SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

[WIP] Test TF Flaubert + Add {XLM, Flaubert}{TokenClassification, MultipleC… (#5614)

Browse Source 
* Test TF Flaubert + Add {XLM, Flaubert}{TokenClassification, MultipleChoice} models and tests

* AutoModels


Tiny tweaks

* Style

* Final changes before merge

* Re-order for simpler review

* Final fixes

* Addressing @sgugger's comments

* Test MultipleChoice
This commit is contained in:
Lysandre Debut
2020-07-29 14:26:26 -04:00
committed by GitHub
parent 8a8ae27617
commit 3f94170a10
12 changed files with 652 additions and 21 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
tests/test_modeling_common.py
View File

@@ -66,7 +66,7 @@ class ModelTesterMixin:
if model_class in MODEL_FOR_MULTIPLE_CHOICE_MAPPING.values():
return {
k: v.unsqueeze(1).expand(-1, self.model_tester.num_choices, -1).contiguous()
if isinstance(v, torch.Tensor) and v.ndim != 0
if isinstance(v, torch.Tensor) and v.ndim > 1
else v
for k, v in inputs_dict.items()
}

47
tests/test_modeling_flaubert.py
View File

@@ -32,6 +32,7 @@ if is_torch_available():
FlaubertForQuestionAnsweringSimple,
FlaubertForSequenceClassification,
FlaubertForTokenClassification,
FlaubertForMultipleChoice,
)
from transformers.modeling_flaubert import FLAUBERT_PRETRAINED_MODEL_ARCHIVE_LIST
@@ -90,6 +91,7 @@ class FlaubertModelTester(object):
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)
is_impossible_labels = ids_tensor([self.batch_size], 2).float()
choice_labels = ids_tensor([self.batch_size], self.num_choices)
config = FlaubertConfig(
vocab_size=self.vocab_size,
@@ -118,6 +120,7 @@ class FlaubertModelTester(object):
sequence_labels,
token_labels,
is_impossible_labels,
choice_labels,
input_mask,
)
@@ -133,6 +136,7 @@ class FlaubertModelTester(object):
sequence_labels,
token_labels,
is_impossible_labels,
choice_labels,
input_mask,
):
model = FlaubertModel(config=config)
@@ -158,6 +162,7 @@ class FlaubertModelTester(object):
sequence_labels,
token_labels,
is_impossible_labels,
choice_labels,
input_mask,
):
model = FlaubertWithLMHeadModel(config)
@@ -183,6 +188,7 @@ class FlaubertModelTester(object):
sequence_labels,
token_labels,
is_impossible_labels,
choice_labels,
input_mask,
):
model = FlaubertForQuestionAnsweringSimple(config)
@@ -212,6 +218,7 @@ class FlaubertModelTester(object):
sequence_labels,
token_labels,
is_impossible_labels,
choice_labels,
input_mask,
):
model = FlaubertForQuestionAnswering(config)
@@ -278,6 +285,7 @@ class FlaubertModelTester(object):
sequence_labels,
token_labels,
is_impossible_labels,
choice_labels,
input_mask,
):
model = FlaubertForSequenceClassification(config)
@@ -304,6 +312,7 @@ class FlaubertModelTester(object):
sequence_labels,
token_labels,
is_impossible_labels,
choice_labels,
input_mask,
):
config.num_labels = self.num_labels
@@ -319,6 +328,38 @@ class FlaubertModelTester(object):
self.parent.assertListEqual(list(result["logits"].size()), [self.batch_size, self.seq_length, self.num_labels])
self.check_loss_output(result)
def create_and_check_flaubert_multiple_choice(
self,
config,
input_ids,
token_type_ids,
input_lengths,
sequence_labels,
token_labels,
is_impossible_labels,
choice_labels,
input_mask,
):
config.num_choices = self.num_choices
model = FlaubertForMultipleChoice(config=config)
model.to(torch_device)
model.eval()
multiple_choice_inputs_ids = input_ids.unsqueeze(1).expand(-1, self.num_choices, -1).contiguous()
multiple_choice_token_type_ids = token_type_ids.unsqueeze(1).expand(-1, self.num_choices, -1).contiguous()
multiple_choice_input_mask = input_mask.unsqueeze(1).expand(-1, self.num_choices, -1).contiguous()
loss, logits = model(
multiple_choice_inputs_ids,
attention_mask=multiple_choice_input_mask,
token_type_ids=multiple_choice_token_type_ids,
labels=choice_labels,
)
result = {
"loss": loss,
"logits": logits,
}
self.parent.assertListEqual(list(result["logits"].size()), [self.batch_size, self.num_choices])
self.check_loss_output(result)
def prepare_config_and_inputs_for_common(self):
config_and_inputs = self.prepare_config_and_inputs()
(
@@ -329,6 +370,7 @@ class FlaubertModelTester(object):
sequence_labels,
token_labels,
is_impossible_labels,
choice_labels,
input_mask,
) = config_and_inputs
inputs_dict = {"input_ids": input_ids, "token_type_ids": token_type_ids, "lengths": input_lengths}
@@ -346,6 +388,7 @@ class FlaubertModelTest(ModelTesterMixin, unittest.TestCase):
FlaubertForQuestionAnsweringSimple,
FlaubertForSequenceClassification,
FlaubertForTokenClassification,
FlaubertForMultipleChoice,
)
if is_torch_available()
else ()
@@ -382,6 +425,10 @@ class FlaubertModelTest(ModelTesterMixin, unittest.TestCase):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_flaubert_token_classif(*config_and_inputs)
def test_flaubert_multiple_choice(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_flaubert_multiple_choice(*config_and_inputs)
@slow
def test_model_from_pretrained(self):
for model_name in FLAUBERT_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:

