Revert "[Tests] fix attention masks in Tests (huggingface#6621)"

This reverts commit d608dd6.
fabiocapsouza · Nov 15, 2020 · 4e36949 · 4e36949
1 parent 5e04391
commit 4e36949
Show file tree

Hide file tree

Showing 15 changed files with 31 additions and 35 deletions.
diff --git a/tests/test_modeling_albert.py b/tests/test_modeling_albert.py
@@ -20,7 +20,7 @@
 from transformers.testing_utils import require_torch, slow, torch_device
 
 from .test_configuration_common import ConfigTester
-from .test_modeling_common import ModelTesterMixin, ids_tensor, random_attention_mask
+from .test_modeling_common import ModelTesterMixin, ids_tensor
 
 
 if is_torch_available():
@@ -71,7 +71,7 @@ def prepare_config_and_inputs(self):
 
         input_mask = None
         if self.use_input_mask:
-            input_mask = random_attention_mask([self.batch_size, self.seq_length])
+            input_mask = ids_tensor([self.batch_size, self.seq_length], vocab_size=2)
 
         token_type_ids = None
         if self.use_token_type_ids:

diff --git a/tests/test_modeling_bert.py b/tests/test_modeling_bert.py
@@ -20,7 +20,7 @@
 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, random_attention_mask
+from .test_modeling_common import ModelTesterMixin, floats_tensor, ids_tensor
 
 
 if is_torch_available():
@@ -93,7 +93,7 @@ def prepare_config_and_inputs(self):
 
         input_mask = None
         if self.use_input_mask:
-            input_mask = random_attention_mask([self.batch_size, self.seq_length])
+            input_mask = ids_tensor([self.batch_size, self.seq_length], vocab_size=2)
 
         token_type_ids = None
         if self.use_token_type_ids:

diff --git a/tests/test_modeling_common.py b/tests/test_modeling_common.py
@@ -704,6 +704,9 @@ def recursive_check(tuple_object, dict_object):
                             recursive_check(tuple_iterable_value, dict_iterable_value)
                     elif tuple_object is None:
                         return
+                    elif torch.isinf(tuple_object).any() and torch.isinf(dict_object).any():
+                        # TODO: (Lysandre) - maybe take a look if that's ok here
+                        return
                     else:
                         self.assertTrue(
                             torch.allclose(tuple_object, dict_object, atol=1e-5),
@@ -934,13 +937,6 @@ def ids_tensor(shape, vocab_size, rng=None, name=None):
     return torch.tensor(data=values, dtype=torch.long, device=torch_device).view(shape).contiguous()
 
 
-def random_attention_mask(shape, rng=None, name=None):
-    attn_mask = ids_tensor(shape, vocab_size=2, rng=None, name=None)
-    # make sure that at least one token is attended to for each batch
-    attn_mask[:, -1] = 1
-    return attn_mask
-
-
 def floats_tensor(shape, scale=1.0, rng=None, name=None):
     """Creates a random float32 tensor"""
     if rng is None:

diff --git a/tests/test_modeling_ctrl.py b/tests/test_modeling_ctrl.py
@@ -19,7 +19,7 @@
 from transformers.testing_utils import require_torch, slow, torch_device
 
 from .test_configuration_common import ConfigTester
-from .test_modeling_common import ModelTesterMixin, ids_tensor, random_attention_mask
+from .test_modeling_common import ModelTesterMixin, ids_tensor
 
 
 if is_torch_available():
@@ -60,7 +60,7 @@ def prepare_config_and_inputs(self):
 
         input_mask = None
         if self.use_input_mask:
-            input_mask = random_attention_mask([self.batch_size, self.seq_length])
+            input_mask = ids_tensor([self.batch_size, self.seq_length], vocab_size=2)
 
         token_type_ids = None
         if self.use_token_type_ids:

diff --git a/tests/test_modeling_distilbert.py b/tests/test_modeling_distilbert.py
@@ -20,7 +20,7 @@
 from transformers.testing_utils import require_torch, slow, torch_device
 
 from .test_configuration_common import ConfigTester
-from .test_modeling_common import ModelTesterMixin, ids_tensor, random_attention_mask
+from .test_modeling_common import ModelTesterMixin, ids_tensor
 
