: v0.5-hub-{{ checksum "setup.py" }}
                   paths:
                       - '~/.cache/pip'
-            - run: python -m pytest --max-worker-restart=0 -sv --make-reports=tests_hub $(cat test_preparation/test_list.txt) -m is_staging_test | tee tests_output.txt
+            - run: python -m pytest --max-worker-restart=0 -sv --make-reports=tests_hub $(cat test_preparation/filtered_test_list.txt) -m is_staging_test | tee tests_output.txt
             - store_artifacts:
                   path: ~/transformers/tests_output.txt
             - store_artifacts:
@@ -596,7 +606,7 @@ jobs:
             - attach_workspace:
                 at: ~/transformers/test_preparation
             - run: |
-                if [ ! -s test_preparation/test_list.txt ]; then
+                if [ ! -s test_preparation/filtered_test_list.txt ]; then
                     echo "No tests to run, exiting early!"
                     circleci-agent step halt
                 fi
@@ -610,7 +620,7 @@ jobs:
                   key: v0.5-onnx-{{ checksum "setup.py" }}
                   paths:
                       - '~/.cache/pip'
-            - run: python -m pytest -n 1 --max-worker-restart=0 --dist=loadfile -s --make-reports=tests_onnx $(cat test_preparation/test_list.txt) -k onnx | tee tests_output.txt
+            - run: python -m pytest -n 1 --max-worker-restart=0 --dist=loadfile -s --make-reports=tests_onnx $(cat test_preparation/filtered_test_list.txt) -k onnx | tee tests_output.txt
 
             - store_artifacts:
                   path: ~/transformers/tests_output.txt
@@ -690,7 +700,7 @@ jobs:
         steps:
             - checkout
             - attach_workspace:
-                at: ~/transformers/test_preparation
+                at: ~/transformers/filtered_test_list.txt
             - run: |
                 if [ ! -s test_preparation/test_list.txt ]; then
                     echo "No tests to run, exiting early!"

.github/workflows/self-scheduled.yml

@@ -256,10 +256,8 @@ jobs:
 
       - name: Run all pipeline tests on GPU
         working-directory: /transformers
-        env:
-          RUN_PIPELINE_TESTS: yes
         run: |
-          python3 -m pytest -n 1 -v --dist=loadfile -m is_pipeline_test --make-reports=${{ matrix.machine_type }}_tests_torch_pipeline_gpu tests
+          python3 -m pytest -n 1 -v --dist=loadfile --make-reports=${{ matrix.machine_type }}_tests_torch_pipeline_gpu tests/pipelines
 
       - name: Failure short reports
         if: ${{ failure() }}
@@ -301,10 +299,8 @@ jobs:
 
       - name: Run all pipeline tests on GPU
         working-directory: /transformers
-        env:
-          RUN_PIPELINE_TESTS: yes
         run: |
-          python3 -m pytest -n 1 -v --dist=loadfile -m is_pipeline_test --make-reports=${{ matrix.machine_type }}_tests_tf_pipeline_gpu tests
+          python3 -m pytest -n 1 -v --dist=loadfile --make-reports=${{ matrix.machine_type }}_tests_tf_pipeline_gpu tests/pipelines
 
       - name: Failure short reports
         if: ${{ always() }}

conftest.py

@@ -32,7 +32,6 @@ warnings.simplefilter(action="ignore", category=FutureWarning)
 
 
 def pytest_configure(config):
-    config.addinivalue_line("markers", "is_pipeline_test: mark test to run only when pipeline are tested")
     config.addinivalue_line(
         "markers", "is_pt_tf_cross_test: mark test to run only when PT and TF interactions are tested"
     )

src/transformers/testing_utils.py

@@ -133,7 +133,6 @@ _run_pt_tf_cross_tests = parse_flag_from_env("RUN_PT_TF_CROSS_TESTS", default=Fa
 _run_pt_flax_cross_tests = parse_flag_from_env("RUN_PT_FLAX_CROSS_TESTS", default=False)
 _run_custom_tokenizers = parse_flag_from_env("RUN_CUSTOM_TOKENIZERS", default=False)
 _run_staging = parse_flag_from_env("HUGGINGFACE_CO_STAGING", default=False)
-_run_pipeline_tests = parse_flag_from_env("RUN_PIPELINE_TESTS", 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)
 
@@ -176,25 +175,6 @@ def is_pt_flax_cross_test(test_case):
             return pytest.mark.is_pt_flax_cross_test()(test_case)
 
