Generate: skip left-padding tests on old models (#23437)

tests/generation/test_utils.py

@@ -1597,9 +1597,7 @@ class GenerationTesterMixin:
                 attn_weights = out[attn_name] if attn_name == attention_names[0] else out[attn_name][-1]
                 self.assertEqual(sum([w.sum().item() for w in attn_weights]), 0.0)
 
-    # TODO (joao): this test is actually not slow :) However, it is not passing in some models (e.g. GPTNeoX) and the
-    # fix for some models is quite lengthy. Being slow means it doesn't block our push CI while we fix it.
-    @slow
+    @slow  # TODO (Joao): fix GPTBigCode
     def test_left_padding_compatibility(self):
         # The check done in this test is fairly difficult -- depending on the model architecture, passing the right
         # position index for the position embeddings can still result in a different output, due to numerical masking.

tests/models/bart/test_modeling_bart.py

@@ -1516,3 +1516,7 @@ class BartStandaloneDecoderModelTest(ModelTesterMixin, GenerationTesterMixin, un
 
     def test_save_load_fast_init_from_base(self):
         pass
+
+    @unittest.skip("The model doesn't support left padding")  # and it's not used enough to be worth fixing :)
+    def test_left_padding_compatibility(self):
+        pass

tests/models/bigbird_pegasus/test_modeling_bigbird_pegasus.py

@@ -811,3 +811,7 @@ class BigBirdPegasusStandaloneDecoderModelTest(ModelTesterMixin, GenerationTeste
     def test_retain_grad_hidden_states_attentions(self):
         # decoder cannot keep gradients
         return
+
+    @unittest.skip("The model doesn't support left padding")  # and it's not used enough to be worth fixing :)
+    def test_left_padding_compatibility(self):
+        pass

tests/models/blenderbot/test_modeling_blenderbot.py

@@ -561,3 +561,7 @@ class BlenderbotStandaloneDecoderModelTest(ModelTesterMixin, GenerationTesterMix
     def test_retain_grad_hidden_states_attentions(self):
         # decoder cannot keep gradients
         return
+
+    @unittest.skip("The model doesn't support left padding")  # and it's not used enough to be worth fixing :)
+    def test_left_padding_compatibility(self):
+        pass

tests/models/blenderbot_small/test_modeling_blenderbot_small.py

@@ -566,3 +566,7 @@ class BlenderbotSmallStandaloneDecoderModelTest(ModelTesterMixin, GenerationTest
     def test_retain_grad_hidden_states_attentions(self):
         # decoder cannot keep gradients
         return
+
+    @unittest.skip("The model doesn't support left padding")  # and it's not used enough to be worth fixing :)
+    def test_left_padding_compatibility(self):
+        pass

tests/models/ctrl/test_modeling_ctrl.py

@@ -249,6 +249,10 @@ class CTRLModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixin
             model = CTRLModel.from_pretrained(model_name)
             self.assertIsNotNone(model)
 
+    @unittest.skip("The model doesn't support left padding")  # and it's not used enough to be worth fixing :)
+    def test_left_padding_compatibility(self):
+        pass
+
 
 @require_torch
 class CTRLModelLanguageGenerationTest(unittest.TestCase):

tests/models/imagegpt/test_modeling_imagegpt.py

@@ -520,6 +520,10 @@ class ImageGPTModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterM
 
             self.assertTrue(models_equal)
 
+    @unittest.skip("The model doesn't support left padding")  # and it's not used enough to be worth fixing :)
+    def test_left_padding_compatibility(self):
+        pass
+
 
 # We will verify our results on an image of cute cats
 def prepare_img():

tests/models/marian/test_modeling_marian.py

@@ -862,3 +862,7 @@ class MarianStandaloneDecoderModelTest(ModelTesterMixin, GenerationTesterMixin,
     def test_retain_grad_hidden_states_attentions(self):
         # decoder cannot keep gradients
         return
+
+    @unittest.skip("The model doesn't support left padding")  # and it's not used enough to be worth fixing :)
+    def test_left_padding_compatibility(self):
+        pass

tests/models/mbart/test_modeling_mbart.py

@@ -692,3 +692,7 @@ class MBartStandaloneDecoderModelTest(ModelTesterMixin, GenerationTesterMixin, u
     def test_retain_grad_hidden_states_attentions(self):
         # decoder cannot keep gradients
         return
+
+    @unittest.skip("The model doesn't support left padding")  # and it's not used enough to be worth fixing :)
+    def test_left_padding_compatibility(self):
+        pass

tests/models/mvp/test_modeling_mvp.py

@@ -818,3 +818,7 @@ class MvpStandaloneDecoderModelTest(ModelTesterMixin, GenerationTesterMixin, uni
     def test_retain_grad_hidden_states_attentions(self):
         # decoder cannot keep gradients
         return
+
+    @unittest.skip("The model doesn't support left padding")  # and it's not used enough to be worth fixing :)
+    def test_left_padding_compatibility(self):
+        pass

tests/models/pegasus/test_modeling_pegasus.py

@@ -573,3 +573,7 @@ class PegasusStandaloneDecoderModelTest(ModelTesterMixin, GenerationTesterMixin,
     def test_retain_grad_hidden_states_attentions(self):
         # decoder cannot keep gradients
         return
+
+    @unittest.skip("The model doesn't support left padding")  # and it's not used enough to be worth fixing :)
+    def test_left_padding_compatibility(self):
+        pass

tests/models/plbart/test_modeling_plbart.py

@@ -658,3 +658,7 @@ class PLBartStandaloneDecoderModelTest(ModelTesterMixin, GenerationTesterMixin,
     def test_retain_grad_hidden_states_attentions(self):
         # decoder cannot keep gradients
         return
+
+    @unittest.skip("The model doesn't support left padding")  # and it's not used enough to be worth fixing :)
+    def test_left_padding_compatibility(self):
+        pass

tests/models/prophetnet/test_modeling_prophetnet.py

@@ -1146,6 +1146,10 @@ class ProphetNetStandaloneDecoderModelTest(ModelTesterMixin, GenerationTesterMix
         # decoder cannot keep gradients
         return
 
+    @unittest.skip("The model doesn't support left padding")  # and it's not used enough to be worth fixing :)
+    def test_left_padding_compatibility(self):
+        pass
+
 
 @require_torch
 class ProphetNetStandaloneEncoderModelTest(ModelTesterMixin, unittest.TestCase):

tests/models/reformer/test_modeling_reformer.py

@@ -682,6 +682,10 @@ class ReformerLocalAttnModelTest(ReformerTesterMixin, GenerationTesterMixin, Mod
                 [expected_shape] * len(iter_hidden_states),
             )
 
+    @unittest.skip("The model doesn't support left padding")  # and it's not used enough to be worth fixing :)
+    def test_left_padding_compatibility(self):
+        pass
+
 
 @require_torch
 class ReformerLSHAttnModelTest(
@@ -839,6 +843,10 @@ class ReformerLSHAttnModelTest(
     def test_past_key_values_format(self):
         pass
 
+    @unittest.skip("The model doesn't support left padding")  # and it's not used enough to be worth fixing :)
+    def test_left_padding_compatibility(self):
+        pass
+
 
 @require_torch
 @require_sentencepiece

tests/models/transfo_xl/test_modeling_transfo_xl.py

@@ -486,6 +486,10 @@ class TransfoXLModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTester
                 weight = getattr(module, param)
                 weight.data.fill_(3)
 
+    @unittest.skip("The model doesn't support left padding")  # and it's not used enough to be worth fixing :)
+    def test_left_padding_compatibility(self):
+        pass
+
 
 @require_torch
 class TransfoXLModelLanguageGenerationTest(unittest.TestCase):

tests/models/trocr/test_modeling_trocr.py

@@ -192,3 +192,7 @@ class TrOCRStandaloneDecoderModelTest(ModelTesterMixin, GenerationTesterMixin, P
     # decoder cannot keep gradients
     def test_retain_grad_hidden_states_attentions(self):
         return
+
+    @unittest.skip("The model doesn't support left padding")  # and it's not used enough to be worth fixing :)
+    def test_left_padding_compatibility(self):
+        pass