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

Update tiny model creation script and some others files (#22006)

Browse Source 
* Update 1

* Update 2

* Update 3

* Update 4

* Update 5

* Update 6

* Update 7

* Update 8

* Update 9

* Update 10

---------

Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>
This commit is contained in:
Yih-Dar
2023-03-07 22:31:14 +01:00
committed by GitHub
parent 8abe4930d3
commit b338414e61
11 changed files with 36 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
src/transformers/models/auto/feature_extraction_auto.py
View File

@@ -82,6 +82,7 @@ FEATURE_EXTRACTOR_MAPPING_NAMES = OrderedDict(
("swinv2", "ViTFeatureExtractor"),
("table-transformer", "DetrFeatureExtractor"),
("timesformer", "VideoMAEFeatureExtractor"),
("tvlt", "TvltFeatureExtractor"),
("unispeech", "Wav2Vec2FeatureExtractor"),
("unispeech-sat", "Wav2Vec2FeatureExtractor"),
("van", "ConvNextFeatureExtractor"),

1
src/transformers/models/auto/image_processing_auto.py
View File

@@ -87,6 +87,7 @@ IMAGE_PROCESSOR_MAPPING_NAMES = OrderedDict(
("swinv2", "ViTImageProcessor"),
("table-transformer", "DetrImageProcessor"),
("timesformer", "VideoMAEImageProcessor"),
("tvlt", "TvltImageProcessor"),
("upernet", "SegformerImageProcessor"),
("van", "ConvNextImageProcessor"),
("videomae", "VideoMAEImageProcessor"),

1
src/transformers/models/auto/processing_auto.py
View File

@@ -65,6 +65,7 @@ PROCESSOR_MAPPING_NAMES = OrderedDict(
("speech_to_text_2", "Speech2Text2Processor"),
("speecht5", "SpeechT5Processor"),
("trocr", "TrOCRProcessor"),
("tvlt", "TvltProcessor"),
("unispeech", "Wav2Vec2Processor"),
("unispeech-sat", "Wav2Vec2Processor"),
("vilt", "ViltProcessor"),

3
src/transformers/models/gptsan_japanese/configuration_gptsan_japanese.py
View File

@@ -31,8 +31,7 @@ class GPTSanJapaneseConfig(PretrainedConfig):
This is the configuration class to store the configuration of a [`GPTSanJapaneseModel`]. It is used to instantiate
a GPTSANJapanese model according to the specified arguments, defining the model architecture. Instantiating a
configuration with the defaults will yield a similar configuration to that of the GPTSANJapanese
[tanreinama/GPTSAN-2.8B-spout_is_uniform](https://huggingface.co/tanreinama/GPTSAN-2.8B-spout_is_uniform)
architecture.
[Tanrei/GPTSAN-japanese](https://huggingface.co/Tanrei/GPTSAN-japanese) architecture.
Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
documentation from [`PretrainedConfig`] for more information.

3
src/transformers/models/timesformer/configuration_timesformer.py
View File

@@ -30,7 +30,8 @@ class TimesformerConfig(PretrainedConfig):
This is the configuration class to store the configuration of a [`TimesformerModel`]. It is used to instantiate a
TimeSformer model according to the specified arguments, defining the model architecture. Instantiating a
configuration with the defaults will yield a similar configuration to that of the TimeSformer
[facebook/timesformer](https://huggingface.co/facebook/timesformer-base-finetuned-k600) architecture.
[facebook/timesformer-base-finetuned-k600](https://huggingface.co/facebook/timesformer-base-finetuned-k600)
architecture.
Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
documentation from [`PretrainedConfig`] for more information.

2
src/transformers/models/tvlt/configuration_tvlt.py
View File

@@ -30,7 +30,7 @@ class TvltConfig(PretrainedConfig):
This is the configuration class to store the configuration of a [`TvltModel`]. It is used to instantiate a TVLT
model according to the specified arguments, defining the model architecture. Instantiating a configuration with the
defaults will yield a similar configuration to that of the TVLT
[TVLT/tvlt-base](https://huggingface.co/ZinengTang/tvlt-base) architecture.
[ZinengTang/tvlt-base](https://huggingface.co/ZinengTang/tvlt-base) architecture.
Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
documentation from [`PretrainedConfig`] for more information.

4
src/transformers/models/xmod/configuration_xmod.py
View File

@@ -41,8 +41,8 @@ class XmodConfig(PretrainedConfig):
r"""
This is the configuration class to store the configuration of a [`XmodModel`]. It is used to instantiate an X-MOD
model according to the specified arguments, defining the model architecture. Instantiating a configuration with the
defaults will yield a similar configuration to that of the [xmod-base](https://huggingface.co/facebook/xmod-base)
architecture.
defaults will yield a similar configuration to that of the
[facebook/xmod-base](https://huggingface.co/facebook/xmod-base) architecture.
Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
documentation from [`PretrainedConfig`] for more information.

17
tests/models/oneformer/test_modeling_oneformer.py
View File

@@ -56,6 +56,7 @@ class OneFormerModelTester:
parent,
batch_size=2,
is_training=True,
vocab_size=99,
use_auxiliary_loss=False,
num_queries=10,
num_channels=3,
@@ -69,6 +70,7 @@ class OneFormerModelTester:
self.parent = parent
self.batch_size = batch_size
self.is_training = is_training
self.vocab_size = vocab_size
self.use_auxiliary_loss = use_auxiliary_loss
self.num_queries = num_queries
self.num_channels = num_channels
@@ -84,12 +86,16 @@ class OneFormerModelTester:
torch_device
)
task_inputs = torch.randint(high=49408, size=(self.batch_size, self.sequence_length)).to(torch_device).long()
task_inputs = (
torch.randint(high=self.vocab_size, size=(self.batch_size, self.sequence_length)).to(torch_device).long()
)
pixel_mask = torch.ones([self.batch_size, self.min_size, self.max_size], device=torch_device)
text_inputs = (
torch.randint(high=49408, size=(self.batch_size, self.num_queries - self.n_ctx, self.sequence_length))
torch.randint(
high=self.vocab_size, size=(self.batch_size, self.num_queries - self.n_ctx, self.sequence_length)
)
.to(torch_device)
.long()
)
@@ -104,6 +110,7 @@ class OneFormerModelTester:
def get_config(self):
config = OneFormerConfig(
text_encoder_vocab_size=self.vocab_size,
hidden_size=self.hidden_dim,
)
@@ -303,8 +310,10 @@ class OneFormerModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCas
size = (self.model_tester.min_size,) * 2
inputs = {
"pixel_values": torch.randn((2, 3, *size), device=torch_device),
"task_inputs": torch.randint(high=49408, size=(2, 77), device=torch_device).long(),
"text_inputs": torch.randint(high=49408, size=(2, 134, 77), device=torch_device).long(),
"task_inputs": torch.randint(high=self.model_tester.vocab_size, size=(2, 77), device=torch_device).long(),
"text_inputs": torch.randint(
high=self.model_tester.vocab_size, size=(2, 134, 77), device=torch_device
).long(),
"mask_labels": torch.randn((2, 150, *size), device=torch_device),
"class_labels": torch.zeros(2, 150, device=torch_device).long(),
}

3
tests/models/speecht5/test_modeling_speecht5.py
View File

@@ -103,6 +103,7 @@ class SpeechT5ModelTester:
batch_size=13,
seq_length=7,
is_training=False,
vocab_size=81,
hidden_size=24,
num_hidden_layers=4,
num_attention_heads=2,
@@ -112,6 +113,7 @@ class SpeechT5ModelTester:
self.batch_size = batch_size
self.seq_length = seq_length
self.is_training = is_training
self.vocab_size = vocab_size
self.hidden_size = hidden_size
self.num_hidden_layers = num_hidden_layers
self.num_attention_heads = num_attention_heads
@@ -140,6 +142,7 @@ class SpeechT5ModelTester:
def get_config(self):
return SpeechT5Config(
vocab_size=self.vocab_size,
hidden_size=self.hidden_size,
encoder_layers=self.num_hidden_layers,
decoder_layers=self.num_hidden_layers,

10
utils/check_config_docstrings.py
View File

@@ -51,10 +51,12 @@ def get_checkpoint_from_config_class(config_class):
config_source = inspect.getsource(config_class)
checkpoints = _re_checkpoint.findall(config_source)
for checkpoint in checkpoints:
# Each `checkpoint` is a tuple of a checkpoint name and a checkpoint link.
# For example, `('bert-base-uncased', 'https://huggingface.co/bert-base-uncased')`
ckpt_name, ckpt_link = checkpoint
# Each `checkpoint` is a tuple of a checkpoint name and a checkpoint link.
# For example, `('bert-base-uncased', 'https://huggingface.co/bert-base-uncased')`
for ckpt_name, ckpt_link in checkpoints:
# allow the link to end with `/`
if ckpt_link.endswith("/"):
ckpt_link = ckpt_link[:-1]
# verify the checkpoint name corresponds to the checkpoint link
ckpt_link_from_name = f"https://huggingface.co/{ckpt_name}"

5
utils/create_dummy_models.py
View File

@@ -782,6 +782,11 @@ def get_config_overrides(config_class, processors):
# CLIP-like models have `text_model_tester` and `vision_model_tester`, and we need to pass `vocab_size` to
# `text_model_tester` via `text_kwargs`. The same trick is also necessary for `Flava`.
if config_class.__name__ in [
"AlignConfig",
"AltCLIPConfig",
"ChineseCLIPConfig",
"CLIPSegConfig",
"ClapConfig",
"CLIPConfig",
"GroupViTConfig",
"OwlViTConfig",