 
 if is_torch_available():
@@ -89,7 +89,7 @@ def prepare_config_and_inputs(self):
 
             input_mask = None
             if self.use_input_mask:
-                input_mask = random_attention_mask([self.batch_size, self.seq_length])
+                input_mask = ids_tensor([self.batch_size, self.seq_length], vocab_size=2)
 
             sequence_labels = None
             token_labels = None

diff --git a/tests/test_modeling_dpr.py b/tests/test_modeling_dpr.py
@@ -20,7 +20,7 @@
 from transformers.testing_utils import require_torch, slow, torch_device
 
 from .test_configuration_common import ConfigTester
-from .test_modeling_common import ModelTesterMixin, ids_tensor, random_attention_mask
+from .test_modeling_common import ModelTesterMixin, ids_tensor
 
 
 if is_torch_available():
@@ -88,7 +88,7 @@ def prepare_config_and_inputs(self):
 
         input_mask = None
         if self.use_input_mask:
-            input_mask = random_attention_mask([self.batch_size, self.seq_length])
+            input_mask = ids_tensor([self.batch_size, self.seq_length], vocab_size=2)
 
         token_type_ids = None
         if self.use_token_type_ids:

diff --git a/tests/test_modeling_electra.py b/tests/test_modeling_electra.py
@@ -20,7 +20,7 @@
 from transformers.testing_utils import require_torch, slow, torch_device
 
 from .test_configuration_common import ConfigTester
-from .test_modeling_common import ModelTesterMixin, ids_tensor, random_attention_mask
+from .test_modeling_common import ModelTesterMixin, ids_tensor
 
 
 if is_torch_available():
@@ -69,7 +69,7 @@ def prepare_config_and_inputs(self):
 
         input_mask = None
         if self.use_input_mask:
-            input_mask = random_attention_mask([self.batch_size, self.seq_length])
+            input_mask = ids_tensor([self.batch_size, self.seq_length], vocab_size=2)
 
         token_type_ids = None
         if self.use_token_type_ids:

diff --git a/tests/test_modeling_flaubert.py b/tests/test_modeling_flaubert.py
@@ -20,7 +20,7 @@
 from transformers.testing_utils import require_torch, slow, torch_device
 
 from .test_configuration_common import ConfigTester
-from .test_modeling_common import ModelTesterMixin, ids_tensor, random_attention_mask
+from .test_modeling_common import ModelTesterMixin, ids_tensor
 
 
 if is_torch_available():
@@ -72,7 +72,7 @@ def __init__(
 
     def prepare_config_and_inputs(self):
         input_ids = ids_tensor([self.batch_size, self.seq_length], self.vocab_size)
-        input_mask = random_attention_mask([self.batch_size, self.seq_length])
+        input_mask = ids_tensor([self.batch_size, self.seq_length], 2).float()
 
         input_lengths = None
         if self.use_input_lengths:

diff --git a/tests/test_modeling_gpt2.py b/tests/test_modeling_gpt2.py
@@ -20,7 +20,7 @@
 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, random_attention_mask
+from .test_modeling_common import ModelTesterMixin, floats_tensor, ids_tensor
 
 
 if is_torch_available():
@@ -92,7 +92,7 @@ def prepare_config_and_inputs(self):
 
         input_mask = None
         if self.use_input_mask:
-            input_mask = random_attention_mask([self.batch_size, self.seq_length])
+            input_mask = ids_tensor([self.batch_size, self.seq_length], vocab_size=2)
 
         token_type_ids = None
         if self.use_token_type_ids:

diff --git a/tests/test_modeling_longformer.py b/tests/test_modeling_longformer.py
@@ -20,7 +20,7 @@
 from transformers.testing_utils import require_torch, slow, torch_device
 
 from .test_configuration_common import ConfigTester
-from .test_modeling_common import ModelTesterMixin, ids_tensor, random_attention_mask
+from .test_modeling_common import ModelTesterMixin, ids_tensor
 
 
 if is_torch_available():
@@ -82,7 +82,7 @@ def prepare_config_and_inputs(self):
 
         input_mask = None
         if self.use_input_mask:
-            input_mask = random_attention_mask([self.batch_size, self.seq_length])
+            input_mask = ids_tensor([self.batch_size, self.seq_length], vocab_size=2)
 
