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

Move test model folders (#17034)

Browse Source 
* move test model folders (TODO: fix imports and others)

* fix (potentially partially) imports (in model test modules)

* fix (potentially partially) imports (in tokenization test modules)

* fix (potentially partially) imports (in feature extraction test modules)

* fix import utils.test_modeling_tf_core

* fix path ../fixtures/

* fix imports about generation.test_generation_flax_utils

* fix more imports

* fix fixture path

* fix get_test_dir

* update module_to_test_file

* fix get_tests_dir from wrong transformers.utils

* update config.yml (CircleCI)

* fix style

* remove missing imports

* update new model script

* update check_repo

* update SPECIAL_MODULE_TO_TEST_MAP

* fix style

* add __init__

* update self-scheduled

* fix add_new_model scripts

* check one way to get location back

* python setup.py build install

* fix import in test auto

* update self-scheduled.yml

* update slack notification script

* Add comments about artifact names

* fix for yolos

Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>
This commit is contained in:
Yih-Dar
2022-05-03 14:42:02 +02:00
committed by GitHub
parent cd9274d010
commit 19420fd99e
408 changed files with 616 additions and 607 deletions
Download Patch File Download Diff File Expand all files Collapse all files

0
tests/models/vit/__init__.py Normal file
View File

191
tests/models/vit/test_feature_extraction_vit.py Normal file
View File

@@ -0,0 +1,191 @@
# coding=utf-8
# Copyright 2021 HuggingFace Inc.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# 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.
import unittest
import numpy as np
from transformers.testing_utils import require_torch, require_vision
from transformers.utils import is_torch_available, is_vision_available
from ...test_feature_extraction_common import FeatureExtractionSavingTestMixin, prepare_image_inputs
if is_torch_available():
import torch
if is_vision_available():
from PIL import Image
from transformers import ViTFeatureExtractor
class ViTFeatureExtractionTester(unittest.TestCase):
def __init__(
self,
parent,
batch_size=7,
num_channels=3,
image_size=18,
min_resolution=30,
max_resolution=400,
do_resize=True,
size=18,
do_normalize=True,
image_mean=[0.5, 0.5, 0.5],
image_std=[0.5, 0.5, 0.5],
):
self.parent = parent
self.batch_size = batch_size
self.num_channels = num_channels
self.image_size = image_size
self.min_resolution = min_resolution
self.max_resolution = max_resolution
self.do_resize = do_resize
self.size = size
self.do_normalize = do_normalize
self.image_mean = image_mean
self.image_std = image_std
def prepare_feat_extract_dict(self):
return {
"image_mean": self.image_mean,
"image_std": self.image_std,
"do_normalize": self.do_normalize,
"do_resize": self.do_resize,
"size": self.size,
}
@require_torch
@require_vision
class ViTFeatureExtractionTest(FeatureExtractionSavingTestMixin, unittest.TestCase):
feature_extraction_class = ViTFeatureExtractor if is_vision_available() else None
def setUp(self):
self.feature_extract_tester = ViTFeatureExtractionTester(self)
@property
def feat_extract_dict(self):
