Sentence-pair tasks handling. Using common tests on RoBERTa. Forced push to fix indentation.
This commit is contained in:
LysandreJik
@@ -12,58 +12,172 @@
|
||||
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||
# See the License for the specific language governing permissions and
|
||||
# limitations under the License.
|
||||
from __future__ import (absolute_import, division, print_function,
|
||||
unicode_literals)
|
||||
from __future__ import absolute_import
|
||||
from __future__ import division
|
||||
from __future__ import print_function
|
||||
|
||||
import os
|
||||
import unittest
|
||||
import shutil
|
||||
import pytest
|
||||
import torch
|
||||
|
||||
from pytorch_transformers.modeling_roberta import (RobertaForMaskedLM,
|
||||
RobertaModel)
|
||||
from pytorch_transformers import (RobertaConfig, RobertaModel, RobertaForMaskedLM)
|
||||
from pytorch_transformers.modeling_roberta import ROBERTA_PRETRAINED_MODEL_ARCHIVE_MAP
|
||||
|
||||
from .modeling_common_test import (CommonTestCases, ConfigTester, ids_tensor)
|
||||
|
||||
|
||||
class RobertaModelTest(unittest.TestCase):
|
||||
class RobertaModelTest(CommonTestCases.CommonModelTester):
|
||||
|
||||
# @pytest.mark.slow
|
||||
def test_inference_masked_lm(self):
|
||||
model = RobertaForMaskedLM.from_pretrained('roberta-base')
|
||||
|
||||
input_ids = torch.tensor([[ 0, 31414, 232, 328, 740, 1140, 12695, 69, 46078, 1588, 2]])
|
||||
output = model(input_ids)[0]
|
||||
expected_shape = torch.Size((1, 11, 50265))
|
||||
self.assertEqual(
|
||||
output.shape,
|
||||
expected_shape
|
||||
)
|
||||
# compare the actual values for a slice.
|
||||
expected_slice = torch.Tensor(
|
||||
[[[33.8843, -4.3107, 22.7779],
|
||||
[ 4.6533, -2.8099, 13.6252],
|
||||
[ 1.8222, -3.6898, 8.8600]]]
|
||||
)
|
||||
self.assertTrue(
|
||||
torch.allclose(output[:, :3, :3], expected_slice, atol=1e-3)
|
||||
)
|
||||
all_model_classes = (RobertaForMaskedLM, RobertaModel)
|
||||
|
||||
# @pytest.mark.slow
|
||||
def test_inference_no_head(self):
|
||||
model = RobertaModel.from_pretrained('roberta-base')
|
||||
|
||||
input_ids = torch.tensor([[ 0, 31414, 232, 328, 740, 1140, 12695, 69, 46078, 1588, 2]])
|
||||
output = model(input_ids)[0]
|
||||
# compare the actual values for a slice.
|
||||
expected_slice = torch.Tensor(
|
||||
[[[-0.0231, 0.0782, 0.0074],
|
||||
[-0.1854, 0.0539, -0.0174],
|
||||
[ 0.0548, 0.0799, 0.1687]]]
|
||||
)
|
||||
self.assertTrue(
|
||||
torch.allclose(output[:, :3, :3], expected_slice, atol=1e-3)
|
||||
)
|
||||
class RobertaModelTester(object):
|
||||
|
||||
def __init__(self,
|
||||
parent,
|
||||
batch_size=13,
|
||||
seq_length=7,
|
||||
is_training=True,
|
||||
use_input_mask=True,
|
||||
use_token_type_ids=True,
|
||||
use_labels=True,
|
||||
vocab_size=99,
|
||||
hidden_size=32,
|
||||
num_hidden_layers=5,
|
||||
num_attention_heads=4,
|
||||
intermediate_size=37,
|
||||
hidden_act="gelu",
|
||||
hidden_dropout_prob=0.1,
|
||||
attention_probs_dropout_prob=0.1,
|
||||
max_position_embeddings=512,
|
||||
type_vocab_size=16,
|
||||
type_sequence_label_size=2,
|
||||
initializer_range=0.02,
|
||||
num_labels=3,
|
||||
num_choices=4,
|
||||
scope=None,
|
||||
):
|
||||
self.parent = parent
|
||||
self.batch_size = batch_size
|
||||
self.seq_length = seq_length
|
||||
self.is_training = is_training
|
||||
self.use_input_mask = use_input_mask
|
||||
self.use_token_type_ids = use_token_type_ids
|
||||
self.use_labels = use_labels
|
||||
self.vocab_size = vocab_size
|
||||
self.hidden_size = hidden_size
|
||||
self.num_hidden_layers = num_hidden_layers
|
||||
self.num_attention_heads = num_attention_heads
|
||||
self.intermediate_size = intermediate_size
|
||||
self.hidden_act = hidden_act
|
||||
self.hidden_dropout_prob = hidden_dropout_prob