         token_type_ids = None
         if self.use_token_type_ids:

diff --git a/tests/test_modeling_mobilebert.py b/tests/test_modeling_mobilebert.py
@@ -20,7 +20,7 @@
 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, random_attention_mask
+from .test_modeling_common import ModelTesterMixin, floats_tensor, ids_tensor
 
 
 if is_torch_available():
@@ -94,7 +94,7 @@ def prepare_config_and_inputs(self):
 
         input_mask = None
         if self.use_input_mask:
-            input_mask = random_attention_mask([self.batch_size, self.seq_length])
+            input_mask = ids_tensor([self.batch_size, self.seq_length], vocab_size=2)
 
         token_type_ids = None
         if self.use_token_type_ids:

diff --git a/tests/test_modeling_reformer.py b/tests/test_modeling_reformer.py
@@ -19,7 +19,7 @@
 from transformers.testing_utils import require_multigpu, require_torch, slow, torch_device
 
 from .test_configuration_common import ConfigTester
-from .test_modeling_common import ModelTesterMixin, floats_tensor, ids_tensor, random_attention_mask
+from .test_modeling_common import ModelTesterMixin, floats_tensor, ids_tensor
 
 
 if is_torch_available():
@@ -133,7 +133,7 @@ def prepare_config_and_inputs(self):
 
         input_mask = None
         if self.use_input_mask:
-            input_mask = random_attention_mask([self.batch_size, self.seq_length])
+            input_mask = ids_tensor([self.batch_size, self.seq_length], vocab_size=2)
 
         choice_labels = None
         if self.use_labels:

diff --git a/tests/test_modeling_roberta.py b/tests/test_modeling_roberta.py
@@ -20,7 +20,7 @@
 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, random_attention_mask
+from .test_modeling_common import ModelTesterMixin, floats_tensor, ids_tensor
 
 
 if is_torch_available():
@@ -71,7 +71,7 @@ def prepare_config_and_inputs(self):
 
         input_mask = None
         if self.use_input_mask:
-            input_mask = random_attention_mask([self.batch_size, self.seq_length])
+            input_mask = ids_tensor([self.batch_size, self.seq_length], vocab_size=2)
 
         token_type_ids = None
         if self.use_token_type_ids:

diff --git a/tests/test_modeling_xlm.py b/tests/test_modeling_xlm.py
@@ -20,7 +20,7 @@
 from transformers.testing_utils import require_torch, slow, torch_device
 
 from .test_configuration_common import ConfigTester
-from .test_modeling_common import ModelTesterMixin, ids_tensor, random_attention_mask
+from .test_modeling_common import ModelTesterMixin, ids_tensor
 
 
 if is_torch_available():
@@ -73,7 +73,7 @@ def __init__(
 
     def prepare_config_and_inputs(self):
         input_ids = ids_tensor([self.batch_size, self.seq_length], self.vocab_size)
-        input_mask = random_attention_mask([self.batch_size, self.seq_length])
+        input_mask = ids_tensor([self.batch_size, self.seq_length], 2).float()
 
         input_lengths = None
         if self.use_input_lengths:

diff --git a/tests/test_modeling_xlnet.py b/tests/test_modeling_xlnet.py
@@ -21,7 +21,7 @@
 from transformers.testing_utils import require_torch, slow, torch_device
 
 from .test_configuration_common import ConfigTester
-from .test_modeling_common import ModelTesterMixin, ids_tensor, random_attention_mask
+from .test_modeling_common import ModelTesterMixin, ids_tensor
 
 
 if is_torch_available():
@@ -100,7 +100,7 @@ def prepare_config_and_inputs(self):
         input_ids_1 = ids_tensor([self.batch_size, self.seq_length], self.vocab_size)
         input_ids_2 = ids_tensor([self.batch_size, self.seq_length], self.vocab_size)
         segment_ids = ids_tensor([self.batch_size, self.seq_length], self.type_vocab_size)
-        input_mask = random_attention_mask([self.batch_size, self.seq_length])
+        input_mask = ids_tensor([self.batch_size, self.seq_length], 2).float()
 
         input_ids_q = ids_tensor([self.batch_size, self.seq_length + 1], self.vocab_size)
         perm_mask = torch.zeros(