return self.feature_extract_tester.prepare_feat_extract_dict()
def test_feat_extract_properties(self):
feature_extractor = self.feature_extraction_class(**self.feat_extract_dict)
self.assertTrue(hasattr(feature_extractor, "image_mean"))
self.assertTrue(hasattr(feature_extractor, "image_std"))
self.assertTrue(hasattr(feature_extractor, "do_normalize"))
self.assertTrue(hasattr(feature_extractor, "do_resize"))
self.assertTrue(hasattr(feature_extractor, "size"))
def test_batch_feature(self):
pass
def test_call_pil(self):
# Initialize feature_extractor
feature_extractor = self.feature_extraction_class(**self.feat_extract_dict)
# create random PIL images
image_inputs = prepare_image_inputs(self.feature_extract_tester, equal_resolution=False)
for image in image_inputs:
self.assertIsInstance(image, Image.Image)
# Test not batched input
encoded_images = feature_extractor(image_inputs[0], return_tensors="pt").pixel_values
self.assertEqual(
encoded_images.shape,
(
1,
self.feature_extract_tester.num_channels,
self.feature_extract_tester.size,
self.feature_extract_tester.size,
),
)
# Test batched
encoded_images = feature_extractor(image_inputs, return_tensors="pt").pixel_values
self.assertEqual(
encoded_images.shape,
(
self.feature_extract_tester.batch_size,
self.feature_extract_tester.num_channels,
self.feature_extract_tester.size,
self.feature_extract_tester.size,
),
)
def test_call_numpy(self):
# Initialize feature_extractor
feature_extractor = self.feature_extraction_class(**self.feat_extract_dict)
# create random numpy tensors
image_inputs = prepare_image_inputs(self.feature_extract_tester, equal_resolution=False, numpify=True)
for image in image_inputs:
self.assertIsInstance(image, np.ndarray)
# Test not batched input
encoded_images = feature_extractor(image_inputs[0], return_tensors="pt").pixel_values
self.assertEqual(
encoded_images.shape,
(
1,
self.feature_extract_tester.num_channels,
self.feature_extract_tester.size,
self.feature_extract_tester.size,
),
)
# Test batched
encoded_images = feature_extractor(image_inputs, return_tensors="pt").pixel_values
self.assertEqual(
encoded_images.shape,
(
self.feature_extract_tester.batch_size,
self.feature_extract_tester.num_channels,
self.feature_extract_tester.size,
self.feature_extract_tester.size,
),
)
def test_call_pytorch(self):
# Initialize feature_extractor
feature_extractor = self.feature_extraction_class(**self.feat_extract_dict)
# create random PyTorch tensors
image_inputs = prepare_image_inputs(self.feature_extract_tester, equal_resolution=False, torchify=True)
for image in image_inputs:
self.assertIsInstance(image, torch.Tensor)
# Test not batched input
encoded_images = feature_extractor(image_inputs[0], return_tensors="pt").pixel_values
self.assertEqual(
encoded_images.shape,
(
1,
self.feature_extract_tester.num_channels,
self.feature_extract_tester.size,
self.feature_extract_tester.size,
),
)
# Test batched
encoded_images = feature_extractor(image_inputs, return_tensors="pt").pixel_values
self.assertEqual(
encoded_images.shape,
(
self.feature_extract_tester.batch_size,
self.feature_extract_tester.num_channels,
self.feature_extract_tester.size,
self.feature_extract_tester.size,
),
)

168
tests/models/vit/test_modeling_flax_vit.py Normal file
View File

@@ -0,0 +1,168 @@
# Copyright 2021 The HuggingFace Team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# 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.
import inspect
import unittest
import numpy as np
from transformers import ViTConfig, is_flax_available
from transformers.testing_utils import require_flax, slow
from ...test_configuration_common import ConfigTester
from ...test_modeling_flax_common import FlaxModelTesterMixin, floats_tensor
if is_flax_available():
import jax
from transformers.models.vit.modeling_flax_vit import FlaxViTForImageClassification, FlaxViTModel
class FlaxViTModelTester(unittest.TestCase):
def __init__(
self,
parent,
batch_size=13,
image_size=30,
patch_size=2,
num_channels=3,
is_training=True,
use_labels=True,
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,
type_sequence_label_size=10,
initializer_range=0.02,
):
self.parent = parent
self.batch_size = batch_size
self.image_size = image_size
self.patch_size = patch_size
self.num_channels = num_channels
self.is_training = is_training
self.use_labels = use_labels
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.type_sequence_label_size = type_sequence_label_size
self.initializer_range = initializer_range
# in ViT, the seq length equals the number of patches + 1 (we add 1 for the [CLS] token)
num_patches = (image_size // patch_size) ** 2
self.seq_length = num_patches + 1
def prepare_config_and_inputs(self):
pixel_values = floats_tensor([self.batch_size, self.num_channels, self.image_size, self.image_size])
config = ViTConfig(
image_size=self.image_size,
patch_size=self.patch_size,
num_channels=self.num_channels,
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,
is_decoder=False,
initializer_range=self.initializer_range,
)
return config, pixel_values
def create_and_check_model(self, config, pixel_values, labels):
model = FlaxViTModel(config=config)
result = model(pixel_values)
# expected sequence length = num_patches + 1 (we add 1 for the [CLS] token)
image_size = (self.image_size, self.image_size)
patch_size = (self.patch_size, self.patch_size)
num_patches = (image_size[1] // patch_size[1]) * (image_size[0] // patch_size[0])
self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, num_patches + 1, self.hidden_size))
def prepare_config_and_inputs_for_common(self):
config_and_inputs = self.prepare_config_and_inputs()
(
config,
pixel_values,
) = config_and_inputs
inputs_dict = {"pixel_values": pixel_values}
return config, inputs_dict
@require_flax
class FlaxViTModelTest(FlaxModelTesterMixin, unittest.TestCase):
all_model_classes = (FlaxViTModel, FlaxViTForImageClassification) if is_flax_available() else ()
def setUp(self) -> None:
self.model_tester = FlaxViTModelTester(self)
self.config_tester = ConfigTester(self, config_class=ViTConfig, has_text_modality=False, hidden_size=37)
def test_config(self):
self.config_tester.run_common_tests()
# We neeed to override this test because ViT's forward signature is different than text models.
def test_forward_signature(self):
config, _ = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
model = model_class(config)
signature = inspect.signature(model.__call__)
# signature.parameters is an OrderedDict => so arg_names order is deterministic
arg_names = [*signature.parameters.keys()]
expected_arg_names = ["pixel_values"]
self.assertListEqual(arg_names[:1], expected_arg_names)
# We need to override this test because ViT expects pixel_values instead of input_ids
def test_jit_compilation(self):
config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
with self.subTest(model_class.__name__):
prepared_inputs_dict = self._prepare_for_class(inputs_dict, model_class)
model = model_class(config)
@jax.jit
def model_jitted(pixel_values, **kwargs):
return model(pixel_values=pixel_values, **kwargs)
with self.subTest("JIT Enabled"):
jitted_outputs = model_jitted(**prepared_inputs_dict).to_tuple()
with self.subTest("JIT Disabled"):
with jax.disable_jit():
outputs = model_jitted(**prepared_inputs_dict).to_tuple()
self.assertEqual(len(outputs), len(jitted_outputs))
for jitted_output, output in zip(jitted_outputs, outputs):
self.assertEqual(jitted_output.shape, output.shape)
@slow
def test_model_from_pretrained(self):
for model_class_name in self.all_model_classes:
model = model_class_name.from_pretrained("google/vit-base-patch16-224")
outputs = model(np.ones((1, 3, 224, 224)))
self.assertIsNotNone(outputs)

