Skip to content

GitLab

Commit a75c64d8 authored by Lysandre's avatar Lysandre
Browse files

Black 20 release

parent e78c1103
Show whitespace changes
Inline Side-by-side
Showing with 228 additions and 67 deletions
tests/test_modeling_longformer.py
View file @ a75c64d8
......@@ -40,7 +40,8 @@ if is_torch_available():
class LongformerModelTester:
def __init__(
self, parent,
self,
parent,
):
self.parent = parent
self.batch_size = 13
......
tests/test_modeling_mobilebert.py
View file @ a75c64d8
......@@ -217,7 +217,10 @@ class MobileBertModelTester:
model.to(torch_device)
model.eval()
result = model(
input_ids, attention_mask=input_mask, token_type_ids=token_type_ids, next_sentence_label=sequence_labels,
input_ids,
attention_mask=input_mask,
token_type_ids=token_type_ids,
next_sentence_label=sequence_labels,
)
self.parent.assertEqual(result.logits.shape, (self.batch_size, 2))
......@@ -397,7 +400,11 @@ class MobileBertModelTest(ModelTesterMixin, unittest.TestCase):
def _long_tensor(tok_lst):
return torch.tensor(tok_lst, dtype=torch.long, device=torch_device,)
return torch.tensor(
tok_lst,
dtype=torch.long,
device=torch_device,
)
TOLERANCE = 1e-3
......
tests/test_modeling_openai.py
View file @ a75c64d8
......@@ -37,7 +37,8 @@ if is_torch_available():
class OpenAIGPTModelTester:
def __init__(
self, parent,
self,
parent,
):
self.parent = parent
self.batch_size = 13
......
tests/test_modeling_reformer.py
View file @ a75c64d8
......@@ -240,14 +240,19 @@ class ReformerModelTester:
half_input_ids = input_ids[:, :half_seq_len]
# normal padded
attn_mask = torch.cat([torch.ones_like(half_input_ids), torch.zeros_like(half_input_ids)], dim=-1,)
attn_mask = torch.cat(
[torch.ones_like(half_input_ids), torch.zeros_like(half_input_ids)],
dim=-1,
)
input_ids_padded = torch.cat(
[half_input_ids, ids_tensor((self.batch_size, half_seq_len), self.vocab_size)], dim=-1,
[half_input_ids, ids_tensor((self.batch_size, half_seq_len), self.vocab_size)],
dim=-1,
)
# shifted padded
input_ids_roll = torch.cat(
[half_input_ids, ids_tensor((self.batch_size, half_seq_len), self.vocab_size)], dim=-1,
[half_input_ids, ids_tensor((self.batch_size, half_seq_len), self.vocab_size)],
dim=-1,
)
input_ids_roll = torch.roll(input_ids_roll, roll, dims=-1)
attn_mask_roll = torch.roll(attn_mask, roll, dims=-1)
......@@ -283,13 +288,21 @@ class ReformerModelTester:
torch.manual_seed(layer.attention_seed)
attn_outputs = layer.attention(hidden_states, attention_mask=input_mask)
self.parent.assertTrue(
torch.allclose(prev_attn_output + attn_outputs.hidden_states, next_attn_output, atol=1e-3,)
torch.allclose(
prev_attn_output + attn_outputs.hidden_states,
next_attn_output,
atol=1e-3,
)
)
torch.manual_seed(layer.feed_forward_seed)
feed_forward_hidden_states = layer.feed_forward(next_attn_output)
self.parent.assertTrue(
torch.allclose(next_hidden_states, hidden_states + feed_forward_hidden_states, atol=1e-3,)
torch.allclose(
next_hidden_states,
hidden_states + feed_forward_hidden_states,
atol=1e-3,
)
)
def create_and_check_reformer_feed_backward_chunking(self, config, input_ids, input_mask, choice_labels):
......@@ -416,7 +429,10 @@ class ReformerModelTester:
model.to(torch_device)
model.eval()
result = model(
input_ids, attention_mask=input_mask, start_positions=choice_labels, end_positions=choice_labels,
input_ids,
attention_mask=input_mask,