4
tests/test_modeling_tf_common.py
View File

@@ -80,8 +80,8 @@ class TFModelTesterMixin:
def _prepare_for_class(self, inputs_dict, model_class, return_labels=False):
if model_class in TF_MODEL_FOR_MULTIPLE_CHOICE_MAPPING.values():
inputs_dict = {
k: tf.tile(tf.expand_dims(v, 1), (1, self.model_tester.num_choices, 1))
if isinstance(v, tf.Tensor) and v.ndim != 0
k: tf.tile(tf.expand_dims(v, 1), (1, self.model_tester.num_choices) + (1,) * (v.ndim - 1))
if isinstance(v, tf.Tensor) and v.ndim > 0
else v
for k, v in inputs_dict.items()
}

331
tests/test_modeling_tf_flaubert.py
View File

@@ -18,11 +18,340 @@ import unittest
from transformers import is_tf_available
from transformers.testing_utils import require_tf, slow
from .test_configuration_common import ConfigTester
from .test_modeling_tf_common import TFModelTesterMixin, ids_tensor
if is_tf_available():
import tensorflow as tf
import numpy as np
from transformers import TFFlaubertModel
from transformers import (
FlaubertConfig,
TFFlaubertModel,
TFFlaubertWithLMHeadModel,
TFFlaubertForSequenceClassification,
TFFlaubertForQuestionAnsweringSimple,
TFFlaubertForTokenClassification,
TFFlaubertForMultipleChoice,
TF_FLAUBERT_PRETRAINED_MODEL_ARCHIVE_LIST,
)
class TFFlaubertModelTester:
def __init__(
self, parent,
):
self.parent = parent
self.batch_size = 13
self.seq_length = 7
self.is_training = True
self.use_input_lengths = True
self.use_token_type_ids = True
self.use_labels = True
self.gelu_activation = True
self.sinusoidal_embeddings = False
self.causal = False
self.asm = False
self.n_langs = 2
self.vocab_size = 99
self.n_special = 0
self.hidden_size = 32
self.num_hidden_layers = 5
self.num_attention_heads = 4
self.hidden_dropout_prob = 0.1
self.attention_probs_dropout_prob = 0.1
self.max_position_embeddings = 512
self.type_vocab_size = 16
self.type_sequence_label_size = 2
self.initializer_range = 0.02
self.num_labels = 3
self.num_choices = 4
self.summary_type = "last"
self.use_proj = True
self.scope = None
self.bos_token_id = 0
def prepare_config_and_inputs(self):
input_ids = ids_tensor([self.batch_size, self.seq_length], self.vocab_size)
input_mask = ids_tensor([self.batch_size, self.seq_length], 2, dtype=tf.float32)
input_lengths = None
if self.use_input_lengths:
input_lengths = (
ids_tensor([self.batch_size], vocab_size=2) + self.seq_length - 2
) # small variation of seq_length
token_type_ids = None
if self.use_token_type_ids:
token_type_ids = ids_tensor([self.batch_size, self.seq_length], self.n_langs)
sequence_labels = None
token_labels = None
is_impossible_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)
is_impossible_labels = ids_tensor([self.batch_size], 2, dtype=tf.float32)
choice_labels = ids_tensor([self.batch_size], self.num_choices)
config = FlaubertConfig(
vocab_size=self.vocab_size,
n_special=self.n_special,
emb_dim=self.hidden_size,
n_layers=self.num_hidden_layers,
n_heads=self.num_attention_heads,
dropout=self.hidden_dropout_prob,
attention_dropout=self.attention_probs_dropout_prob,
gelu_activation=self.gelu_activation,
sinusoidal_embeddings=self.sinusoidal_embeddings,
asm=self.asm,
causal=self.causal,
n_langs=self.n_langs,
max_position_embeddings=self.max_position_embeddings,
initializer_range=self.initializer_range,
summary_type=self.summary_type,
use_proj=self.use_proj,
bos_token_id=self.bos_token_id,
)
return (
config,
input_ids,
token_type_ids,
input_lengths,
sequence_labels,
token_labels,
is_impossible_labels,
choice_labels,
input_mask,
)
def create_and_check_flaubert_model(
self,
config,
input_ids,
token_type_ids,
input_lengths,
sequence_labels,
token_labels,
is_impossible_labels,
choice_labels,
input_mask,
):
model = TFFlaubertModel(config=config)
inputs = {"input_ids": input_ids, "lengths": input_lengths, "langs": token_type_ids}
outputs = model(inputs)
inputs = [input_ids, input_mask]
outputs = model(inputs)
sequence_output = outputs[0]
result = {
"sequence_output": sequence_output.numpy(),
}
self.parent.assertListEqual(
list(result["sequence_output"].shape), [self.batch_size, self.seq_length, self.hidden_size]
)
def create_and_check_flaubert_lm_head(
self,
config,
input_ids,
token_type_ids,
input_lengths,
sequence_labels,
token_labels,
is_impossible_labels,
choice_labels,
input_mask,
):
model = TFFlaubertWithLMHeadModel(config)
inputs = {"input_ids": input_ids, "lengths": input_lengths, "langs": token_type_ids}
outputs = model(inputs)
logits = outputs[0]
result = {
"logits": logits.numpy(),
}
self.parent.assertListEqual(list(result["logits"].shape), [self.batch_size, self.seq_length, self.vocab_size])
def create_and_check_flaubert_qa(
self,
config,
input_ids,
token_type_ids,
input_lengths,
sequence_labels,
token_labels,
is_impossible_labels,
choice_labels,
input_mask,
):
model = TFFlaubertForQuestionAnsweringSimple(config)
inputs = {"input_ids": input_ids, "lengths": input_lengths}
start_logits, end_logits = model(inputs)
result = {
"start_logits": start_logits.numpy(),
"end_logits": end_logits.numpy(),
}
self.parent.assertListEqual(list(result["start_logits"].shape), [self.batch_size, self.seq_length])
self.parent.assertListEqual(list(result["end_logits"].shape), [self.batch_size, self.seq_length])
def create_and_check_flaubert_sequence_classif(
self,
config,
input_ids,
token_type_ids,
input_lengths,
sequence_labels,
token_labels,
is_impossible_labels,
choice_labels,
input_mask,
):
model = TFFlaubertForSequenceClassification(config)
inputs = {"input_ids": input_ids, "lengths": input_lengths}
(logits,) = model(inputs)
result = {
"logits": logits.numpy(),
}
self.parent.assertListEqual(list(result["logits"].shape), [self.batch_size, self.type_sequence_label_size])
def create_and_check_flaubert_for_token_classification(
self,
config,
input_ids,
token_type_ids,
input_lengths,
sequence_labels,
token_labels,
is_impossible_labels,
choice_labels,
input_mask,
):
config.num_labels = self.num_labels
model = TFFlaubertForTokenClassification(config=config)
inputs = {"input_ids": input_ids, "attention_mask": input_mask, "token_type_ids": token_type_ids}
(logits,) = model(inputs)
result = {
"logits": logits.numpy(),
}