|
||||
self.attention_probs_dropout_prob = attention_probs_dropout_prob
|
||||
self.max_position_embeddings = max_position_embeddings
|
||||
self.type_vocab_size = type_vocab_size
|
||||
self.type_sequence_label_size = type_sequence_label_size
|
||||
self.initializer_range = initializer_range
|
||||
self.num_labels = num_labels
|
||||
self.num_choices = num_choices
|
||||
self.scope = scope
|
||||
|
||||
def prepare_config_and_inputs(self):
|
||||
input_ids = ids_tensor([self.batch_size, self.seq_length], self.vocab_size)
|
||||
|
||||
if __name__ == '__main__':
|
||||
input_mask = None
|
||||
if self.use_input_mask:
|
||||
input_mask = ids_tensor([self.batch_size, self.seq_length], vocab_size=2)
|
||||
|
||||
token_type_ids = None
|
||||
if self.use_token_type_ids:
|
||||
token_type_ids = ids_tensor([self.batch_size, self.seq_length], self.type_vocab_size)
|
||||
|
||||
sequence_labels = None
|
||||
token_labels = None
|
||||
choice_labels = None
|
||||
if self.use_labels:
|
||||
sequence_labels = ids_tensor([self.batch_size], self.type_sequence_label_size)
|
||||
token_labels = ids_tensor([self.batch_size, self.seq_length], self.num_labels)
|
||||
choice_labels = ids_tensor([self.batch_size], self.num_choices)
|
||||
|
||||
config = RobertaConfig(
|
||||
vocab_size_or_config_json_file=self.vocab_size,
|
||||
hidden_size=self.hidden_size,
|
||||
num_hidden_layers=self.num_hidden_layers,
|
||||
num_attention_heads=self.num_attention_heads,
|
||||
intermediate_size=self.intermediate_size,
|
||||
hidden_act=self.hidden_act,
|
||||
hidden_dropout_prob=self.hidden_dropout_prob,
|
||||
attention_probs_dropout_prob=self.attention_probs_dropout_prob,
|
||||
max_position_embeddings=self.max_position_embeddings,
|
||||
type_vocab_size=self.type_vocab_size,
|
||||
initializer_range=self.initializer_range)
|
||||
|
||||
return config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
|
||||
|
||||
def check_loss_output(self, result):
|
||||
self.parent.assertListEqual(
|
||||
list(result["loss"].size()),
|
||||
[])
|
||||
|
||||
def create_and_check_roberta_model(self, config, input_ids, token_type_ids, input_mask, sequence_labels,
|
||||
token_labels, choice_labels):
|
||||
model = RobertaModel(config=config)
|
||||
model.eval()
|
||||
sequence_output, pooled_output = model(input_ids, token_type_ids, input_mask)
|
||||
sequence_output, pooled_output = model(input_ids, token_type_ids)
|
||||
sequence_output, pooled_output = model(input_ids)
|
||||
|
||||
result = {
|
||||
"sequence_output": sequence_output,
|
||||
"pooled_output": pooled_output,
|
||||
}
|
||||
self.parent.assertListEqual(
|
||||
list(result["sequence_output"].size()),
|
||||
[self.batch_size, self.seq_length, self.hidden_size])
|
||||
self.parent.assertListEqual(list(result["pooled_output"].size()), [self.batch_size, self.hidden_size])
|
||||
|
||||
def create_and_check_roberta_for_masked_lm(self, config, input_ids, token_type_ids, input_mask, sequence_labels,
|
||||
token_labels, choice_labels):
|
||||
model = RobertaForMaskedLM(config=config)
|
||||
model.eval()
|
||||
loss, prediction_scores = model(input_ids, token_type_ids, input_mask, token_labels)
|
||||
result = {
|
||||
"loss": loss,
|
||||
"prediction_scores": prediction_scores,
|
||||
}
|
||||
self.parent.assertListEqual(
|
||||
list(result["prediction_scores"].size()),
|
||||
[self.batch_size, self.seq_length, self.vocab_size])
|
||||
self.check_loss_output(result)
|
||||
|
||||
def prepare_config_and_inputs_for_common(self):
|
||||
config_and_inputs = self.prepare_config_and_inputs()
|
||||
(config, input_ids, token_type_ids, input_mask,
|
||||
sequence_labels, token_labels, choice_labels) = config_and_inputs
|
||||