start_positions=choice_labels,
end_positions=choice_labels,
)
self.parent.assertEqual(result.start_logits.shape, (self.batch_size, self.seq_length))
self.parent.assertEqual(result.end_logits.shape, (self.batch_size, self.seq_length))
......@@ -887,7 +903,9 @@ class ReformerIntegrationTests(unittest.TestCase):
reformer_output = layer(prev_attn_output=hidden_states.clone(), hidden_states=hidden_states)
output_slice = reformer_output.hidden_states[0, 0, :5]
expected_output_slice = torch.tensor(
[1.6879, -1.3083, -0.4708, 1.3555, -0.6292], dtype=torch.float, device=torch_device,
[1.6879, -1.3083, -0.4708, 1.3555, -0.6292],
dtype=torch.float,
device=torch_device,
)
self.assertTrue(torch.allclose(output_slice, expected_output_slice, atol=1e-3))
......@@ -902,11 +920,15 @@ class ReformerIntegrationTests(unittest.TestCase):
layer = ReformerLayer(ReformerConfig(**config)).to(torch_device)
layer.eval()
reformer_output = layer(
prev_attn_output=hidden_states.clone(), hidden_states=hidden_states, attention_mask=attn_mask,
prev_attn_output=hidden_states.clone(),
hidden_states=hidden_states,
attention_mask=attn_mask,
)
output_slice = reformer_output.hidden_states[0, 0, :5]
expected_output_slice = torch.tensor(
[1.6439, -1.2306, -0.5108, 1.3006, -0.6537], dtype=torch.float, device=torch_device,
[1.6439, -1.2306, -0.5108, 1.3006, -0.6537],
dtype=torch.float,
device=torch_device,
)
self.assertTrue(torch.allclose(output_slice, expected_output_slice, atol=1e-3))
......@@ -922,7 +944,9 @@ class ReformerIntegrationTests(unittest.TestCase):
reformer_output = layer(prev_attn_output=hidden_states, hidden_states=hidden_states)
output_slice = reformer_output.hidden_states[0, 0, :5]
expected_output_slice = torch.tensor(
[1.4212, -2.0576, -0.9688, 1.4599, -0.1344], dtype=torch.float, device=torch_device,
[1.4212, -2.0576, -0.9688, 1.4599, -0.1344],
dtype=torch.float,
device=torch_device,
)
self.assertTrue(torch.allclose(output_slice, expected_output_slice, atol=1e-3))
......@@ -935,10 +959,16 @@ class ReformerIntegrationTests(unittest.TestCase):
torch.manual_seed(0)
layer = ReformerLayer(ReformerConfig(**config)).to(torch_device)
layer.eval()
reformer_output = layer(prev_attn_output=hidden_states, hidden_states=hidden_states, attention_mask=attn_mask,)
reformer_output = layer(
prev_attn_output=hidden_states,
hidden_states=hidden_states,
attention_mask=attn_mask,
)
output_slice = reformer_output.hidden_states[0, 0, :5]
expected_output_slice = torch.tensor(
[1.4750, -2.0235, -0.9743, 1.4463, -0.1269], dtype=torch.float, device=torch_device,
[1.4750, -2.0235, -0.9743, 1.4463, -0.1269],
dtype=torch.float,
device=torch_device,
)
self.assertTrue(torch.allclose(output_slice, expected_output_slice, atol=1e-3))
......@@ -953,7 +983,9 @@ class ReformerIntegrationTests(unittest.TestCase):
hidden_states = model(input_ids=input_ids, attention_mask=attn_mask)[0]
output_slice = hidden_states[0, 0, :5]
expected_output_slice = torch.tensor(
[-0.9896, -0.9396, -1.0831, -0.0597, 0.2456], dtype=torch.float, device=torch_device,
[-0.9896, -0.9396, -1.0831, -0.0597, 0.2456],
dtype=torch.float,
device=torch_device,
)
self.assertTrue(torch.allclose(output_slice, expected_output_slice, atol=1e-3))
......@@ -967,7 +999,9 @@ class ReformerIntegrationTests(unittest.TestCase):
hidden_states = model(input_ids=input_ids, attention_mask=attn_mask)[0]