self.parent.assertListEqual(list(result["logits"].shape), [self.batch_size, self.seq_length, self.num_labels])
def create_and_check_flaubert_for_multiple_choice(
self,
config,
input_ids,
token_type_ids,
input_lengths,
sequence_labels,
token_labels,
is_impossible_labels,
choice_labels,
input_mask,
):
config.num_choices = self.num_choices
model = TFFlaubertForMultipleChoice(config=config)
multiple_choice_inputs_ids = tf.tile(tf.expand_dims(input_ids, 1), (1, self.num_choices, 1))
multiple_choice_input_mask = tf.tile(tf.expand_dims(input_mask, 1), (1, self.num_choices, 1))
multiple_choice_token_type_ids = tf.tile(tf.expand_dims(token_type_ids, 1), (1, self.num_choices, 1))
inputs = {
"input_ids": multiple_choice_inputs_ids,
"attention_mask": multiple_choice_input_mask,
"token_type_ids": multiple_choice_token_type_ids,
}
(logits,) = model(inputs)
result = {"logits": logits.numpy()}
self.parent.assertListEqual(list(result["logits"].shape), [self.batch_size, self.num_choices])
def prepare_config_and_inputs_for_common(self):
config_and_inputs = self.prepare_config_and_inputs()
(
config,
input_ids,
token_type_ids,
input_lengths,
sequence_labels,
token_labels,
is_impossible_labels,
choice_labels,
input_mask,
) = config_and_inputs
inputs_dict = {
"input_ids": input_ids,
"token_type_ids": token_type_ids,
"langs": token_type_ids,
"lengths": input_lengths,
}
return config, inputs_dict
@require_tf
class TFFlaubertModelTest(TFModelTesterMixin, unittest.TestCase):
all_model_classes = (
(
TFFlaubertModel,
TFFlaubertWithLMHeadModel,
TFFlaubertForSequenceClassification,
TFFlaubertForQuestionAnsweringSimple,
TFFlaubertForTokenClassification,
TFFlaubertForMultipleChoice,
)
if is_tf_available()
else ()
)
all_generative_model_classes = (
(TFFlaubertWithLMHeadModel,) if is_tf_available() else ()
) # TODO (PVP): Check other models whether language generation is also applicable
def setUp(self):
self.model_tester = TFFlaubertModelTester(self)
self.config_tester = ConfigTester(self, config_class=FlaubertConfig, emb_dim=37)
def test_config(self):
self.config_tester.run_common_tests()
def test_flaubert_model(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_flaubert_model(*config_and_inputs)
def test_flaubert_lm_head(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_flaubert_lm_head(*config_and_inputs)
def test_flaubert_qa(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_flaubert_qa(*config_and_inputs)
def test_flaubert_sequence_classif(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_flaubert_sequence_classif(*config_and_inputs)
def test_for_token_classification(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_flaubert_for_token_classification(*config_and_inputs)
def test_for_multiple_choice(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_flaubert_for_multiple_choice(*config_and_inputs)
@slow
def test_model_from_pretrained(self):
for model_name in TF_FLAUBERT_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
model = TFFlaubertModel.from_pretrained(model_name)
self.assertIsNotNone(model)
@require_tf