inputs_dict = {'input_ids': input_ids, 'token_type_ids': token_type_ids, 'attention_mask': input_mask}
|
||||
return config, inputs_dict
|
||||
|
||||
def setUp(self):
|
||||
self.model_tester = RobertaModelTest.RobertaModelTester(self)
|
||||
self.config_tester = ConfigTester(self, config_class=RobertaConfig, hidden_size=37)
|
||||
|
||||
def test_config(self):
|
||||
self.config_tester.run_common_tests()
|
||||
|
||||
def test_roberta_model(self):
|
||||
config_and_inputs = self.model_tester.prepare_config_and_inputs()
|
||||
self.model_tester.create_and_check_roberta_model(*config_and_inputs)
|
||||
|
||||
def test_for_masked_lm(self):
|
||||
config_and_inputs = self.model_tester.prepare_config_and_inputs()
|
||||
self.model_tester.create_and_check_roberta_for_masked_lm(*config_and_inputs)
|
||||
|
||||
@pytest.mark.slow
|
||||
def test_model_from_pretrained(self):
|
||||
cache_dir = "/tmp/pytorch_transformers_test/"
|
||||
for model_name in list(ROBERTA_PRETRAINED_MODEL_ARCHIVE_MAP.keys())[:1]:
|
||||
model = RobertaModel.from_pretrained(model_name, cache_dir=cache_dir)
|
||||
shutil.rmtree(cache_dir)
|
||||
self.assertIsNotNone(model)
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
|
||||
@@ -12,32 +12,45 @@
|
||||
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||
# See the License for the specific language governing permissions and
|
||||
# limitations under the License.
|
||||
from __future__ import (absolute_import, division, print_function,
|
||||
unicode_literals)
|
||||
from __future__ import absolute_import, division, print_function, unicode_literals
|
||||
|
||||
import os
|
||||
import unittest
|
||||
import pytest
|
||||
import six
|
||||
|
||||
from pytorch_transformers.tokenization_roberta import RobertaTokenizer
|
||||
from pytorch_transformers.tokenization_roberta import RobertaTokenizer, VOCAB_FILES_NAMES
|
||||
from .tokenization_tests_commons import create_and_check_tokenizer_commons, TemporaryDirectory
|
||||
|
||||
|
||||
class RobertaTokenizationTest(unittest.TestCase):
|
||||
|
||||
# @pytest.mark.slow
|
||||
def test_full_tokenizer(self):
|
||||
tokenizer = RobertaTokenizer.from_pretrained('roberta-base')
|
||||
self.assertListEqual(
|
||||
tokenizer.encode('Hello world!'),
|
||||
[0, 31414, 232, 328, 2]
|
||||
)
|
||||
if six.PY3:
|
||||
self.assertListEqual(
|
||||
tokenizer.encode('Hello world! cécé herlolip'),
|
||||
[0, 31414, 232, 328, 740, 1140, 12695, 69, 46078, 1588, 2]
|
||||
)
|
||||
""" Adapted from Sennrich et al. 2015 and https://github.com/rsennrich/subword-nmt """
|
||||
vocab = ["l", "o", "w", "e", "r", "s", "t", "i", "d", "n",
|
||||
"lo", "low", "er",
|
||||
"low", "lowest", "newer", "wider", "<unk>"]
|
||||
vocab_tokens = dict(zip(vocab, range(len(vocab))))
|
||||
special_tokens_map = {"unk_token": "<unk>"}
|
||||
|
||||
with TemporaryDirectory() as tmpdirname:
|
||||
vocab_file = os.path.join(tmpdirname, VOCAB_FILES_NAMES['vocab_file'])
|
||||
with open(vocab_file, "w") as fp:
|
||||
[fp.write(f"{vocab} {index}\n") for index, vocab in enumerate(vocab_tokens)]
|
||||
|
||||
input_text = u"lower newer"
|
||||
output_text = u"lower<unk>newer"
|
||||
|
||||
create_and_check_tokenizer_commons(self, input_text, output_text, RobertaTokenizer, tmpdirname, **special_tokens_map)
|
||||
|
||||
tokenizer = RobertaTokenizer(vocab_file, **special_tokens_map)
|
||||
text = "lower"
|
||||
bpe_tokens = ["low", "er"]
|
||||
tokens = tokenizer.tokenize(text)
|
||||
self.assertListEqual(tokens, bpe_tokens)
|
||||
|
||||
input_tokens = tokens + [tokenizer.unk_token]
|
||||
input_bpe_tokens = [13, 12, 17]
|
||||
self.assertListEqual(
|
||||
tokenizer.convert_tokens_to_ids(input_tokens), input_bpe_tokens)
|
||||
|
||||
|
||||
if __name__ == '__main__':
|
||||
|
||||
Reference in New Issue
Block a user