output_slice = hidden_states[0, 0, :5]
expected_output_slice = torch.tensor(
[-1.6791, 0.7171, 0.1594, 0.4063, 1.2584], dtype=torch.float, device=torch_device,
[-1.6791, 0.7171, 0.1594, 0.4063, 1.2584],
dtype=torch.float,
device=torch_device,
)
self.assertTrue(torch.allclose(output_slice, expected_output_slice, atol=1e-3))
......@@ -983,7 +1017,9 @@ class ReformerIntegrationTests(unittest.TestCase):
hidden_states = model(input_ids=input_ids, attention_mask=attn_mask)[0]
output_slice = hidden_states[1, -1, :5]
expected_output_slice = torch.tensor(
[0.0256, -0.0121, 0.0636, 0.0024, -0.0393], dtype=torch.float, device=torch_device,
[0.0256, -0.0121, 0.0636, 0.0024, -0.0393],
dtype=torch.float,
device=torch_device,
)
self.assertTrue(torch.allclose(output_slice, expected_output_slice, atol=1e-3))
......@@ -1005,15 +1041,21 @@ class ReformerIntegrationTests(unittest.TestCase):
# check last grads to cover all proable errors
grad_slice_word = model.reformer.embeddings.word_embeddings.weight.grad[0, :5]
expected_grad_slice_word = torch.tensor(
[-0.0005, 0.0001, 0.0002, 0.0003, 0.0006], dtype=torch.float, device=torch_device,
[-0.0005, 0.0001, 0.0002, 0.0003, 0.0006],
dtype=torch.float,
device=torch_device,
)
grad_slice_position_factor_1 = model.reformer.embeddings.position_embeddings.weights[0][1, 0, -5:]
expected_grad_slice_pos_fac_1 = torch.tensor(
[0.0037, -1.3793, -1.0231, -1.5230, -2.5306], dtype=torch.float, device=torch_device,
[0.0037, -1.3793, -1.0231, -1.5230, -2.5306],
dtype=torch.float,
device=torch_device,
)
grad_slice_position_factor_2 = model.reformer.embeddings.position_embeddings.weights[1][0, 1, :5]
expected_grad_slice_pos_fac_2 = torch.tensor(
[-1.3165, 0.5168, 0.7785, 1.0811, -0.9830], dtype=torch.float, device=torch_device,
[-1.3165, 0.5168, 0.7785, 1.0811, -0.9830],
dtype=torch.float,
device=torch_device,
)
self.assertTrue(torch.allclose(grad_slice_word, expected_grad_slice_word, atol=1e-3))
self.assertTrue(torch.allclose(grad_slice_position_factor_1, expected_grad_slice_pos_fac_1, atol=1e-3))
......@@ -1038,15 +1080,21 @@ class ReformerIntegrationTests(unittest.TestCase):
# check last grads to cover all proable errors
grad_slice_word = model.reformer.embeddings.word_embeddings.weight.grad[0, :5]
expected_grad_slice_word = torch.tensor(
[2.6357e-05, 4.3358e-04, -8.4985e-04, 1.0094e-04, 3.8954e-04], dtype=torch.float, device=torch_device,
[2.6357e-05, 4.3358e-04, -8.4985e-04, 1.0094e-04, 3.8954e-04],
dtype=torch.float,
device=torch_device,
)
grad_slice_position_factor_1 = model.reformer.embeddings.position_embeddings.weights[0][1, 0, -5:]
expected_grad_slice_pos_fac_1 = torch.tensor(
[-0.0984, 0.6283, 0.4282, 1.2960, 0.6897], dtype=torch.float, device=torch_device,
[-0.0984, 0.6283, 0.4282, 1.2960, 0.6897],
dtype=torch.float,
device=torch_device,
)
grad_slice_position_factor_2 = model.reformer.embeddings.position_embeddings.weights[1][0, 1, :5]
expected_grad_slice_pos_fac_2 = torch.tensor(
[0.4626, -0.0231, -0.0172, 0.1081, 0.3805], dtype=torch.float, device=torch_device,
[0.4626, -0.0231, -0.0172, 0.1081, 0.3805],
dtype=torch.float,
device=torch_device,
)
self.assertTrue(torch.allclose(grad_slice_word, expected_grad_slice_word, atol=1e-3))
self.assertTrue(torch.allclose(grad_slice_position_factor_1, expected_grad_slice_pos_fac_1, atol=1e-3))
......