 
-def is_pipeline_test(test_case):
-    """
-    Decorator marking a test as a pipeline test.
-
-    Pipeline tests are skipped by default and we can run only them by setting RUN_PIPELINE_TESTS environment variable
-    to a truthy value and selecting the is_pipeline_test pytest mark.
-
-    """
-    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 is_staging_test(test_case):
     """
     Decorator marking a test as a staging test.
@@ -309,6 +289,18 @@ def require_torch(test_case):
     return unittest.skipUnless(is_torch_available(), "test requires PyTorch")(test_case)
 
 
+def require_torch_or_tf(test_case):
+    """
+    Decorator marking a test that requires PyTorch or TensorFlow.
+
+    These tests are skipped when neither PyTorch not TensorFlow is installed.
+
+    """
+    return unittest.skipUnless(is_torch_available() or is_tf_available(), "test requires PyTorch or TensorFlow")(
+        test_case
+    )
+
+
 def require_intel_extension_for_pytorch(test_case):
     """
     Decorator marking a test that requires Intel Extension for PyTorch.

tests/pipelines/test_pipelines_audio_classification.py

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

tests/pipelines/test_pipelines_automatic_speech_recognition.py

@@ -31,7 +31,6 @@ from transformers.pipelines import AutomaticSpeechRecognitionPipeline, pipeline
 from transformers.pipelines.audio_utils import chunk_bytes_iter
 from transformers.pipelines.automatic_speech_recognition import chunk_iter
 from transformers.testing_utils import (
-    is_pipeline_test,
     is_torch_available,
     nested_simplify,
     require_pyctcdecode,
@@ -52,7 +51,6 @@ if is_torch_available():
 # from .test_pipelines_common import CustomInputPipelineCommonMixin
 
 
-@is_pipeline_test
 class AutomaticSpeechRecognitionPipelineTests(unittest.TestCase, metaclass=PipelineTestCaseMeta):
     model_mapping = {
         k: v

tests/pipelines/test_pipelines_common.py

@@ -48,13 +48,13 @@ from transformers.testing_utils import (
     USER,
     CaptureLogger,
     RequestCounter,
-    is_pipeline_test,
     is_staging_test,
     nested_simplify,
     require_scatter,
     require_tensorflow_probability,
     require_tf,
     require_torch,
+    require_torch_or_tf,
     slow,
 )
 from transformers.utils import is_tf_available, is_torch_available
@@ -307,7 +307,6 @@ class PipelineTestCaseMeta(type):
         return type.__new__(mcs, name, bases, dct)
 
 
-@is_pipeline_test
 class CommonPipelineTest(unittest.TestCase):
     @require_torch
     def test_pipeline_iteration(self):
@@ -416,7 +415,6 @@ class CommonPipelineTest(unittest.TestCase):
         self.assertEqual(len(outputs), 20)
 
 
-@is_pipeline_test
 class PipelinePadTest(unittest.TestCase):
     @require_torch
     def test_pipeline_padding(self):
@@ -498,7 +496,6 @@ class PipelinePadTest(unittest.TestCase):
         )
 
 
-@is_pipeline_test
 class PipelineUtilsTest(unittest.TestCase):
     @require_torch
     def test_pipeline_dataset(self):
@@ -795,7 +792,6 @@ 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()
@@ -835,6 +831,7 @@ class CustomPipelineTest(unittest.TestCase):
         # Clean registry for next tests.
         del PIPELINE_REGISTRY.supported_tasks["custom-text-classification"]
 
+    @require_torch_or_tf
     def test_dynamic_pipeline(self):
         PIPELINE_REGISTRY.register_pipeline(
             "pair-classification",
@@ -886,6 +883,7 @@ class CustomPipelineTest(unittest.TestCase):
             [{"label": "LABEL_0", "score": 0.505}],
         )
 
+    @require_torch_or_tf
     def test_cached_pipeline_has_minimum_calls_to_head(self):
         # Make sure we have cached the pipeline.
         _ = pipeline("text-classification", model="hf-internal-testing/tiny-random-bert")

tests/pipelines/test_pipelines_conversational.py

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

tests/pipelines/test_pipelines_document_question_answering.py

@@ -18,7 +18,6 @@ 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,
@@ -53,7 +52,6 @@ INVOICE_URL = (
 )
 
 
-@is_pipeline_test
 @require_torch
 @require_vision
 class DocumentQuestionAnsweringPipelineTests(unittest.TestCase, metaclass=PipelineTestCaseMeta):

tests/pipelines/test_pipelines_feature_extraction.py

@@ -22,12 +22,11 @@ from transformers import (
     LxmertConfig,
     pipeline,
 )
-from transformers.testing_utils import is_pipeline_test, nested_simplify, require_tf, require_torch
+from transformers.testing_utils import nested_simplify, require_tf, require_torch
 
 from .test_pipelines_common import PipelineTestCaseMeta
 
 
-@is_pipeline_test
 class FeatureExtractionPipelineTests(unittest.TestCase, metaclass=PipelineTestCaseMeta):
     model_mapping = MODEL_MAPPING
     tf_model_mapping = TF_MODEL_MAPPING

tests/pipelines/test_pipelines_fill_mask.py

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

tests/pipelines/test_pipelines_image_classification.py

@@ -22,10 +22,10 @@ from transformers import (
 )
 from transformers.pipelines import ImageClassificationPipeline, pipeline
 from transformers.testing_utils import (
-    is_pipeline_test,
     nested_simplify,
     require_tf,
     require_torch,
+    require_torch_or_tf,
     require_vision,
     slow,
 )
@@ -43,7 +43,7 @@ else:
             pass
 
 
-@is_pipeline_test
+@require_torch_or_tf
 @require_vision
 class ImageClassificationPipelineTests(unittest.TestCase, metaclass=PipelineTestCaseMeta):
     model_mapping = MODEL_FOR_IMAGE_CLASSIFICATION_MAPPING

tests/pipelines/test_pipelines_image_segmentation.py

@@ -31,15 +31,7 @@ from transformers import (
     is_vision_available,
     pipeline,
 )
-from transformers.testing_utils import (
-    is_pipeline_test,
-    nested_simplify,
-    require_tf,
-    require_timm,
-    require_torch,
-    require_vision,
-    slow,
-)
+from transformers.testing_utils import nested_simplify, require_tf, require_timm, require_torch, require_vision, slow
 
 from .test_pipelines_common import ANY, PipelineTestCaseMeta
 
@@ -62,7 +54,6 @@ def hashimage(image: Image) -> str:
 @require_vision
 @require_timm
 @require_torch
-@is_pipeline_test
 class ImageSegmentationPipelineTests(unittest.TestCase, metaclass=PipelineTestCaseMeta):
     model_mapping = {
         k: v

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

tests/pipelines/test_pipelines_object_detection.py

@@ -22,15 +22,7 @@ from transformers import (
     is_vision_available,
     pipeline,
 )
-from transformers.testing_utils import (
-    is_pipeline_test,
-    nested_simplify,
-    require_tf,
-    require_timm,
-    require_torch,
-    require_vision,
-    slow,
-)
+from transformers.testing_utils import nested_simplify, require_tf, require_timm, require_torch, require_vision, slow
 
 from .test_pipelines_common import ANY, PipelineTestCaseMeta
 
@@ -48,7 +40,6 @@ else:
 @require_vision
 @require_timm
 @require_torch
-@is_pipeline_test
 class ObjectDetectionPipelineTests(unittest.TestCase, metaclass=PipelineTestCaseMeta):
     model_mapping = MODEL_FOR_OBJECT_DETECTION_MAPPING
 

tests/pipelines/test_pipelines_question_answering.py

@@ -22,12 +22,11 @@ from transformers import (
 )
 from transformers.data.processors.squad import SquadExample
 from transformers.pipelines import QuestionAnsweringArgumentHandler, pipeline
-from transformers.testing_utils import is_pipeline_test, nested_simplify, require_tf, require_torch, slow
+from transformers.testing_utils import nested_simplify, require_tf, require_torch, require_torch_or_tf, slow
 
 from .test_pipelines_common import ANY, PipelineTestCaseMeta
 
 
-@is_pipeline_test
 class QAPipelineTests(unittest.TestCase, metaclass=PipelineTestCaseMeta):
     model_mapping = MODEL_FOR_QUESTION_ANSWERING_MAPPING
     tf_model_mapping = TF_MODEL_FOR_QUESTION_ANSWERING_MAPPING
@@ -345,7 +344,7 @@ between them. It's straightforward to train your models with one before loading
         self.assertEqual(nested_simplify(outputs), {"score": 0.979, "start": 27, "end": 32, "answer": "Paris"})
 
 
-@is_pipeline_test
+@require_torch_or_tf
 class QuestionAnsweringArgumentHandlerTests(unittest.TestCase):
     def test_argument_handler(self):
         qa = QuestionAnsweringArgumentHandler()

tests/pipelines/test_pipelines_summarization.py

@@ -23,7 +23,7 @@ from transformers import (
     T5Config,
     pipeline,
 )
-from transformers.testing_utils import is_pipeline_test, require_tf, require_torch, slow, torch_device
+from transformers.testing_utils import require_tf, require_torch, slow, torch_device
 from transformers.tokenization_utils import TruncationStrategy
 
 from .test_pipelines_common import ANY, PipelineTestCaseMeta
@@ -32,7 +32,6 @@ from .test_pipelines_common import ANY, PipelineTestCaseMeta
 DEFAULT_DEVICE_NUM = -1 if torch_device == "cpu" else 0
 
 
-@is_pipeline_test
 class SummarizationPipelineTests(unittest.TestCase, metaclass=PipelineTestCaseMeta):
     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

@@ -23,7 +23,6 @@ from transformers import (
     pipeline,
 )
 from transformers.testing_utils import (
-    is_pipeline_test,
     require_pandas,
     require_tensorflow_probability,
     require_tf,
@@ -35,7 +34,6 @@ from transformers.testing_utils import (
 from .test_pipelines_common import PipelineTestCaseMeta
 
 
-@is_pipeline_test
 class TQAPipelineTests(unittest.TestCase, metaclass=PipelineTestCaseMeta):
     # 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 is_pipeline_test, require_tf, require_torch
+from transformers.testing_utils import require_tf, require_torch
 from transformers.utils import is_torch_available
 
 from .test_pipelines_common import ANY, PipelineTestCaseMeta
@@ -30,7 +30,6 @@ if is_torch_available():
     import torch
 
 
-@is_pipeline_test
 class Text2TextGenerationPipelineTests(unittest.TestCase, metaclass=PipelineTestCaseMeta):
     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

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

tests/pipelines/test_pipelines_text_generation.py

@@ -16,17 +16,17 @@ import unittest
 
 from transformers import MODEL_FOR_CAUSAL_LM_MAPPING, TF_MODEL_FOR_CAUSAL_LM_MAPPING, TextGenerationPipeline, pipeline
 from transformers.testing_utils import (
-    is_pipeline_test,
     require_accelerate,
     require_tf,
     require_torch,
     require_torch_gpu,
+    require_torch_or_tf,
 )
 
 from .test_pipelines_common import ANY, PipelineTestCaseMeta
 
 
-@is_pipeline_test
+@require_torch_or_tf
 class TextGenerationPipelineTests(unittest.TestCase, metaclass=PipelineTestCaseMeta):
     model_mapping = MODEL_FOR_CAUSAL_LM_MAPPING
     tf_model_mapping = TF_MODEL_FOR_CAUSAL_LM_MAPPING

tests/pipelines/test_pipelines_token_classification.py

@@ -25,14 +25,7 @@ from transformers import (
     pipeline,
 )
 from transformers.pipelines import AggregationStrategy, TokenClassificationArgumentHandler
-from transformers.testing_utils import (
-    is_pipeline_test,
-    nested_simplify,
-    require_tf,
-    require_torch,
-    require_torch_gpu,
-    slow,
-)
+from transformers.testing_utils import nested_simplify, require_tf, require_torch, require_torch_gpu, slow
 
 from .test_pipelines_common import ANY, PipelineTestCaseMeta
 
@@ -40,7 +33,6 @@ from .test_pipelines_common import ANY, PipelineTestCaseMeta
 VALID_INPUTS = ["A simple string", ["list of strings", "A simple string that is quite a bit longer"]]
 
 
-@is_pipeline_test
 class TokenClassificationPipelineTests(unittest.TestCase, metaclass=PipelineTestCaseMeta):
     model_mapping = MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING
     tf_model_mapping = TF_MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING
@@ -770,7 +762,6 @@ class TokenClassificationPipelineTests(unittest.TestCase, metaclass=PipelineTest
         )
 
 
-@is_pipeline_test
 class TokenClassificationArgumentHandlerTestCase(unittest.TestCase):
     def setUp(self):
         self.args_parser = TokenClassificationArgumentHandler()

tests/pipelines/test_pipelines_translation.py

@@ -25,12 +25,11 @@ from transformers import (
     TranslationPipeline,
     pipeline,
 )
-from transformers.testing_utils import is_pipeline_test, require_tf, require_torch, slow
+from transformers.testing_utils import require_tf, require_torch, slow
 
 from .test_pipelines_common import ANY, PipelineTestCaseMeta
 
 
-@is_pipeline_test
 class TranslationPipelineTests(unittest.TestCase, metaclass=PipelineTestCaseMeta):
     model_mapping = MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING
     tf_model_mapping = TF_MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING
@@ -118,7 +117,6 @@ class TranslationPipelineTests(unittest.TestCase, metaclass=PipelineTestCaseMeta
         )
 
 
-@is_pipeline_test
 class TranslationNewFormatPipelineTests(unittest.TestCase):
     @require_torch
     @slow

tests/pipelines/test_pipelines_visual_question_answering.py

@@ -16,14 +16,7 @@ import unittest
 
 from transformers import MODEL_FOR_VISUAL_QUESTION_ANSWERING_MAPPING, is_vision_available
 from transformers.pipelines import pipeline
-from transformers.testing_utils import (
-    is_pipeline_test,
-    nested_simplify,
-    require_tf,
-    require_torch,
-    require_vision,
-    slow,
-)
+from transformers.testing_utils import nested_simplify, require_tf, require_torch, require_vision, slow
 
 from .test_pipelines_common import ANY, PipelineTestCaseMeta
 
@@ -38,7 +31,6 @@ else:
             pass
 
 
-@is_pipeline_test
 @require_torch
 @require_vision
 class VisualQuestionAnsweringPipelineTests(unittest.TestCase, metaclass=PipelineTestCaseMeta):

tests/pipelines/test_pipelines_zero_shot.py

@@ -21,12 +21,11 @@ from transformers import (
     ZeroShotClassificationPipeline,
     pipeline,
 )
-from transformers.testing_utils import is_pipeline_test, nested_simplify, require_tf, require_torch, slow
+from transformers.testing_utils import nested_simplify, require_tf, require_torch, slow
 
 from .test_pipelines_common import ANY, PipelineTestCaseMeta
 
 
-@is_pipeline_test
 class ZeroShotClassificationPipelineTests(unittest.TestCase, metaclass=PipelineTestCaseMeta):
     model_mapping = MODEL_FOR_SEQUENCE_CLASSIFICATION_MAPPING
     tf_model_mapping = TF_MODEL_FOR_SEQUENCE_CLASSIFICATION_MAPPING

tests/pipelines/test_pipelines_zero_shot_image_classification.py

@@ -16,14 +16,7 @@ import unittest
 
 from transformers import is_vision_available
 from transformers.pipelines import pipeline
-from transformers.testing_utils import (
-    is_pipeline_test,
-    nested_simplify,
-    require_tf,
-    require_torch,
-    require_vision,
-    slow,
-)
+from transformers.testing_utils import nested_simplify, require_tf, require_torch, require_vision, slow
 
 from .test_pipelines_common import ANY, PipelineTestCaseMeta
 
@@ -39,7 +32,6 @@ else:
 
 
 @require_vision
-@is_pipeline_test
 class ZeroShotImageClassificationPipelineTests(unittest.TestCase, metaclass=PipelineTestCaseMeta):
     # Deactivating auto tests since we don't have a good MODEL_FOR_XX mapping,
     # and only CLIP would be there for now.

utils/tests_fetcher.py

@@ -619,6 +619,25 @@ def infer_tests_to_run(output_file, diff_with_last_commit=False, filters=None, j
                 json.dump(test_map, fp, ensure_ascii=False)
 
 
+def filter_pipeline_tests(output_file):
+    if not os.path.isfile(output_file):
+        print("No test file found.")
+        return
+    with open(output_file, "r", encoding="utf-8") as f:
+        test_files = f.read().split(" ")
+
+    if len(test_files) == 0:
+        print("No tests to filter.")
+        return
+    if test_files == ["tests"]:
+        test_files = [os.path.join("tests", f) for f in os.listdir("tests") if f not in ["__init__.py", "pipelines"]]
+    else:
+        test_files = [f for f in test_files if not f.startswith(os.path.join("tests", "pipelines"))]
+
+    with open(output_file, "w", encoding="utf-8") as f:
+        f.write(" ".join(test_files))
+
+
 if __name__ == "__main__":
     parser = argparse.ArgumentParser()
     parser.add_argument(
@@ -645,6 +664,11 @@ if __name__ == "__main__":
         default=["tests"],
         help="Only keep the test files matching one of those filters.",
     )
+    parser.add_argument(
+        "--filter_pipeline_tests",
+        action="store_true",
+        help="Will filter the pipeline tests outside of the generated list of tests.",
+    )
     parser.add_argument(
         "--print_dependencies_of",
         type=str,
@@ -656,6 +680,8 @@ if __name__ == "__main__":
         print_tree_deps_of(args.print_dependencies_of)
     elif args.sanity_check:
         sanity_check()
+    elif args.filter_pipeline_tests:
+        filter_pipeline_tests(args.output_file)
     else:
         repo = Repo(PATH_TO_TRANFORMERS)