237
tests/models/vit/test_modeling_tf_vit.py Normal file
View File

@@ -0,0 +1,237 @@
# coding=utf-8
# Copyright 2021 The HuggingFace Inc. team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# 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.
""" Testing suite for the TensorFlow ViT model. """
import inspect
import unittest
from transformers import ViTConfig
from transformers.testing_utils import require_tf, require_vision, slow
from transformers.utils import cached_property, is_tf_available, is_vision_available
from ...test_configuration_common import ConfigTester
from ...test_modeling_tf_common import TFModelTesterMixin, floats_tensor, ids_tensor
if is_tf_available():
import tensorflow as tf
from transformers import TFViTForImageClassification, TFViTModel
if is_vision_available():
from PIL import Image
from transformers import ViTFeatureExtractor
class TFViTModelTester:
def __init__(
self,
parent,
batch_size=13,
image_size=30,
patch_size=2,
num_channels=3,
is_training=True,
use_labels=True,
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,
type_sequence_label_size=10,
initializer_range=0.02,
num_labels=3,
scope=None,
):
self.parent = parent
self.batch_size = batch_size
self.image_size = image_size
self.patch_size = patch_size
self.num_channels = num_channels
self.is_training = is_training
self.use_labels = use_labels
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.type_sequence_label_size = type_sequence_label_size
self.initializer_range = initializer_range
self.scope = scope
# in ViT, the seq length equals the number of patches + 1 (we add 1 for the [CLS] token)
num_patches = (image_size // patch_size) ** 2
self.seq_length = num_patches + 1
def prepare_config_and_inputs(self):
pixel_values = floats_tensor([self.batch_size, self.num_channels, self.image_size, self.image_size])
labels = None
if self.use_labels:
labels = ids_tensor([self.batch_size], self.type_sequence_label_size)
config = self.get_config()
return config, pixel_values, labels
def get_config(self):
return ViTConfig(
image_size=self.image_size,
patch_size=self.patch_size,
num_channels=self.num_channels,
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,
is_decoder=False,
initializer_range=self.initializer_range,
)
def create_and_check_model(self, config, pixel_values, labels):
model = TFViTModel(config=config)
result = model(pixel_values, training=False)
self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, self.seq_length, self.hidden_size))
# Test with an image with different size than the one specified in config.
image_size = self.image_size // 2
pixel_values = pixel_values[:, :, :image_size, :image_size]
result = model(pixel_values, interpolate_pos_encoding=True, training=False)
seq_length = (image_size // self.patch_size) ** 2 + 1
self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, seq_length, self.hidden_size))
def create_and_check_for_image_classification(self, config, pixel_values, labels):
config.num_labels = self.type_sequence_label_size
model = TFViTForImageClassification(config)
result = model(pixel_values, labels=labels, training=False)
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.type_sequence_label_size))
# Test with an image with different size than the one specified in config.
image_size = self.image_size // 2
pixel_values = pixel_values[:, :, :image_size, :image_size]
result = model(pixel_values, interpolate_pos_encoding=True, training=False)
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.type_sequence_label_size))
def prepare_config_and_inputs_for_common(self):
config_and_inputs = self.prepare_config_and_inputs()
config, pixel_values, labels = config_and_inputs
inputs_dict = {"pixel_values": pixel_values}
return config, inputs_dict
@require_tf
class TFViTModelTest(TFModelTesterMixin, unittest.TestCase):
"""
Here we also overwrite some of the tests of test_modeling_tf_common.py, as ViT does not use input_ids, inputs_embeds,
attention_mask and seq_length.
"""
all_model_classes = (TFViTModel, TFViTForImageClassification) if is_tf_available() else ()
test_resize_embeddings = False
test_head_masking = False
test_onnx = False
def setUp(self):
self.model_tester = TFViTModelTester(self)
self.config_tester = ConfigTester(self, config_class=ViTConfig, has_text_modality=False, hidden_size=37)
def test_config(self):
self.config_tester.run_common_tests()
@unittest.skip(reason="ViT does not use inputs_embeds")
def test_inputs_embeds(self):
pass
@unittest.skip(reason="ViT does not use inputs_embeds")
def test_graph_mode_with_inputs_embeds(self):
pass
def test_model_common_attributes(self):
config, _ = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
model = model_class(config)
self.assertIsInstance(model.get_input_embeddings(), (tf.keras.layers.Layer))
x = model.get_output_embeddings()
self.assertTrue(x is None or isinstance(x, tf.keras.layers.Layer))
def test_forward_signature(self):
config, _ = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
model = model_class(config)
signature = inspect.signature(model.call)
# signature.parameters is an OrderedDict => so arg_names order is deterministic
arg_names = [*signature.parameters.keys()]
expected_arg_names = ["pixel_values"]
self.assertListEqual(arg_names[:1], expected_arg_names)
def test_model(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_model(*config_and_inputs)
def test_for_image_classification(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_image_classification(*config_and_inputs)
@slow
def test_model_from_pretrained(self):
model = TFViTModel.from_pretrained("google/vit-base-patch16-224")
self.assertIsNotNone(model)
# We will verify our results on an image of cute cats
def prepare_img():
image = Image.open("./tests/fixtures/tests_samples/COCO/000000039769.png")
return image
@require_tf
@require_vision
class TFViTModelIntegrationTest(unittest.TestCase):
@cached_property
def default_feature_extractor(self):
return ViTFeatureExtractor.from_pretrained("google/vit-base-patch16-224") if is_vision_available() else None
@slow
def test_inference_image_classification_head(self):
model = TFViTForImageClassification.from_pretrained("google/vit-base-patch16-224")
feature_extractor = self.default_feature_extractor
image = prepare_img()
inputs = feature_extractor(images=image, return_tensors="tf")
# forward pass
outputs = model(**inputs)
# verify the logits
expected_shape = tf.TensorShape((1, 1000))
self.assertEqual(outputs.logits.shape, expected_shape)
expected_slice = tf.constant([-0.2744, 0.8215, -0.0836])
tf.debugging.assert_near(outputs.logits[0, :3], expected_slice, atol=1e-4)

241
tests/models/vit/test_modeling_vit.py Normal file
View File

@@ -0,0 +1,241 @@
# coding=utf-8
# Copyright 2021 The HuggingFace Inc. team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# 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.
""" Testing suite for the PyTorch ViT model. """
import inspect
import unittest
from transformers import ViTConfig
from transformers.testing_utils import require_torch, require_vision, slow, torch_device
from transformers.utils import cached_property, is_torch_available, is_vision_available
from ...test_configuration_common import ConfigTester
from ...test_modeling_common import ModelTesterMixin, floats_tensor, ids_tensor
if is_torch_available():
import torch
from torch import nn
from transformers import ViTForImageClassification, ViTForMaskedImageModeling, ViTModel
from transformers.models.vit.modeling_vit import VIT_PRETRAINED_MODEL_ARCHIVE_LIST
if is_vision_available():
from PIL import Image
from transformers import ViTFeatureExtractor
class ViTModelTester:
def __init__(
self,
parent,
batch_size=13,
image_size=30,
patch_size=2,
num_channels=3,
is_training=True,
use_labels=True,
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,
type_sequence_label_size=10,
initializer_range=0.02,
scope=None,
encoder_stride=2,
):
self.parent = parent
self.batch_size = batch_size
self.image_size = image_size
self.patch_size = patch_size
self.num_channels = num_channels
self.is_training = is_training
self.use_labels = use_labels
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.type_sequence_label_size = type_sequence_label_size
self.initializer_range = initializer_range
self.scope = scope
self.encoder_stride = encoder_stride
# in ViT, the seq length equals the number of patches + 1 (we add 1 for the [CLS] token)
num_patches = (image_size // patch_size) ** 2
self.seq_length = num_patches + 1
def prepare_config_and_inputs(self):
pixel_values = floats_tensor([self.batch_size, self.num_channels, self.image_size, self.image_size])
labels = None
if self.use_labels:
labels = ids_tensor([self.batch_size], self.type_sequence_label_size)
config = self.get_config()
return config, pixel_values, labels
def get_config(self):
return ViTConfig(
image_size=self.image_size,
patch_size=self.patch_size,
num_channels=self.num_channels,
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,
is_decoder=False,
initializer_range=self.initializer_range,
encoder_stride=self.encoder_stride,
)
def create_and_check_model(self, config, pixel_values, labels):
model = ViTModel(config=config)
model.to(torch_device)
model.eval()
result = model(pixel_values)
self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, self.seq_length, self.hidden_size))
def create_and_check_for_image_classification(self, config, pixel_values, labels):
config.num_labels = self.type_sequence_label_size
model = ViTForImageClassification(config)
model.to(torch_device)
model.eval()
result = model(pixel_values, labels=labels)
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.type_sequence_label_size))
def prepare_config_and_inputs_for_common(self):
config_and_inputs = self.prepare_config_and_inputs()
(
config,
pixel_values,
labels,
) = config_and_inputs
inputs_dict = {"pixel_values": pixel_values}
return config, inputs_dict
@require_torch
class ViTModelTest(ModelTesterMixin, unittest.TestCase):
"""
Here we also overwrite some of the tests of test_modeling_common.py, as ViT does not use input_ids, inputs_embeds,
attention_mask and seq_length.
"""
all_model_classes = (
(
ViTModel,
ViTForImageClassification,
ViTForMaskedImageModeling,
)
if is_torch_available()
else ()
)
test_pruning = False
test_resize_embeddings = False
test_head_masking = False
def setUp(self):
self.model_tester = ViTModelTester(self)
self.config_tester = ConfigTester(self, config_class=ViTConfig, has_text_modality=False, hidden_size=37)
def test_config(self):
self.config_tester.run_common_tests()
@unittest.skip(reason="ViT does not use inputs_embeds")
def test_inputs_embeds(self):
pass
def test_model_common_attributes(self):
config, _ = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
model = model_class(config)
self.assertIsInstance(model.get_input_embeddings(), (nn.Module))
x = model.get_output_embeddings()
self.assertTrue(x is None or isinstance(x, nn.Linear))
def test_forward_signature(self):
config, _ = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
model = model_class(config)
signature = inspect.signature(model.forward)
# signature.parameters is an OrderedDict => so arg_names order is deterministic
arg_names = [*signature.parameters.keys()]
expected_arg_names = ["pixel_values"]
self.assertListEqual(arg_names[:1], expected_arg_names)
def test_model(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_model(*config_and_inputs)
def test_for_image_classification(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_for_image_classification(*config_and_inputs)
@slow
def test_model_from_pretrained(self):
for model_name in VIT_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
model = ViTModel.from_pretrained(model_name)
self.assertIsNotNone(model)
# We will verify our results on an image of cute cats
def prepare_img():
image = Image.open("./tests/fixtures/tests_samples/COCO/000000039769.png")
return image
@require_torch
@require_vision
class ViTModelIntegrationTest(unittest.TestCase):
@cached_property
def default_feature_extractor(self):
return ViTFeatureExtractor.from_pretrained("google/vit-base-patch16-224") if is_vision_available() else None
@slow
def test_inference_image_classification_head(self):
model = ViTForImageClassification.from_pretrained("google/vit-base-patch16-224").to(torch_device)
feature_extractor = self.default_feature_extractor
image = prepare_img()
inputs = feature_extractor(images=image, return_tensors="pt").to(torch_device)
# forward pass
with torch.no_grad():
outputs = model(**inputs)
# verify the logits
expected_shape = torch.Size((1, 1000))
self.assertEqual(outputs.logits.shape, expected_shape)
expected_slice = torch.tensor([-0.2744, 0.8215, -0.0836]).to(torch_device)
self.assertTrue(torch.allclose(outputs.logits[0, :3], expected_slice, atol=1e-4))