tests/test_modeling_roberta.py
View file @ a75c64d8
......@@ -45,7 +45,8 @@ if is_torch_available():
class RobertaModelTester:
def __init__(
self, parent,
self,
parent,
):
self.parent = parent
self.batch_size = 13
......@@ -352,7 +353,7 @@ class RobertaModelTest(ModelTesterMixin, unittest.TestCase):
self.assertIsNotNone(model)
def test_create_position_ids_respects_padding_index(self):
""" Ensure that the default position ids only assign a sequential . This is a regression
"""Ensure that the default position ids only assign a sequential . This is a regression
test for https://github.com/huggingface/transformers/issues/1761
The position ids should be masked with the embedding object's padding index. Therefore, the
......@@ -371,7 +372,7 @@ class RobertaModelTest(ModelTesterMixin, unittest.TestCase):
self.assertTrue(torch.all(torch.eq(position_ids, expected_positions)))
def test_create_position_ids_from_inputs_embeds(self):
""" Ensure that the default position ids only assign a sequential . This is a regression
"""Ensure that the default position ids only assign a sequential . This is a regression
test for https://github.com/huggingface/transformers/issues/1761
The position ids should be masked with the embedding object's padding index. Therefore, the
......
tests/test_modeling_t5.py
View file @ a75c64d8
......@@ -101,7 +101,13 @@ class T5ModelTester:
)
def check_prepare_lm_labels_via_shift_left(
self, config, input_ids, decoder_input_ids, attention_mask, decoder_attention_mask, lm_labels,
self,
config,
input_ids,
decoder_input_ids,
attention_mask,
decoder_attention_mask,
lm_labels,
):
model = T5Model(config=config)
model.to(torch_device)
......@@ -134,7 +140,13 @@ class T5ModelTester:
self.parent.assertListEqual(decoder_input_ids_slice[1:].tolist(), lm_labels_slice[:-1].tolist())
def create_and_check_model(
self, config, input_ids, decoder_input_ids, attention_mask, decoder_attention_mask, lm_labels,
self,
config,
input_ids,
decoder_input_ids,
attention_mask,
decoder_attention_mask,
lm_labels,
):
model = T5Model(config=config)
model.to(torch_device)
......@@ -160,7 +172,13 @@ class T5ModelTester:
self.parent.assertEqual(len(decoder_past[1][0]), 4)
def create_and_check_with_lm_head(
self, config, input_ids, decoder_input_ids, attention_mask, decoder_attention_mask, lm_labels,
self,
config,
input_ids,
decoder_input_ids,
attention_mask,
decoder_attention_mask,
lm_labels,
):
model = T5ForConditionalGeneration(config=config).to(torch_device).eval()
outputs = model(
......@@ -174,7 +192,13 @@ class T5ModelTester:
self.parent.assertEqual(outputs["loss"].size(), ())
def create_and_check_decoder_model_past(
self, config, input_ids, decoder_input_ids, attention_mask, decoder_attention_mask, lm_labels,
self,
config,
input_ids,
decoder_input_ids,
attention_mask,
decoder_attention_mask,
lm_labels,
):
model = T5Model(config=config).get_decoder().to(torch_device).eval()
# first forward pass
......@@ -205,7 +229,13 @@ class T5ModelTester:
self.parent.assertTrue(torch.allclose(output_from_past_slice, output_from_no_past_slice, atol=1e-3))
def create_and_check_decoder_model_attention_mask_past(
self, config, input_ids, decoder_input_ids, attention_mask, decoder_attention_mask, lm_labels,
self,
config,
input_ids,
decoder_input_ids,
attention_mask,
decoder_attention_mask,
lm_labels,
):
model = T5Model(config=config).get_decoder()
model.to(torch_device)
......@@ -231,7 +261,8 @@ class T5ModelTester:
# append to next input_ids and attn_mask
next_input_ids = torch.cat([input_ids, next_tokens], dim=-1)
attn_mask = torch.cat(
[attn_mask, torch.ones((attn_mask.shape[0], 1), dtype=torch.long, device=torch_device)], dim=1,
[attn_mask, torch.ones((attn_mask.shape[0], 1), dtype=torch.long, device=torch_device)],
dim=1,
)
# get two different outputs
......@@ -249,7 +280,13 @@ class T5ModelTester:
self.parent.assertTrue(torch.allclose(output_from_past_slice, output_from_no_past_slice, atol=1e-3))
def create_and_check_generate_with_past_key_value_states(
self, config, input_ids, decoder_input_ids, attention_mask, decoder_attention_mask, lm_labels,
self,
config,
input_ids,
decoder_input_ids,
attention_mask,
decoder_attention_mask,
lm_labels,
):
model = T5ForConditionalGeneration(config=config).to(torch_device).eval()
torch.manual_seed(0)
......@@ -261,14 +298,26 @@ class T5ModelTester:
self.parent.assertTrue(torch.all(output_with_past_cache == output_without_past_cache))
def create_and_check_model_fp16_forward(
self, config, input_ids, decoder_input_ids, attention_mask, decoder_attention_mask, lm_labels,
self,
config,
input_ids,
decoder_input_ids,
attention_mask,
decoder_attention_mask,
lm_labels,
):
model = T5Model(config=config).to(torch_device).half().eval()
output = model(input_ids, decoder_input_ids=input_ids, attention_mask=attention_mask)["last_hidden_state"]
self.parent.assertFalse(torch.isnan(output).any().item())
def create_and_check_encoder_decoder_shared_weights(
self, config, input_ids, decoder_input_ids, attention_mask, decoder_attention_mask, lm_labels,
self,
config,
input_ids,
decoder_input_ids,
attention_mask,
decoder_attention_mask,
lm_labels,
):
for model_class in [T5Model, T5ForConditionalGeneration]:
torch.manual_seed(0)
......@@ -339,7 +388,14 @@ class T5ModelTester:
def prepare_config_and_inputs_for_common(self):
config_and_inputs = self.prepare_config_and_inputs()
(config, input_ids, decoder_input_ids, attention_mask, decoder_attention_mask, lm_labels,) = config_and_inputs
(
config,
input_ids,
decoder_input_ids,
attention_mask,
decoder_attention_mask,
lm_labels,
) = config_and_inputs
inputs_dict = {
"input_ids": input_ids,
......@@ -412,7 +468,11 @@ class T5ModelTest(ModelTesterMixin, unittest.TestCase):
model = T5Model(config_and_inputs[0]).to(torch_device)
with tempfile.TemporaryDirectory() as tmpdirname:
torch.onnx.export(
model, config_and_inputs[1], f"{tmpdirname}/t5_test.onnx", export_params=True, opset_version=9,
model,
config_and_inputs[1],
f"{tmpdirname}/t5_test.onnx",
export_params=True,
opset_version=9,
)
......@@ -469,7 +529,8 @@ class T5ModelIntegrationTests(unittest.TestCase):
)
decoded = tok.batch_decode(hypotheses_batch, skip_special_tokens=True, clean_up_tokenization_spaces=False)
self.assertListEqual(
expected_summaries, decoded,
expected_summaries,
decoded,
)
@slow
......
tests/test_modeling_tf_camembert.py
View file @ a75c64d8
......@@ -33,7 +33,8 @@ class TFCamembertModelIntegrationTest(unittest.TestCase):
model = TFCamembertModel.from_pretrained("jplu/tf-camembert-base")
input_ids = tf.convert_to_tensor(
[[5, 121, 11, 660, 16, 730, 25543, 110, 83, 6]], dtype=tf.int32,
[[5, 121, 11, 660, 16, 730, 25543, 110, 83, 6]],
dtype=tf.int32,
) # J'aime le camembert !"
output = model(input_ids)["last_hidden_state"]
......@@ -41,7 +42,8 @@ class TFCamembertModelIntegrationTest(unittest.TestCase):
self.assertEqual(output.shape, expected_shape)
# compare the actual values for a slice.
expected_slice = tf.convert_to_tensor(
[[[-0.0254, 0.0235, 0.1027], [0.0606, -0.1811, -0.0418], [-0.1561, -0.1127, 0.2687]]], dtype=tf.float32,
[[[-0.0254, 0.0235, 0.1027], [0.0606, -0.1811, -0.0418], [-0.1561, -0.1127, 0.2687]]],
dtype=tf.float32,
)
# camembert = torch.hub.load('pytorch/fairseq', 'camembert.v0')
# camembert.eval()
......
tests/test_modeling_tf_common.py
View file @ a75c64d8
......@@ -155,7 +155,8 @@ class TFModelTesterMixin:
self.assertEqual(len(outputs), num_out)
self.assertEqual(len(hidden_states), self.model_tester.num_hidden_layers + 1)
self.assertListEqual(
list(hidden_states[0].shape[-2:]), [self.model_tester.seq_length, self.model_tester.hidden_size],
list(hidden_states[0].shape[-2:]),
[self.model_tester.seq_length, self.model_tester.hidden_size],
)
@slow
......@@ -486,7 +487,8 @@ class TFModelTesterMixin:
hidden_states = [t.numpy() for t in outputs[-1]]
self.assertEqual(len(hidden_states), self.model_tester.num_hidden_layers + 1)
self.assertListEqual(
list(hidden_states[0].shape[-2:]), [self.model_tester.seq_length, self.model_tester.hidden_size],
list(hidden_states[0].shape[-2:]),
[self.model_tester.seq_length, self.model_tester.hidden_size],
)
for model_class in self.all_model_classes:
......@@ -591,9 +593,15 @@ class TFModelTesterMixin:
x = wte([input_ids, None, None, None], mode="embedding")
except Exception:
if hasattr(self.model_tester, "embedding_size"):
x = tf.ones(input_ids.shape + [self.model_tester.embedding_size], dtype=tf.dtypes.float32,)
x = tf.ones(
input_ids.shape + [self.model_tester.embedding_size],
dtype=tf.dtypes.float32,
)
else:
x = tf.ones(input_ids.shape + [self.model_tester.hidden_size], dtype=tf.dtypes.float32,)
x = tf.ones(
input_ids.shape + [self.model_tester.hidden_size],
dtype=tf.dtypes.float32,
)
return x
def test_inputs_embeds(self):
......@@ -700,7 +708,14 @@ class TFModelTesterMixin:
model.generate(input_ids, do_sample=False, num_return_sequences=3, num_beams=2)
# num_return_sequences > 1, sample
self._check_generated_ids(model.generate(input_ids, do_sample=True, num_beams=2, num_return_sequences=2,))
self._check_generated_ids(
model.generate(
input_ids,
do_sample=True,
num_beams=2,
num_return_sequences=2,
)
)
# num_return_sequences > 1, greedy
self._check_generated_ids(model.generate(input_ids, do_sample=False, num_beams=2, num_return_sequences=2))
......@@ -895,7 +910,8 @@ class UtilsFunctionsTest(unittest.TestCase):
)
non_inf_expected_idx = tf.convert_to_tensor(
[[0, 0], [0, 9], [0, 10], [0, 25], [0, 26], [1, 13], [1, 17], [1, 18], [1, 20], [1, 27]], dtype=tf.int32,
[[0, 0], [0, 9], [0, 10], [0, 25], [0, 26], [1, 13], [1, 17], [1, 18], [1, 20], [1, 27]],
dtype=tf.int32,
) # expected non filtered idx as noted above
non_inf_expected_output = tf.convert_to_tensor(
......@@ -907,7 +923,8 @@ class UtilsFunctionsTest(unittest.TestCase):
non_inf_output = output[output != -float("inf")]
non_inf_idx = tf.cast(
tf.where(tf.not_equal(output, tf.constant(-float("inf"), dtype=tf.float32))), dtype=tf.int32,
tf.where(tf.not_equal(output, tf.constant(-float("inf"), dtype=tf.float32))),
dtype=tf.int32,
)
tf.debugging.assert_near(non_inf_output, non_inf_expected_output, rtol=1e-12)
......
tests/test_modeling_tf_ctrl.py
View file @ a75c64d8
......@@ -31,7 +31,8 @@ if is_tf_available():
class TFCTRLModelTester(object):
def __init__(
self, parent,
self,
parent,
):
self.parent = parent
self.batch_size = 13
......
tests/test_modeling_tf_distilbert.py
View file @ a75c64d8
......@@ -39,7 +39,8 @@ if is_tf_available():
class TFDistilBertModelTester:
def __init__(
self, parent,
self,
parent,
):
self.parent = parent
self.batch_size = 13
......
tests/test_modeling_tf_electra.py
View file @ a75c64d8
......@@ -39,7 +39,8 @@ if is_tf_available():
class TFElectraModelTester:
def __init__(
self, parent,
self,
parent,
):
self.parent = parent
self.batch_size = 13
......
tests/test_modeling_tf_flaubert.py
View file @ a75c64d8
......@@ -40,7 +40,8 @@ if is_tf_available():
class TFFlaubertModelTester:
def __init__(
self, parent,
self,
parent,
):
self.parent = parent
self.batch_size = 13
......@@ -337,7 +338,8 @@ class TFFlaubertModelIntegrationTest(unittest.TestCase):
model = TFFlaubertModel.from_pretrained("jplu/tf-flaubert-small-cased")
input_ids = tf.convert_to_tensor(
[[0, 158, 735, 2592, 1424, 6727, 82, 1]], dtype=tf.int32,
[[0, 158, 735, 2592, 1424, 6727, 82, 1]],
dtype=tf.int32,
) # "J'aime flaubert !"
output = model(input_ids)[0]
......
tests/test_modeling_tf_gpt2.py
View file @ a75c64d8
......@@ -37,7 +37,8 @@ if is_tf_available():
class TFGPT2ModelTester:
def __init__(
self, parent,
self,
parent,
):
self.parent = parent
self.batch_size = 13
......
tests/test_modeling_tf_longformer.py
View file @ a75c64d8
......@@ -45,7 +45,8 @@ if is_tf_available():
class TFLongformerModelTester:
def __init__(
self, parent,
self,
parent,
):
self.parent = parent
self.batch_size = 13
......@@ -228,7 +229,8 @@ class TFLongformerModelTester:
# global attention mask has to be partly defined
# to trace all weights
global_attention_mask = tf.concat(
[tf.zeros_like(input_ids)[:, :-1], tf.ones_like(input_ids)[:, -1:]], axis=-1,
[tf.zeros_like(input_ids)[:, :-1], tf.ones_like(input_ids)[:, -1:]],
axis=-1,
)
inputs_dict = {
......@@ -267,7 +269,13 @@ class TFLongformerModelTest(TFModelTesterMixin, unittest.TestCase):
test_torchscript = False
all_model_classes = (
(TFLongformerModel, TFLongformerForMaskedLM, TFLongformerForQuestionAnswering,) if is_tf_available() else ()
(
TFLongformerModel,
TFLongformerForMaskedLM,
TFLongformerForQuestionAnswering,
)
if is_tf_available()
else ()
)
def setUp(self):
......
tests/test_modeling_tf_openai.py
View file @ a75c64d8
......@@ -36,7 +36,8 @@ if is_tf_available():
class TFOpenAIGPTModelTester:
def __init__(
self, parent,
self,
parent,
):
self.parent = parent
self.batch_size = 13
......
tests/test_modeling_tf_roberta.py
View file @ a75c64d8
......@@ -40,7 +40,8 @@ if is_tf_available():
class TFRobertaModelTester:
def __init__(
self, parent,
self,
parent,
):
self.parent = parent
self.batch_size = 13
......
tests/test_modeling_tf_t5.py
View file @ a75c64d8
......@@ -32,7 +32,8 @@ if is_tf_available():
class TFT5ModelTester:
def __init__(
self, parent,
self,
parent,
):
self.parent = parent
self.batch_size = 13
......@@ -181,7 +182,10 @@ class TFT5ModelTester:
# append to next input_ids and attn_mask
next_input_ids = tf.concat([input_ids, next_tokens], axis=-1)
attn_mask = tf.concat([attn_mask, tf.ones((attn_mask.shape[0], 1), dtype=tf.int32)], axis=1,)
attn_mask = tf.concat(
[attn_mask, tf.ones((attn_mask.shape[0], 1), dtype=tf.int32)],
axis=1,
)
# get two different outputs
output_from_no_past = model(next_input_ids, attention_mask=attn_mask)[0]
......
tests/test_modeling_tf_transfo_xl.py
View file @ a75c64d8
......@@ -32,7 +32,8 @@ if is_tf_available():
class TFTransfoXLModelTester:
def __init__(
self, parent,
self,
parent,
):
self.parent = parent
self.batch_size = 13
......
tests/test_modeling_tf_xlm.py
View file @ a75c64d8
......@@ -40,7 +40,8 @@ if is_tf_available():
class TFXLMModelTester:
def __init__(
self, parent,
self,
parent,
):
self.parent = parent
self.batch_size = 13
......
tests/test_modeling_tf_xlnet.py
View file @ a75c64d8
......@@ -40,7 +40,8 @@ if is_tf_available():
class TFXLNetModelTester:
def __init__(
self, parent,
self,
parent,
):
self.parent = parent
self.batch_size = 13
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment