check_repo.py 31.4 KB
Newer Older
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
# coding=utf-8
# Copyright 2020 The HuggingFace Inc. team.
#
# 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.

16
17
18
19
import importlib
import inspect
import os
import re
20
import warnings
21
from collections import OrderedDict
22
from difflib import get_close_matches
23
from pathlib import Path
24

25
from transformers import is_flax_available, is_tf_available, is_torch_available
26
from transformers.models.auto import get_values
27
from transformers.utils import ENV_VARS_TRUE_VALUES
28

29
30
31
32
33

# All paths are set with the intent you should run this script from the root of the repo with the command
# python utils/check_repo.py
PATH_TO_TRANSFORMERS = "src/transformers"
PATH_TO_TESTS = "tests"
34
PATH_TO_DOC = "docs/source/en"
35

36
37
38
# Update this list with models that are supposed to be private.
PRIVATE_MODELS = [
    "DPRSpanPredictor",
Daniel Stancl's avatar
Daniel Stancl committed
39
    "LongT5Stack",
Li-Huai (Allan) Lin's avatar
Li-Huai (Allan) Lin committed
40
    "RealmBertModel",
41
42
43
44
    "T5Stack",
    "TFDPRSpanPredictor",
]

45
46
# Update this list for models that are not tested with a comment explaining the reason it should not be.
# Being in this list is an exception and should **not** be the rule.
47
IGNORE_NON_TESTED = PRIVATE_MODELS.copy() + [
48
    # models to ignore for not tested
Younes Belkada's avatar
Younes Belkada committed
49
    "OPTDecoder",  # Building part of bigger (tested) model.
50
    "DecisionTransformerGPT2Model",  # Building part of bigger (tested) model.
NielsRogge's avatar
NielsRogge committed
51
    "SegformerDecodeHead",  # Building part of bigger (tested) model.
Gunjan Chhablani's avatar
Gunjan Chhablani committed
52
53
54
    "PLBartEncoder",  # Building part of bigger (tested) model.
    "PLBartDecoder",  # Building part of bigger (tested) model.
    "PLBartDecoderWrapper",  # Building part of bigger (tested) model.
Vasudev Gupta's avatar
Vasudev Gupta committed
55
56
57
    "BigBirdPegasusEncoder",  # Building part of bigger (tested) model.
    "BigBirdPegasusDecoder",  # Building part of bigger (tested) model.
    "BigBirdPegasusDecoderWrapper",  # Building part of bigger (tested) model.
NielsRogge's avatar
NielsRogge committed
58
59
60
    "DetrEncoder",  # Building part of bigger (tested) model.
    "DetrDecoder",  # Building part of bigger (tested) model.
    "DetrDecoderWrapper",  # Building part of bigger (tested) model.
Suraj Patil's avatar
Suraj Patil committed
61
62
    "M2M100Encoder",  # Building part of bigger (tested) model.
    "M2M100Decoder",  # Building part of bigger (tested) model.
Chan Woo Kim's avatar
Chan Woo Kim committed
63
    "MCTCTEncoder",  # Building part of bigger (tested) model.
Suraj Patil's avatar
Suraj Patil committed
64
65
    "Speech2TextEncoder",  # Building part of bigger (tested) model.
    "Speech2TextDecoder",  # Building part of bigger (tested) model.
Patrick von Platen's avatar
Patrick von Platen committed
66
67
    "LEDEncoder",  # Building part of bigger (tested) model.
    "LEDDecoder",  # Building part of bigger (tested) model.
68
    "BartDecoderWrapper",  # Building part of bigger (tested) model.
69
    "BartEncoder",  # Building part of bigger (tested) model.
70
    "BertLMHeadModel",  # Needs to be setup as decoder.
71
    "BlenderbotSmallEncoder",  # Building part of bigger (tested) model.
72
    "BlenderbotSmallDecoderWrapper",  # Building part of bigger (tested) model.
73
    "BlenderbotEncoder",  # Building part of bigger (tested) model.
74
    "BlenderbotDecoderWrapper",  # Building part of bigger (tested) model.
75
    "MBartEncoder",  # Building part of bigger (tested) model.
76
    "MBartDecoderWrapper",  # Building part of bigger (tested) model.
77
78
79
80
    "MegatronBertLMHeadModel",  # Building part of bigger (tested) model.
    "MegatronBertEncoder",  # Building part of bigger (tested) model.
    "MegatronBertDecoder",  # Building part of bigger (tested) model.
    "MegatronBertDecoderWrapper",  # Building part of bigger (tested) model.
StevenTang1998's avatar
StevenTang1998 committed
81
82
    "MvpDecoderWrapper",  # Building part of bigger (tested) model.
    "MvpEncoder",  # Building part of bigger (tested) model.
83
    "PegasusEncoder",  # Building part of bigger (tested) model.
84
    "PegasusDecoderWrapper",  # Building part of bigger (tested) model.
Jason Phang's avatar
Jason Phang committed
85
86
87
    "PegasusXEncoder",  # Building part of bigger (tested) model.
    "PegasusXDecoder",  # Building part of bigger (tested) model.
    "PegasusXDecoderWrapper",  # Building part of bigger (tested) model.
88
    "DPREncoder",  # Building part of bigger (tested) model.
89
    "ProphetNetDecoderWrapper",  # Building part of bigger (tested) model.
Li-Huai (Allan) Lin's avatar
Li-Huai (Allan) Lin committed
90
91
92
93
    "RealmBertModel",  # Building part of bigger (tested) model.
    "RealmReader",  # Not regular model.
    "RealmScorer",  # Not regular model.
    "RealmForOpenQA",  # Not regular model.
94
    "ReformerForMaskedLM",  # Needs to be setup as decoder.
95
    "Speech2Text2DecoderWrapper",  # Building part of bigger (tested) model.
Ratthachat (Jung)'s avatar
Ratthachat (Jung) committed
96
    "TFDPREncoder",  # Building part of bigger (tested) model.
97
98
    "TFElectraMainLayer",  # Building part of bigger (tested) model (should it be a TFPreTrainedModel ?)
    "TFRobertaForMultipleChoice",  # TODO: fix
99
    "TrOCRDecoderWrapper",  # Building part of bigger (tested) model.
abhishek thakur's avatar
abhishek thakur committed
100
    "SeparableConv1D",  # Building part of bigger (tested) model.
101
    "FlaxBartForCausalLM",  # Building part of bigger (tested) model.
102
    "FlaxBertForCausalLM",  # Building part of bigger (tested) model. Tested implicitly through FlaxRobertaForCausalLM.
Younes Belkada's avatar
Younes Belkada committed
103
    "OPTDecoderWrapper",
104
    "TFSegformerDecodeHead",  # Not a regular model.
105
106
107
108
109
]

# Update this list with test files that don't have a tester with a `all_model_classes` variable and which don't
# trigger the common tests.
TEST_FILES_WITH_NO_COMMON_TESTS = [
Yih-Dar's avatar
Yih-Dar committed
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
    "models/decision_transformer/test_modeling_decision_transformer.py",
    "models/camembert/test_modeling_camembert.py",
    "models/mt5/test_modeling_flax_mt5.py",
    "models/mbart/test_modeling_mbart.py",
    "models/mt5/test_modeling_mt5.py",
    "models/pegasus/test_modeling_pegasus.py",
    "models/camembert/test_modeling_tf_camembert.py",
    "models/mt5/test_modeling_tf_mt5.py",
    "models/xlm_roberta/test_modeling_tf_xlm_roberta.py",
    "models/xlm_roberta/test_modeling_flax_xlm_roberta.py",
    "models/xlm_prophetnet/test_modeling_xlm_prophetnet.py",
    "models/xlm_roberta/test_modeling_xlm_roberta.py",
    "models/vision_text_dual_encoder/test_modeling_vision_text_dual_encoder.py",
    "models/vision_text_dual_encoder/test_modeling_flax_vision_text_dual_encoder.py",
    "models/decision_transformer/test_modeling_decision_transformer.py",
125
126
]

127
128
# Update this list for models that are not in any of the auto MODEL_XXX_MAPPING. Being in this list is an exception and
# should **not** be the rule.
129
IGNORE_NON_AUTO_CONFIGURED = PRIVATE_MODELS.copy() + [
130
    # models to ignore for model xxx mapping
Jason Phang's avatar
Jason Phang committed
131
132
133
134
135
136
    "PegasusXEncoder",
    "PegasusXDecoder",
    "PegasusXDecoderWrapper",
    "PegasusXEncoder",
    "PegasusXDecoder",
    "PegasusXDecoderWrapper",
NielsRogge's avatar
NielsRogge committed
137
    "DPTForDepthEstimation",
138
    "DecisionTransformerGPT2Model",
NielsRogge's avatar
NielsRogge committed
139
    "GLPNForDepthEstimation",
NielsRogge's avatar
NielsRogge committed
140
141
    "ViltForImagesAndTextClassification",
    "ViltForImageAndTextRetrieval",
142
    "ViltForTokenClassification",
NielsRogge's avatar
NielsRogge committed
143
    "ViltForMaskedLM",
Suraj Patil's avatar
Suraj Patil committed
144
145
146
    "XGLMEncoder",
    "XGLMDecoder",
    "XGLMDecoderWrapper",
NielsRogge's avatar
NielsRogge committed
147
148
    "PerceiverForMultimodalAutoencoding",
    "PerceiverForOpticalFlow",
NielsRogge's avatar
NielsRogge committed
149
    "SegformerDecodeHead",
150
    "TFSegformerDecodeHead",
Kamal Raj's avatar
Kamal Raj committed
151
    "FlaxBeitForMaskedImageModeling",
Gunjan Chhablani's avatar
Gunjan Chhablani committed
152
153
154
    "PLBartEncoder",
    "PLBartDecoder",
    "PLBartDecoderWrapper",
NielsRogge's avatar
NielsRogge committed
155
    "BeitForMaskedImageModeling",
Suraj Patil's avatar
Suraj Patil committed
156
157
    "CLIPTextModel",
    "CLIPVisionModel",
158
159
    "GroupViTTextModel",
    "GroupViTVisionModel",
Yih-Dar's avatar
Yih-Dar committed
160
161
    "TFCLIPTextModel",
    "TFCLIPVisionModel",
Suraj Patil's avatar
Suraj Patil committed
162
163
    "FlaxCLIPTextModel",
    "FlaxCLIPVisionModel",
164
    "FlaxWav2Vec2ForCTC",
NielsRogge's avatar
NielsRogge committed
165
    "DetrForSegmentation",
166
167
    "DPRReader",
    "FlaubertForQuestionAnswering",
168
169
170
171
    "FlavaImageCodebook",
    "FlavaTextModel",
    "FlavaImageModel",
    "FlavaMultimodalModel",
172
    "GPT2DoubleHeadsModel",
173
    "LayoutLMForQuestionAnswering",
Ryokan RI's avatar
Ryokan RI committed
174
    "LukeForMaskedLM",
NielsRogge's avatar
NielsRogge committed
175
176
177
    "LukeForEntityClassification",
    "LukeForEntityPairClassification",
    "LukeForEntitySpanClassification",
178
    "OpenAIGPTDoubleHeadsModel",
179
180
181
    "OwlViTTextModel",
    "OwlViTVisionModel",
    "OwlViTForObjectDetection",
182
183
184
    "RagModel",
    "RagSequenceForGeneration",
    "RagTokenForGeneration",
Li-Huai (Allan) Lin's avatar
Li-Huai (Allan) Lin committed
185
186
187
188
    "RealmEmbedder",
    "RealmForOpenQA",
    "RealmScorer",
    "RealmReader",
Ratthachat (Jung)'s avatar
Ratthachat (Jung) committed
189
    "TFDPRReader",
190
    "TFGPT2DoubleHeadsModel",
191
    "TFLayoutLMForQuestionAnswering",
192
    "TFOpenAIGPTDoubleHeadsModel",
Ratthachat (Jung)'s avatar
Ratthachat (Jung) committed
193
194
195
    "TFRagModel",
    "TFRagSequenceForGeneration",
    "TFRagTokenForGeneration",
196
    "Wav2Vec2ForCTC",
Patrick von Platen's avatar
Patrick von Platen committed
197
    "HubertForCTC",
198
199
    "SEWForCTC",
    "SEWDForCTC",
200
201
    "XLMForQuestionAnswering",
    "XLNetForQuestionAnswering",
abhishek thakur's avatar
abhishek thakur committed
202
    "SeparableConv1D",
Gunjan Chhablani's avatar
Gunjan Chhablani committed
203
204
205
206
    "VisualBertForRegionToPhraseAlignment",
    "VisualBertForVisualReasoning",
    "VisualBertForQuestionAnswering",
    "VisualBertForMultipleChoice",
Will Rice's avatar
Will Rice committed
207
    "TFWav2Vec2ForCTC",
Will Rice's avatar
Will Rice committed
208
    "TFHubertForCTC",
209
    "MaskFormerForInstanceSegmentation",
210
211
]

212
213
214
215
216
217
# Update this list for models that have multiple model types for the same
# model doc
MODEL_TYPE_TO_DOC_MAPPING = OrderedDict(
    [
        ("data2vec-text", "data2vec"),
        ("data2vec-audio", "data2vec"),
218
        ("data2vec-vision", "data2vec"),
NielsRogge's avatar
NielsRogge committed
219
        ("donut-swin", "donut"),
220
221
222
223
    ]
)


224
225
226
227
228
229
230
231
232
# This is to make sure the transformers module imported is the one in the repo.
spec = importlib.util.spec_from_file_location(
    "transformers",
    os.path.join(PATH_TO_TRANSFORMERS, "__init__.py"),
    submodule_search_locations=[PATH_TO_TRANSFORMERS],
)
transformers = spec.loader.load_module()


233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
def check_model_list():
    """Check the model list inside the transformers library."""
    # Get the models from the directory structure of `src/transformers/models/`
    models_dir = os.path.join(PATH_TO_TRANSFORMERS, "models")
    _models = []
    for model in os.listdir(models_dir):
        model_dir = os.path.join(models_dir, model)
        if os.path.isdir(model_dir) and "__init__.py" in os.listdir(model_dir):
            _models.append(model)

    # Get the models from the directory structure of `src/transformers/models/`
    models = [model for model in dir(transformers.models) if not model.startswith("__")]

    missing_models = sorted(list(set(_models).difference(models)))
    if missing_models:
        raise Exception(
            f"The following models should be included in {models_dir}/__init__.py: {','.join(missing_models)}."
        )


253
254
255
# If some modeling modules should be ignored for all checks, they should be added in the nested list
# _ignore_modules of this function.
def get_model_modules():
Patrick von Platen's avatar
Patrick von Platen committed
256
    """Get the model modules inside the transformers library."""
257
258
259
260
261
262
263
264
    _ignore_modules = [
        "modeling_auto",
        "modeling_encoder_decoder",
        "modeling_marian",
        "modeling_mmbt",
        "modeling_outputs",
        "modeling_retribert",
        "modeling_utils",
Sylvain Gugger's avatar
Sylvain Gugger committed
265
        "modeling_flax_auto",
266
        "modeling_flax_encoder_decoder",
Stas Bekman's avatar
Stas Bekman committed
267
        "modeling_flax_utils",
268
        "modeling_speech_encoder_decoder",
269
        "modeling_flax_speech_encoder_decoder",
270
        "modeling_flax_vision_encoder_decoder",
271
272
        "modeling_transfo_xl_utilities",
        "modeling_tf_auto",
273
        "modeling_tf_encoder_decoder",
274
275
276
277
        "modeling_tf_outputs",
        "modeling_tf_pytorch_utils",
        "modeling_tf_utils",
        "modeling_tf_transfo_xl_utilities",
278
        "modeling_tf_vision_encoder_decoder",
279
        "modeling_vision_encoder_decoder",
280
281
    ]
    modules = []
Sylvain Gugger's avatar
Sylvain Gugger committed
282
283
284
285
286
287
288
289
290
    for model in dir(transformers.models):
        # There are some magic dunder attributes in the dir, we ignore them
        if not model.startswith("__"):
            model_module = getattr(transformers.models, model)
            for submodule in dir(model_module):
                if submodule.startswith("modeling") and submodule not in _ignore_modules:
                    modeling_module = getattr(model_module, submodule)
                    if inspect.ismodule(modeling_module):
                        modules.append(modeling_module)
291
292
293
    return modules


294
def get_models(module, include_pretrained=False):
Patrick von Platen's avatar
Patrick von Platen committed
295
    """Get the objects in module that are models."""
296
    models = []
297
    model_classes = (transformers.PreTrainedModel, transformers.TFPreTrainedModel, transformers.FlaxPreTrainedModel)
298
    for attr_name in dir(module):
299
        if not include_pretrained and ("Pretrained" in attr_name or "PreTrained" in attr_name):
300
301
302
303
304
305
306
            continue
        attr = getattr(module, attr_name)
        if isinstance(attr, type) and issubclass(attr, model_classes) and attr.__module__ == module.__name__:
            models.append((attr_name, attr))
    return models


307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
def is_a_private_model(model):
    """Returns True if the model should not be in the main init."""
    if model in PRIVATE_MODELS:
        return True

    # Wrapper, Encoder and Decoder are all privates
    if model.endswith("Wrapper"):
        return True
    if model.endswith("Encoder"):
        return True
    if model.endswith("Decoder"):
        return True
    return False


def check_models_are_in_init():
    """Checks all models defined in the library are in the main init."""
    models_not_in_init = []
    dir_transformers = dir(transformers)
    for module in get_model_modules():
        models_not_in_init += [
            model[0] for model in get_models(module, include_pretrained=True) if model[0] not in dir_transformers
        ]

    # Remove private models
    models_not_in_init = [model for model in models_not_in_init if not is_a_private_model(model)]
    if len(models_not_in_init) > 0:
        raise Exception(f"The following models should be in the main init: {','.join(models_not_in_init)}.")


337
338
339
# If some test_modeling files should be ignored when checking models are all tested, they should be added in the
# nested list _ignore_files of this function.
def get_model_test_files():
Yih-Dar's avatar
Yih-Dar committed
340
341
342
343
344
345
    """Get the model test files.

    The returned files should NOT contain the `tests` (i.e. `PATH_TO_TESTS` defined in this script). They will be
    considered as paths relative to `tests`. A caller has to use `os.path.join(PATH_TO_TESTS, ...)` to access the files.
    """

346
347
348
    _ignore_files = [
        "test_modeling_common",
        "test_modeling_encoder_decoder",
349
        "test_modeling_flax_encoder_decoder",
350
        "test_modeling_flax_speech_encoder_decoder",
351
352
        "test_modeling_marian",
        "test_modeling_tf_common",
353
        "test_modeling_tf_encoder_decoder",
354
355
    ]
    test_files = []
Yih-Dar's avatar
Yih-Dar committed
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
    # Check both `PATH_TO_TESTS` and `PATH_TO_TESTS/models`
    model_test_root = os.path.join(PATH_TO_TESTS, "models")
    model_test_dirs = []
    for x in os.listdir(model_test_root):
        x = os.path.join(model_test_root, x)
        if os.path.isdir(x):
            model_test_dirs.append(x)

    for target_dir in [PATH_TO_TESTS] + model_test_dirs:
        for file_or_dir in os.listdir(target_dir):
            path = os.path.join(target_dir, file_or_dir)
            if os.path.isfile(path):
                filename = os.path.split(path)[-1]
                if "test_modeling" in filename and not os.path.splitext(filename)[0] in _ignore_files:
                    file = os.path.join(*path.split(os.sep)[1:])
                    test_files.append(file)

373
374
375
376
377
378
    return test_files


# This is a bit hacky but I didn't find a way to import the test_file as a module and read inside the tester class
# for the all_model_classes variable.
def find_tested_models(test_file):
Patrick von Platen's avatar
Patrick von Platen committed
379
    """Parse the content of test_file to detect what's in all_model_classes"""
Sylvain Gugger's avatar
Sylvain Gugger committed
380
    # This is a bit hacky but I didn't find a way to import the test_file as a module and read inside the class
381
    with open(os.path.join(PATH_TO_TESTS, test_file), "r", encoding="utf-8", newline="\n") as f:
382
        content = f.read()
Sylvain Gugger's avatar
Sylvain Gugger committed
383
    all_models = re.findall(r"all_model_classes\s+=\s+\(\s*\(([^\)]*)\)", content)
384
385
    # Check with one less parenthesis as well
    all_models += re.findall(r"all_model_classes\s+=\s+\(([^\)]*)\)", content)
Sylvain Gugger's avatar
Sylvain Gugger committed
386
    if len(all_models) > 0:
387
        model_tested = []
Sylvain Gugger's avatar
Sylvain Gugger committed
388
389
390
391
392
        for entry in all_models:
            for line in entry.split(","):
                name = line.strip()
                if len(name) > 0:
                    model_tested.append(name)
393
394
395
396
        return model_tested


def check_models_are_tested(module, test_file):
Patrick von Platen's avatar
Patrick von Platen committed
397
    """Check models defined in module are tested in test_file."""
398
    # XxxPreTrainedModel are not tested
399
400
401
    defined_models = get_models(module)
    tested_models = find_tested_models(test_file)
    if tested_models is None:
402
        if test_file.replace(os.path.sep, "/") in TEST_FILES_WITH_NO_COMMON_TESTS:
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
            return
        return [
            f"{test_file} should define `all_model_classes` to apply common tests to the models it tests. "
            + "If this intentional, add the test filename to `TEST_FILES_WITH_NO_COMMON_TESTS` in the file "
            + "`utils/check_repo.py`."
        ]
    failures = []
    for model_name, _ in defined_models:
        if model_name not in tested_models and model_name not in IGNORE_NON_TESTED:
            failures.append(
                f"{model_name} is defined in {module.__name__} but is not tested in "
                + f"{os.path.join(PATH_TO_TESTS, test_file)}. Add it to the all_model_classes in that file."
                + "If common tests should not applied to that model, add its name to `IGNORE_NON_TESTED`"
                + "in the file `utils/check_repo.py`."
            )
    return failures


def check_all_models_are_tested():
Patrick von Platen's avatar
Patrick von Platen committed
422
    """Check all models are properly tested."""
423
424
425
426
    modules = get_model_modules()
    test_files = get_model_test_files()
    failures = []
    for module in modules:
427
428
        test_file = [file for file in test_files if f"test_{module.__name__.split('.')[-1]}.py" in file]
        if len(test_file) == 0:
429
            failures.append(f"{module.__name__} does not have its corresponding test file {test_file}.")
430
431
432
433
        elif len(test_file) > 1:
            failures.append(f"{module.__name__} has several test files: {test_file}.")
        else:
            test_file = test_file[0]
434
435
436
            new_failures = check_models_are_tested(module, test_file)
            if new_failures is not None:
                failures += new_failures
437
438
439
440
    if len(failures) > 0:
        raise Exception(f"There were {len(failures)} failures:\n" + "\n".join(failures))


441
def get_all_auto_configured_models():
Patrick von Platen's avatar
Patrick von Platen committed
442
    """Return the list of all models in at least one auto class."""
443
    result = set()  # To avoid duplicates we concatenate all model classes in a set.
444
445
    if is_torch_available():
        for attr_name in dir(transformers.models.auto.modeling_auto):
446
            if attr_name.startswith("MODEL_") and attr_name.endswith("MAPPING_NAMES"):
447
448
449
                result = result | set(get_values(getattr(transformers.models.auto.modeling_auto, attr_name)))
    if is_tf_available():
        for attr_name in dir(transformers.models.auto.modeling_tf_auto):
450
            if attr_name.startswith("TF_MODEL_") and attr_name.endswith("MAPPING_NAMES"):
451
452
453
                result = result | set(get_values(getattr(transformers.models.auto.modeling_tf_auto, attr_name)))
    if is_flax_available():
        for attr_name in dir(transformers.models.auto.modeling_flax_auto):
454
            if attr_name.startswith("FLAX_MODEL_") and attr_name.endswith("MAPPING_NAMES"):
455
                result = result | set(get_values(getattr(transformers.models.auto.modeling_flax_auto, attr_name)))
456
    return [cls for cls in result]
457
458


459
460
461
462
463
464
465
466
467
468
469
def ignore_unautoclassed(model_name):
    """Rules to determine if `name` should be in an auto class."""
    # Special white list
    if model_name in IGNORE_NON_AUTO_CONFIGURED:
        return True
    # Encoder and Decoder should be ignored
    if "Encoder" in model_name or "Decoder" in model_name:
        return True
    return False


470
def check_models_are_auto_configured(module, all_auto_models):
Patrick von Platen's avatar
Patrick von Platen committed
471
    """Check models defined in module are each in an auto class."""
472
473
474
    defined_models = get_models(module)
    failures = []
    for model_name, _ in defined_models:
475
        if model_name not in all_auto_models and not ignore_unautoclassed(model_name):
476
477
478
479
480
481
482
483
484
            failures.append(
                f"{model_name} is defined in {module.__name__} but is not present in any of the auto mapping. "
                "If that is intended behavior, add its name to `IGNORE_NON_AUTO_CONFIGURED` in the file "
                "`utils/check_repo.py`."
            )
    return failures


def check_all_models_are_auto_configured():
Patrick von Platen's avatar
Patrick von Platen committed
485
    """Check all models are each in an auto class."""
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
    missing_backends = []
    if not is_torch_available():
        missing_backends.append("PyTorch")
    if not is_tf_available():
        missing_backends.append("TensorFlow")
    if not is_flax_available():
        missing_backends.append("Flax")
    if len(missing_backends) > 0:
        missing = ", ".join(missing_backends)
        if os.getenv("TRANSFORMERS_IS_CI", "").upper() in ENV_VARS_TRUE_VALUES:
            raise Exception(
                "Full quality checks require all backends to be installed (with `pip install -e .[dev]` in the "
                f"Transformers repo, the following are missing: {missing}."
            )
        else:
            warnings.warn(
                "Full quality checks require all backends to be installed (with `pip install -e .[dev]` in the "
                f"Transformers repo, the following are missing: {missing}. While it's probably fine as long as you "
                "didn't make any change in one of those backends modeling files, you should probably execute the "
                "command above to be on the safe side."
            )
507
508
509
510
511
512
513
514
515
516
517
    modules = get_model_modules()
    all_auto_models = get_all_auto_configured_models()
    failures = []
    for module in modules:
        new_failures = check_models_are_auto_configured(module, all_auto_models)
        if new_failures is not None:
            failures += new_failures
    if len(failures) > 0:
        raise Exception(f"There were {len(failures)} failures:\n" + "\n".join(failures))


Sylvain Gugger's avatar
Sylvain Gugger committed
518
519
520
521
_re_decorator = re.compile(r"^\s*@(\S+)\s+$")


def check_decorator_order(filename):
Patrick von Platen's avatar
Patrick von Platen committed
522
    """Check that in the test file `filename` the slow decorator is always last."""
523
    with open(filename, "r", encoding="utf-8", newline="\n") as f:
Sylvain Gugger's avatar
Sylvain Gugger committed
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
        lines = f.readlines()
    decorator_before = None
    errors = []
    for i, line in enumerate(lines):
        search = _re_decorator.search(line)
        if search is not None:
            decorator_name = search.groups()[0]
            if decorator_before is not None and decorator_name.startswith("parameterized"):
                errors.append(i)
            decorator_before = decorator_name
        elif decorator_before is not None:
            decorator_before = None
    return errors


def check_all_decorator_order():
Patrick von Platen's avatar
Patrick von Platen committed
540
    """Check that in all test files, the slow decorator is always last."""
Sylvain Gugger's avatar
Sylvain Gugger committed
541
542
543
544
545
546
547
548
549
    errors = []
    for fname in os.listdir(PATH_TO_TESTS):
        if fname.endswith(".py"):
            filename = os.path.join(PATH_TO_TESTS, fname)
            new_errors = check_decorator_order(filename)
            errors += [f"- {filename}, line {i}" for i in new_errors]
    if len(errors) > 0:
        msg = "\n".join(errors)
        raise ValueError(
Sylvain Gugger's avatar
Sylvain Gugger committed
550
551
            "The parameterized decorator (and its variants) should always be first, but this is not the case in the"
            f" following files:\n{msg}"
Sylvain Gugger's avatar
Sylvain Gugger committed
552
553
554
        )


555
def find_all_documented_objects():
Patrick von Platen's avatar
Patrick von Platen committed
556
    """Parse the content of all doc files to detect which classes and functions it documents"""
557
558
    documented_obj = []
    for doc_file in Path(PATH_TO_DOC).glob("**/*.rst"):
Julien Plu's avatar
Julien Plu committed
559
        with open(doc_file, "r", encoding="utf-8", newline="\n") as f:
560
561
562
            content = f.read()
        raw_doc_objs = re.findall(r"(?:autoclass|autofunction):: transformers.(\S+)\s+", content)
        documented_obj += [obj.split(".")[-1] for obj in raw_doc_objs]
Sylvain Gugger's avatar
Sylvain Gugger committed
563
564
565
566
567
    for doc_file in Path(PATH_TO_DOC).glob("**/*.mdx"):
        with open(doc_file, "r", encoding="utf-8", newline="\n") as f:
            content = f.read()
        raw_doc_objs = re.findall("\[\[autodoc\]\]\s+(\S+)\s+", content)
        documented_obj += [obj.split(".")[-1] for obj in raw_doc_objs]
568
569
570
571
572
573
    return documented_obj


# One good reason for not being documented is to be deprecated. Put in this list deprecated objects.
DEPRECATED_OBJECTS = [
    "AutoModelWithLMHead",
574
    "BartPretrainedModel",
575
576
    "DataCollator",
    "DataCollatorForSOP",
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
    "GlueDataset",
    "GlueDataTrainingArguments",
    "LineByLineTextDataset",
    "LineByLineWithRefDataset",
    "LineByLineWithSOPTextDataset",
    "PretrainedBartModel",
    "PretrainedFSMTModel",
    "SingleSentenceClassificationProcessor",
    "SquadDataTrainingArguments",
    "SquadDataset",
    "SquadExample",
    "SquadFeatures",
    "SquadV1Processor",
    "SquadV2Processor",
    "TFAutoModelWithLMHead",
592
    "TFBartPretrainedModel",
593
594
    "TextDataset",
    "TextDatasetForNextSentencePrediction",
595
    "Wav2Vec2ForMaskedLM",
596
    "Wav2Vec2Tokenizer",
597
598
599
600
601
602
603
604
605
606
    "glue_compute_metrics",
    "glue_convert_examples_to_features",
    "glue_output_modes",
    "glue_processors",
    "glue_tasks_num_labels",
    "squad_convert_examples_to_features",
    "xnli_compute_metrics",
    "xnli_output_modes",
    "xnli_processors",
    "xnli_tasks_num_labels",
607
608
    "TFTrainer",
    "TFTrainingArguments",
609
610
611
612
613
614
615
]

# Exceptionally, some objects should not be documented after all rules passed.
# ONLY PUT SOMETHING IN THIS LIST AS A LAST RESORT!
UNDOCUMENTED_OBJECTS = [
    "AddedToken",  # This is a tokenizers class.
    "BasicTokenizer",  # Internal, should never have been in the main init.
616
    "CharacterTokenizer",  # Internal, should never have been in the main init.
617
    "DPRPretrainedReader",  # Like an Encoder.
Sylvain Gugger's avatar
Sylvain Gugger committed
618
    "DummyObject",  # Just picked by mistake sometimes.
619
    "MecabTokenizer",  # Internal, should never have been in the main init.
620
621
622
623
624
625
626
627
628
629
630
    "ModelCard",  # Internal type.
    "SqueezeBertModule",  # Internal building block (should have been called SqueezeBertLayer)
    "TFDPRPretrainedReader",  # Like an Encoder.
    "TransfoXLCorpus",  # Internal type.
    "WordpieceTokenizer",  # Internal, should never have been in the main init.
    "absl",  # External module
    "add_end_docstrings",  # Internal, should never have been in the main init.
    "add_start_docstrings",  # Internal, should never have been in the main init.
    "convert_tf_weight_name_to_pt_weight_name",  # Internal used to convert model weights
    "logger",  # Internal logger
    "logging",  # External module
631
    "requires_backends",  # Internal function
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
]

# This list should be empty. Objects in it should get their own doc page.
SHOULD_HAVE_THEIR_OWN_PAGE = [
    # Benchmarks
    "PyTorchBenchmark",
    "PyTorchBenchmarkArguments",
    "TensorFlowBenchmark",
    "TensorFlowBenchmarkArguments",
]


def ignore_undocumented(name):
    """Rules to determine if `name` should be undocumented."""
    # NOT DOCUMENTED ON PURPOSE.
    # Constants uppercase are not documented.
    if name.isupper():
        return True
    # PreTrainedModels / Encoders / Decoders / Layers / Embeddings / Attention are not documented.
    if (
        name.endswith("PreTrainedModel")
        or name.endswith("Decoder")
        or name.endswith("Encoder")
        or name.endswith("Layer")
        or name.endswith("Embeddings")
        or name.endswith("Attention")
    ):
        return True
    # Submodules are not documented.
    if os.path.isdir(os.path.join(PATH_TO_TRANSFORMERS, name)) or os.path.isfile(
        os.path.join(PATH_TO_TRANSFORMERS, f"{name}.py")
    ):
        return True
    # All load functions are not documented.
    if name.startswith("load_tf") or name.startswith("load_pytorch"):
        return True
    # is_xxx_available functions are not documented.
    if name.startswith("is_") and name.endswith("_available"):
        return True
    # Deprecated objects are not documented.
    if name in DEPRECATED_OBJECTS or name in UNDOCUMENTED_OBJECTS:
        return True
    # MMBT model does not really work.
    if name.startswith("MMBT"):
        return True
    if name in SHOULD_HAVE_THEIR_OWN_PAGE:
        return True
    return False


def check_all_objects_are_documented():
Patrick von Platen's avatar
Patrick von Platen committed
683
    """Check all models are properly documented."""
684
    documented_objs = find_all_documented_objects()
685
686
687
    modules = transformers._modules
    objects = [c for c in dir(transformers) if c not in modules and not c.startswith("_")]
    undocumented_objs = [c for c in objects if c not in documented_objs and not ignore_undocumented(c)]
688
689
690
691
692
    if len(undocumented_objs) > 0:
        raise Exception(
            "The following objects are in the public init so should be documented:\n - "
            + "\n - ".join(undocumented_objs)
        )
693
    check_docstrings_are_in_md()
694
695
696
697
698
699
700
701
702
    check_model_type_doc_match()


def check_model_type_doc_match():
    """Check all doc pages have a corresponding model type."""
    model_doc_folder = Path(PATH_TO_DOC) / "model_doc"
    model_docs = [m.stem for m in model_doc_folder.glob("*.mdx")]

    model_types = list(transformers.models.auto.configuration_auto.MODEL_NAMES_MAPPING.keys())
703
    model_types = [MODEL_TYPE_TO_DOC_MAPPING[m] if m in MODEL_TYPE_TO_DOC_MAPPING else m for m in model_types]
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721

    errors = []
    for m in model_docs:
        if m not in model_types and m != "auto":
            close_matches = get_close_matches(m, model_types)
            error_message = f"{m} is not a proper model identifier."
            if len(close_matches) > 0:
                close_matches = "/".join(close_matches)
                error_message += f" Did you mean {close_matches}?"
            errors.append(error_message)

    if len(errors) > 0:
        raise ValueError(
            "Some model doc pages do not match any existing model type:\n"
            + "\n".join(errors)
            + "\nYou can add any missing model type to the `MODEL_NAMES_MAPPING` constant in "
            "models/auto/configuration_auto.py."
        )
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748


# Re pattern to catch :obj:`xx`, :class:`xx`, :func:`xx` or :meth:`xx`.
_re_rst_special_words = re.compile(r":(?:obj|func|class|meth):`([^`]+)`")
# Re pattern to catch things between double backquotes.
_re_double_backquotes = re.compile(r"(^|[^`])``([^`]+)``([^`]|$)")
# Re pattern to catch example introduction.
_re_rst_example = re.compile(r"^\s*Example.*::\s*$", flags=re.MULTILINE)


def is_rst_docstring(docstring):
    """
    Returns `True` if `docstring` is written in rst.
    """
    if _re_rst_special_words.search(docstring) is not None:
        return True
    if _re_double_backquotes.search(docstring) is not None:
        return True
    if _re_rst_example.search(docstring) is not None:
        return True
    return False


def check_docstrings_are_in_md():
    """Check all docstrings are in md"""
    files_with_rst = []
    for file in Path(PATH_TO_TRANSFORMERS).glob("**/*.py"):
749
        with open(file, encoding="utf-8") as f:
750
751
752
753
754
755
756
757
758
759
760
761
762
            code = f.read()
        docstrings = code.split('"""')

        for idx, docstring in enumerate(docstrings):
            if idx % 2 == 0 or not is_rst_docstring(docstring):
                continue
            files_with_rst.append(file)
            break

    if len(files_with_rst) > 0:
        raise ValueError(
            "The following files have docstrings written in rst:\n"
            + "\n".join([f"- {f}" for f in files_with_rst])
Kamal Raj's avatar
Kamal Raj committed
763
            + "\nTo fix this run `doc-builder convert path_to_py_file` after installing `doc-builder`\n"
764
765
            "(`pip install git+https://github.com/huggingface/doc-builder`)"
        )
766
767


768
def check_repo_quality():
Patrick von Platen's avatar
Patrick von Platen committed
769
    """Check all models are properly tested and documented."""
770
771
    print("Checking all models are included.")
    check_model_list()
772
773
    print("Checking all models are public.")
    check_models_are_in_init()
774
    print("Checking all models are properly tested.")
Sylvain Gugger's avatar
Sylvain Gugger committed
775
    check_all_decorator_order()
776
    check_all_models_are_tested()
777
    print("Checking all objects are properly documented.")
778
    check_all_objects_are_documented()
779
780
    print("Checking all models are in at least one auto class.")
    check_all_models_are_auto_configured()
781
782
783
784


if __name__ == "__main__":
    check_repo_quality()