[`core`/ `GC` / `tests`] Stronger GC tests (#27124)

* stronger GC tests

* better tests and skip failing tests

* break down into 3 sub-tests

* break down into 3 sub-tests

* refactor a bit

* more refactor

* fix

* last nit

* credits contrib and suggestions

* credits contrib and suggestions

---------
Co-authored-by: Yih-Dar <2521628+ydshieh@users.noreply.github.com>
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

tests/models/git/test_modeling_git.py

@@ -174,6 +174,18 @@ class GitVisionModelTest(ModelTesterMixin, unittest.TestCase):
     def test_training_gradient_checkpointing(self):
         pass
 
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant_false(self):
+        pass
+
     @unittest.skip(reason="GitVisionModel has no base class and is not available in MODEL_MAPPING")
     def test_save_load_fast_init_from_base(self):
         pass

tests/models/gpt2/test_modeling_gpt2.py

@@ -562,6 +562,24 @@ class GPT2ModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixin
         config_and_inputs = self.model_tester.prepare_config_and_inputs()
         self.model_tester.create_and_check_gpt2_weight_initialization(*config_and_inputs)
 
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant_false(self):
+        pass
+
     @slow
     def test_batch_generation(self):
         model = GPT2LMHeadModel.from_pretrained("gpt2")

tests/models/groupvit/test_modeling_groupvit.py

@@ -270,6 +270,18 @@ class GroupViTVisionModelTest(ModelTesterMixin, unittest.TestCase):
     def test_training_gradient_checkpointing(self):
         pass
 
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant_false(self):
+        pass
+
     @unittest.skip(reason="GroupViTVisionModel has no base class and is not available in MODEL_MAPPING")
     def test_save_load_fast_init_from_base(self):
         pass
@@ -454,6 +466,18 @@ class GroupViTTextModelTest(ModelTesterMixin, unittest.TestCase):
     def test_training_gradient_checkpointing(self):
         pass
 
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant_false(self):
+        pass
+
     @unittest.skip(reason="GroupViTTextModel does not use inputs_embeds")
     def test_inputs_embeds(self):
         pass

tests/models/idefics/test_modeling_idefics.py

@@ -379,6 +379,18 @@ class IdeficsModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase)
             loss = model(**inputs).loss
             loss.backward()
 
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant_false(self):
+        pass
+
     @unittest.skip(reason="""IDEFICS does not support retaining the gradients of the hidden states and attention""")
     def test_retain_grad_hidden_states_attentions(self):
         return
@@ -496,6 +508,18 @@ class IdeficsForVisionText2TextTest(IdeficsModelTest, unittest.TestCase):
     def test_retain_grad_hidden_states_attentions(self):
         pass
 
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant_false(self):
+        pass
+
 
 @unittest.skipIf(not is_torch_greater_or_equal_than_2_0, reason="pytorch 2.0 or higher is required")
 @require_torch

tests/models/imagegpt/test_modeling_imagegpt.py

@@ -316,6 +316,24 @@ class ImageGPTModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterM
         config_and_inputs = self.model_tester.prepare_config_and_inputs()
         self.model_tester.create_and_check_imagegpt_for_image_classification(*config_and_inputs)
 
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant_false(self):
+        pass
+
     @slow
     def test_model_from_pretrained(self):
         for model_name in IMAGEGPT_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:

tests/models/informer/test_modeling_informer.py

@@ -279,6 +279,24 @@ class InformerModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase
     def test_determinism(self):
         pass
 
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant_false(self):
+        pass
+
     # # Input is 'static_categorical_features' not 'input_ids'
     def test_model_main_input_name(self):
         model_signature = inspect.signature(getattr(InformerModel, "forward"))

tests/models/instructblip/test_modeling_instructblip.py

@@ -199,6 +199,18 @@ class InstructBlipVisionModelTest(ModelTesterMixin, unittest.TestCase):
     def test_training_gradient_checkpointing(self):
         pass
 
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant_false(self):
+        pass
+
     @unittest.skip(reason="InstructBlipVisionModel has no base class and is not available in MODEL_MAPPING")
     def test_save_load_fast_init_from_base(self):
         pass

tests/models/layoutlm/test_modeling_layoutlm.py

@@ -279,6 +279,24 @@ class LayoutLMModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase
         config_and_inputs = self.model_tester.prepare_config_and_inputs()
         self.model_tester.create_and_check_for_question_answering(*config_and_inputs)
 
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant_false(self):
+        pass
+
 
 def prepare_layoutlm_batch_inputs():
     # Here we prepare a batch of 2 sequences to test a LayoutLM forward pass on:

tests/models/lilt/test_modeling_lilt.py

@@ -275,6 +275,24 @@ class LiltModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixin
         config_and_inputs = self.model_tester.prepare_config_and_inputs()
         self.model_tester.create_and_check_for_question_answering(*config_and_inputs)
 
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant_false(self):
+        pass
+
     @slow
     def test_model_from_pretrained(self):
         for model_name in LILT_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:

tests/models/luke/test_modeling_luke.py

@@ -855,6 +855,24 @@ class LukeModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
 
         self.assertIsNotNone(entity_hidden_states.grad)
 
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant_false(self):
+        pass
+
 
 @require_torch
 class LukeModelIntegrationTests(unittest.TestCase):

tests/models/marian/test_modeling_marian.py

@@ -347,6 +347,24 @@ class MarianModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMix
     def test_pipeline_conversational(self):
         pass
 
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant_false(self):
+        pass
+
 
 def assert_tensors_close(a, b, atol=1e-12, prefix=""):
     """If tensors have different shapes, different values or a and b are not both tensors, raise a nice Assertion error."""

tests/models/mra/test_modeling_mra.py

@@ -360,6 +360,24 @@ class MraModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
     def test_attention_outputs(self):
         return
 
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant_false(self):
+        pass
+
 
 @require_torch
 class MraModelIntegrationTest(unittest.TestCase):

tests/models/owlv2/test_modeling_owlv2.py

@@ -190,6 +190,18 @@ class Owlv2VisionModelTest(ModelTesterMixin, unittest.TestCase):
     def test_training_gradient_checkpointing(self):
         pass
 
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant_false(self):
+        pass
+
     @unittest.skip(reason="Owlv2VisionModel has no base class and is not available in MODEL_MAPPING")
     def test_save_load_fast_init_from_base(self):
         pass
@@ -322,6 +334,18 @@ class Owlv2TextModelTest(ModelTesterMixin, unittest.TestCase):
     def test_training_gradient_checkpointing(self):
         pass
 
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant_false(self):
+        pass
+
     @unittest.skip(reason="OWLV2 does not use inputs_embeds")
     def test_inputs_embeds(self):
         pass
@@ -660,6 +684,18 @@ class Owlv2ForObjectDetectionTest(ModelTesterMixin, unittest.TestCase):
     def test_training_gradient_checkpointing(self):
         pass
 
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant_false(self):
+        pass
+
     def _create_and_check_torchscript(self, config, inputs_dict):
         if not self.test_torchscript:
             return

tests/models/owlvit/test_modeling_owlvit.py

@@ -188,6 +188,18 @@ class OwlViTVisionModelTest(ModelTesterMixin, unittest.TestCase):
     def test_training_gradient_checkpointing(self):
         pass
 
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant_false(self):
+        pass
+
     @unittest.skip(reason="OwlViTVisionModel has no base class and is not available in MODEL_MAPPING")
     def test_save_load_fast_init_from_base(self):
         pass
@@ -318,6 +330,18 @@ class OwlViTTextModelTest(ModelTesterMixin, unittest.TestCase):
     def test_training_gradient_checkpointing(self):
         pass
 
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant_false(self):
+        pass
+
     @unittest.skip(reason="OWLVIT does not use inputs_embeds")
     def test_inputs_embeds(self):
         pass
@@ -653,6 +677,18 @@ class OwlViTForObjectDetectionTest(ModelTesterMixin, unittest.TestCase):
     def test_training_gradient_checkpointing(self):
         pass
 
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant_false(self):
+        pass
+
     def _create_and_check_torchscript(self, config, inputs_dict):
         if not self.test_torchscript:
             return

tests/models/pegasus/test_modeling_pegasus.py

@@ -290,6 +290,24 @@ class PegasusModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMi
         model.generate(input_ids, attention_mask=attention_mask)
         model.generate(num_beams=4, do_sample=True, early_stopping=False, num_return_sequences=3)
 
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant_false(self):
+        pass
+
 
 def assert_tensors_close(a, b, atol=1e-12, prefix=""):
     """If tensors have different shapes, different values or a and b are not both tensors, raise a nice Assertion error."""

tests/models/pix2struct/test_modeling_pix2struct.py

@@ -199,6 +199,18 @@ class Pix2StructVisionModelTest(ModelTesterMixin, unittest.TestCase):
     def test_training_gradient_checkpointing(self):
         pass
 
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant_false(self):
+        pass
+
     @unittest.skip(reason="Training is tested directly on `Pix2StructTextImageModelTest`")
     def test_retain_grad_hidden_states_attentions(self):
         pass
@@ -336,6 +348,18 @@ class Pix2StructTextModelTest(ModelTesterMixin, unittest.TestCase):
     def test_training_gradient_checkpointing(self):
         pass
 
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant_false(self):
+        pass
+
     @unittest.skip(reason="Pix2Struct does not use inputs_embeds")
     def test_inputs_embeds(self):
         pass

tests/models/roformer/test_modeling_roformer.py

@@ -486,6 +486,24 @@ class RoFormerModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase
             model = RoFormerModel.from_pretrained(model_name)
             self.assertIsNotNone(model)
 
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant_false(self):
+        pass
+
 
 @require_torch
 class RoFormerModelIntegrationTest(unittest.TestCase):

tests/models/sam/test_modeling_sam.py

@@ -421,6 +421,18 @@ class SamModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
     def test_training_gradient_checkpointing(self):
         pass
 
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant_false(self):
+        pass
+
     @unittest.skip(reason="SamModel has no base class and is not available in MODEL_MAPPING")
     def test_save_load_fast_init_from_base(self):
         pass

tests/models/seamless_m4t/test_modeling_seamless_m4t.py

@@ -490,6 +490,24 @@ class SeamlessM4TModelWithSpeechInputTest(ModelTesterMixin, unittest.TestCase):
     def test_save_load_fast_init_from_base(self):
         pass
 
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant_false(self):
+        pass
+
     def test_attention_outputs(self):
         # expected length is subsampled so need to change a bit this test
         if not self.has_attentions:
@@ -735,6 +753,24 @@ class SeamlessM4TModelWithTextInputTest(ModelTesterMixin, GenerationTesterMixin,
     def test_save_load_fast_init_from_base(self):
         pass
 
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant_false(self):
+        pass
+
 
 @require_torch
 class SeamlessM4TGenerationTest(unittest.TestCase):

tests/models/speech_to_text/test_modeling_speech_to_text.py

@@ -324,6 +324,18 @@ class Speech2TextModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTest
     def test_training_gradient_checkpointing(self):
         pass
 
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant(self):
+        pass
+
+    @unittest.skip(
+        reason="This architecure seem to not compute gradients properly when using GC, check: https://github.com/huggingface/transformers/pull/27124"
+    )
+    def test_training_gradient_checkpointing_use_reentrant_false(self):
+        pass
+
     def test_generate_fp16(self):
         config, input_dict = self.model_tester.prepare_config_and_inputs()
         input_features = input_dict["input_features"]