41
tests/test_modeling_tf_xlm.py
View File

@@ -32,6 +32,7 @@ if is_tf_available():
TFXLMForSequenceClassification,
TFXLMForQuestionAnsweringSimple,
TFXLMForTokenClassification,
TFXLMForMultipleChoice,
TF_XLM_PRETRAINED_MODEL_ARCHIVE_LIST,
)
@@ -91,6 +92,7 @@ class TFXLMModelTester:
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)
is_impossible_labels = ids_tensor([self.batch_size], 2, dtype=tf.float32)
choice_labels = ids_tensor([self.batch_size], self.num_choices)
config = XLMConfig(
vocab_size=self.vocab_size,
@@ -120,6 +122,7 @@ class TFXLMModelTester:
sequence_labels,
token_labels,
is_impossible_labels,
choice_labels,
input_mask,
)
@@ -132,6 +135,7 @@ class TFXLMModelTester:
sequence_labels,
token_labels,
is_impossible_labels,
choice_labels,
input_mask,
):
model = TFXLMModel(config=config)
@@ -157,6 +161,7 @@ class TFXLMModelTester:
sequence_labels,
token_labels,
is_impossible_labels,
choice_labels,
input_mask,
):
model = TFXLMWithLMHeadModel(config)
@@ -181,6 +186,7 @@ class TFXLMModelTester:
sequence_labels,
token_labels,
is_impossible_labels,
choice_labels,
input_mask,
):
model = TFXLMForQuestionAnsweringSimple(config)
@@ -206,6 +212,7 @@ class TFXLMModelTester:
sequence_labels,
token_labels,
is_impossible_labels,
choice_labels,
input_mask,
):
model = TFXLMForSequenceClassification(config)
@@ -229,6 +236,7 @@ class TFXLMModelTester:
sequence_labels,
token_labels,
is_impossible_labels,
choice_labels,
input_mask,
):
config.num_labels = self.num_labels
@@ -240,6 +248,32 @@ class TFXLMModelTester:
}
self.parent.assertListEqual(list(result["logits"].shape), [self.batch_size, self.seq_length, self.num_labels])
def create_and_check_xlm_for_multiple_choice(
self,
config,
input_ids,
token_type_ids,
input_lengths,
sequence_labels,
token_labels,
is_impossible_labels,
choice_labels,
input_mask,
):
config.num_choices = self.num_choices
model = TFXLMForMultipleChoice(config=config)
multiple_choice_inputs_ids = tf.tile(tf.expand_dims(input_ids, 1), (1, self.num_choices, 1))
multiple_choice_input_mask = tf.tile(tf.expand_dims(input_mask, 1), (1, self.num_choices, 1))
multiple_choice_token_type_ids = tf.tile(tf.expand_dims(token_type_ids, 1), (1, self.num_choices, 1))
inputs = {
"input_ids": multiple_choice_inputs_ids,
"attention_mask": multiple_choice_input_mask,
"token_type_ids": multiple_choice_token_type_ids,
}
(logits,) = model(inputs)
result = {"logits": logits.numpy()}
self.parent.assertListEqual(list(result["logits"].shape), [self.batch_size, self.num_choices])
def prepare_config_and_inputs_for_common(self):
config_and_inputs = self.prepare_config_and_inputs()
(
@@ -250,6 +284,7 @@ class TFXLMModelTester:
sequence_labels,
token_labels,
is_impossible_labels,
choice_labels,
input_mask,
) = config_and_inputs
inputs_dict = {
@@ -265,13 +300,13 @@ class TFXLMModelTester:
class TFXLMModelTest(TFModelTesterMixin, unittest.TestCase):
all_model_classes = (
# TODO The multiple choice model is missing and should be added.
(
TFXLMModel,
TFXLMWithLMHeadModel,
TFXLMForSequenceClassification,
TFXLMForQuestionAnsweringSimple,
TFXLMForTokenClassification,
TFXLMForMultipleChoice,
)
if is_tf_available()
else ()
@@ -307,6 +342,10 @@ class TFXLMModelTest(TFModelTesterMixin, unittest.TestCase):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_xlm_for_token_classification(*config_and_inputs)
def test_for_multiple_choice(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_xlm_for_multiple_choice(*config_and_inputs)
@slow
def test_model_from_pretrained(self):
for model_name in TF_XLM_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:

50
tests/test_modeling_xlm.py
View File

@@ -33,6 +33,7 @@ if is_torch_available():
XLMForQuestionAnswering,
XLMForSequenceClassification,
XLMForQuestionAnsweringSimple,
XLMForMultipleChoice,
)
from transformers.modeling_xlm import XLM_PRETRAINED_MODEL_ARCHIVE_LIST
@@ -63,7 +64,7 @@ class XLMModelTester:
self.max_position_embeddings = 512
self.type_sequence_label_size = 2
self.initializer_range = 0.02
self.num_labels = 3
self.num_labels = 2
self.num_choices = 4
self.summary_type = "last"
self.use_proj = True
@@ -91,6 +92,7 @@ class XLMModelTester:
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)
is_impossible_labels = ids_tensor([self.batch_size], 2).float()
choice_labels = ids_tensor([self.batch_size], self.num_choices)
config = XLMConfig(
vocab_size=self.vocab_size,
@@ -109,6 +111,7 @@ class XLMModelTester:
initializer_range=self.initializer_range,
summary_type=self.summary_type,
use_proj=self.use_proj,
num_labels=self.num_labels,
bos_token_id=self.bos_token_id,
)
@@ -120,6 +123,7 @@ class XLMModelTester:
sequence_labels,
token_labels,
is_impossible_labels,
choice_labels,
input_mask,
)
@@ -135,6 +139,7 @@ class XLMModelTester:
sequence_labels,
token_labels,
is_impossible_labels,
choice_labels,
input_mask,
):
model = XLMModel(config=config)
@@ -160,6 +165,7 @@ class XLMModelTester:
sequence_labels,
token_labels,
is_impossible_labels,
choice_labels,
input_mask,
):
model = XLMWithLMHeadModel(config)
@@ -185,6 +191,7 @@ class XLMModelTester:
sequence_labels,
token_labels,
is_impossible_labels,
choice_labels,
input_mask,
):
model = XLMForQuestionAnsweringSimple(config)
@@ -214,6 +221,7 @@ class XLMModelTester:
sequence_labels,
token_labels,
is_impossible_labels,
choice_labels,
input_mask,
):
model = XLMForQuestionAnswering(config)
@@ -280,6 +288,7 @@ class XLMModelTester:
sequence_labels,
token_labels,
is_impossible_labels,
choice_labels,
input_mask,
):
model = XLMForSequenceClassification(config)
@@ -306,6 +315,7 @@ class XLMModelTester:
sequence_labels,
token_labels,
is_impossible_labels,
choice_labels,
input_mask,
):
config.num_labels = self.num_labels
@@ -321,6 +331,38 @@ class XLMModelTester:
self.parent.assertListEqual(list(result["logits"].size()), [self.batch_size, self.seq_length, self.num_labels])
self.check_loss_output(result)
def create_and_check_xlm_for_multiple_choice(
self,
config,
input_ids,
token_type_ids,
input_lengths,
sequence_labels,
token_labels,
is_impossible_labels,
choice_labels,
input_mask,
):
config.num_choices = self.num_choices
model = XLMForMultipleChoice(config=config)
model.to(torch_device)
model.eval()
multiple_choice_inputs_ids = input_ids.unsqueeze(1).expand(-1, self.num_choices, -1).contiguous()
multiple_choice_token_type_ids = token_type_ids.unsqueeze(1).expand(-1, self.num_choices, -1).contiguous()
multiple_choice_input_mask = input_mask.unsqueeze(1).expand(-1, self.num_choices, -1).contiguous()
loss, logits = model(
multiple_choice_inputs_ids,
attention_mask=multiple_choice_input_mask,
token_type_ids=multiple_choice_token_type_ids,
labels=choice_labels,
)
result = {
"loss": loss,
"logits": logits,
}
self.parent.assertListEqual(list(result["logits"].size()), [self.batch_size, self.num_choices])
self.check_loss_output(result)
def prepare_config_and_inputs_for_common(self):
config_and_inputs = self.prepare_config_and_inputs()
(
@@ -331,6 +373,7 @@ class XLMModelTester:
sequence_labels,
token_labels,
is_impossible_labels,
choice_labels,
input_mask,
) = config_and_inputs
inputs_dict = {"input_ids": input_ids, "token_type_ids": token_type_ids, "lengths": input_lengths}
@@ -348,6 +391,7 @@ class XLMModelTest(ModelTesterMixin, unittest.TestCase):
XLMForSequenceClassification,
XLMForQuestionAnsweringSimple,
XLMForTokenClassification,
XLMForMultipleChoice,
)
if is_torch_available()
else ()
@@ -387,6 +431,10 @@ class XLMModelTest(ModelTesterMixin, unittest.TestCase):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_xlm_token_classif(*config_and_inputs)
def test_xlm_for_multiple_choice(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_xlm_for_multiple_choice(*config_and_inputs)
@slow
def test_model_from_pretrained(self):
for model_name in XLM_PRETRAINED_MODEL_ARCHIVE_LIST[:1]: