EngineX-Cambricon/enginex-mlu370-any2any
10
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

init

Browse Source
This commit is contained in:
luopingyi
2025-10-09 16:47:16 +08:00
parent c8feb4deb5
commit e27e3f16bb
5248 changed files with 1778505 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

326
transformers/tests/models/vitdet/test_modeling_vitdet.py Normal file
View File

@@ -0,0 +1,326 @@
# Copyright 2023 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 ViTDet model."""
import unittest
from transformers import VitDetConfig
from transformers.testing_utils import is_flaky, require_torch, torch_device
from transformers.utils import is_torch_available
from ...test_backbone_common import BackboneTesterMixin
from ...test_configuration_common import ConfigTester
from ...test_modeling_common import ModelTesterMixin, floats_tensor, ids_tensor
from ...test_pipeline_mixin import PipelineTesterMixin
if is_torch_available():
import torch
from torch import nn
from transformers import VitDetBackbone, VitDetModel
class VitDetModelTester:
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=2,
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,
):
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.num_patches_one_direction = self.image_size // self.patch_size
self.seq_length = (self.image_size // self.patch_size) ** 2
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 VitDetConfig(
image_size=self.image_size,
pretrain_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 = VitDetModel(config=config)
model.to(torch_device)
model.eval()
result = model(pixel_values)
self.parent.assertEqual(
result.last_hidden_state.shape,
(self.batch_size, self.hidden_size, self.num_patches_one_direction, self.num_patches_one_direction),
)
def create_and_check_backbone(self, config, pixel_values, labels):
model = VitDetBackbone(config=config)
model.to(torch_device)
model.eval()
result = model(pixel_values)
# verify hidden states
self.parent.assertEqual(len(result.feature_maps), len(config.out_features))
self.parent.assertListEqual(
list(result.feature_maps[0].shape),
[self.batch_size, self.hidden_size, self.num_patches_one_direction, self.num_patches_one_direction],
)
# verify channels
self.parent.assertEqual(len(model.channels), len(config.out_features))
self.parent.assertListEqual(model.channels, [config.hidden_size])
# verify backbone works with out_features=None
config.out_features = None
model = VitDetBackbone(config=config)
model.to(torch_device)
model.eval()
result = model(pixel_values)
# verify feature maps
self.parent.assertEqual(len(result.feature_maps), 1)
self.parent.assertListEqual(
list(result.feature_maps[0].shape),
[self.batch_size, self.hidden_size, self.num_patches_one_direction, self.num_patches_one_direction],
)
# verify channels
self.parent.assertEqual(len(model.channels), 1)
self.parent.assertListEqual(model.channels, [config.hidden_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 VitDetModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
"""
Here we also overwrite some of the tests of test_modeling_common.py, as VitDet does not use input_ids, inputs_embeds,
attention_mask and seq_length.
"""
all_model_classes = (VitDetModel, VitDetBackbone) if is_torch_available() else ()
pipeline_model_mapping = {"feature-extraction": VitDetModel} if is_torch_available() else {}
fx_compatible = False
test_pruning = False
test_resize_embeddings = False
test_head_masking = False
test_torch_exportable = True
def setUp(self):
self.model_tester = VitDetModelTester(self)
self.config_tester = ConfigTester(self, config_class=VitDetConfig, has_text_modality=False, hidden_size=37)
@is_flaky(max_attempts=3, description="`torch.nn.init.trunc_normal_` is flaky.")
def test_initialization(self):
super().test_initialization()
# TODO: Fix me (once this model gets more usage)
@unittest.skip(reason="Does not work on the tiny model as we keep hitting edge cases.")
def test_cpu_offload(self):
super().test_cpu_offload()
# TODO: Fix me (once this model gets more usage)
@unittest.skip(reason="Does not work on the tiny model as we keep hitting edge cases.")
def test_disk_offload_bin(self):
super().test_disk_offload()
@unittest.skip(reason="Does not work on the tiny model as we keep hitting edge cases.")
def test_disk_offload_safetensors(self):
super().test_disk_offload()
# TODO: Fix me (once this model gets more usage)
@unittest.skip(reason="Does not work on the tiny model as we keep hitting edge cases.")
def test_model_parallelism(self):
super().test_model_parallelism()
def test_config(self):
self.config_tester.run_common_tests()
@unittest.skip(reason="VitDet does not use inputs_embeds")
def test_inputs_embeds(self):
pass
def test_model_get_set_embeddings(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_model(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_model(*config_and_inputs)
def test_backbone(self):
config_and_inputs = self.model_tester.prepare_config_and_inputs()
self.model_tester.create_and_check_backbone(*config_and_inputs)
def test_hidden_states_output(self):
def check_hidden_states_output(inputs_dict, config, model_class):
model = model_class(config)
model.to(torch_device)
model.eval()
with torch.no_grad():
outputs = model(**self._prepare_for_class(inputs_dict, model_class))
hidden_states = outputs.hidden_states
expected_num_stages = self.model_tester.num_hidden_layers
self.assertEqual(len(hidden_states), expected_num_stages + 1)
# VitDet's feature maps are of shape (batch_size, num_channels, height, width)
self.assertListEqual(
list(hidden_states[0].shape[-2:]),
[
self.model_tester.num_patches_one_direction,
self.model_tester.num_patches_one_direction,
],
)
config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
inputs_dict["output_hidden_states"] = True
check_hidden_states_output(inputs_dict, config, model_class)
# check that output_hidden_states also work using config
del inputs_dict["output_hidden_states"]
config.output_hidden_states = True
check_hidden_states_output(inputs_dict, config, model_class)
# overwrite since VitDet only supports retraining gradients of hidden states
def test_retain_grad_hidden_states_attentions(self):
config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
config.output_hidden_states = True
config.output_attentions = self.has_attentions
# no need to test all models as different heads yield the same functionality
model_class = self.all_model_classes[0]
model = model_class(config)
model.to(torch_device)
inputs = self._prepare_for_class(inputs_dict, model_class)
outputs = model(**inputs)
output = outputs[0]
# Encoder-/Decoder-only models
hidden_states = outputs.hidden_states[0]
hidden_states.retain_grad()
output.flatten()[0].backward(retain_graph=True)
self.assertIsNotNone(hidden_states.grad)
@unittest.skip(reason="VitDet does not support feedforward chunking")
def test_feed_forward_chunking(self):
pass
@unittest.skip(reason="VitDet does not have standalone checkpoints since it used as backbone in other models")
def test_model_from_pretrained(self):
pass
def test_non_square_image(self):
non_square_image_size = (32, 40)
patch_size = (2, 2)
config = self.model_tester.get_config()
config.image_size = non_square_image_size
config.patch_size = patch_size
model = VitDetModel(config=config)
model.to(torch_device)
model.eval()
batch_size = self.model_tester.batch_size
# Create a dummy input tensor with non-square spatial dimensions.
pixel_values = floats_tensor(
[batch_size, config.num_channels, non_square_image_size[0], non_square_image_size[1]]
)
result = model(pixel_values)
expected_height = non_square_image_size[0] / patch_size[0]
expected_width = non_square_image_size[1] / patch_size[1]
expected_shape = (batch_size, config.hidden_size, expected_height, expected_width)
self.assertEqual(result.last_hidden_state.shape, expected_shape)
@require_torch
class VitDetBackboneTest(unittest.TestCase, BackboneTesterMixin):
all_model_classes = (VitDetBackbone,) if is_torch_available() else ()
config_class = VitDetConfig
has_attentions = False
def setUp(self):
self.model_tester = VitDetModelTester(self)