check_repo.py 30.6 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",
Li-Huai (Allan) Lin's avatar
Li-Huai (Allan) Lin committed
39
    "RealmBertModel",
40
41
42
43
    "T5Stack",
    "TFDPRSpanPredictor",
]

44
45
# 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.
46
IGNORE_NON_TESTED = PRIVATE_MODELS.copy() + [
47
    # models to ignore for not tested
Younes Belkada's avatar
Younes Belkada committed
48
    "OPTDecoder",  # Building part of bigger (tested) model.
49
    "DecisionTransformerGPT2Model",  # Building part of bigger (tested) model.
NielsRogge's avatar
NielsRogge committed
50
    "SegformerDecodeHead",  # Building part of bigger (tested) model.
Gunjan Chhablani's avatar
Gunjan Chhablani committed
51
52
53
    "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
54
55
56
    "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
57
58
59
    "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
60
61
    "M2M100Encoder",  # Building part of bigger (tested) model.
    "M2M100Decoder",  # Building part of bigger (tested) model.
Suraj Patil's avatar
Suraj Patil committed
62
63
    "Speech2TextEncoder",  # Building part of bigger (tested) model.
    "Speech2TextDecoder",  # Building part of bigger (tested) model.
Patrick von Platen's avatar
Patrick von Platen committed
64
65
    "LEDEncoder",  # Building part of bigger (tested) model.
    "LEDDecoder",  # Building part of bigger (tested) model.
66
    "BartDecoderWrapper",  # Building part of bigger (tested) model.
67
    "BartEncoder",  # Building part of bigger (tested) model.
68
    "BertLMHeadModel",  # Needs to be setup as decoder.
69
    "BlenderbotSmallEncoder",  # Building part of bigger (tested) model.
70
    "BlenderbotSmallDecoderWrapper",  # Building part of bigger (tested) model.
71
    "BlenderbotEncoder",  # Building part of bigger (tested) model.
72
    "BlenderbotDecoderWrapper",  # Building part of bigger (tested) model.
73
    "MBartEncoder",  # Building part of bigger (tested) model.
74
    "MBartDecoderWrapper",  # Building part of bigger (tested) model.
75
76
77
78
    "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.
79
    "PegasusEncoder",  # Building part of bigger (tested) model.
80
    "PegasusDecoderWrapper",  # Building part of bigger (tested) model.
81
    "DPREncoder",  # Building part of bigger (tested) model.
82
    "ProphetNetDecoderWrapper",  # Building part of bigger (tested) model.
Li-Huai (Allan) Lin's avatar
Li-Huai (Allan) Lin committed
83
84
85
86
    "RealmBertModel",  # Building part of bigger (tested) model.
    "RealmReader",  # Not regular model.
    "RealmScorer",  # Not regular model.
    "RealmForOpenQA",  # Not regular model.
87
    "ReformerForMaskedLM",  # Needs to be setup as decoder.
88
    "Speech2Text2DecoderWrapper",  # Building part of bigger (tested) model.
Ratthachat (Jung)'s avatar
Ratthachat (Jung) committed
89
    "TFDPREncoder",  # Building part of bigger (tested) model.
90
91
    "TFElectraMainLayer",  # Building part of bigger (tested) model (should it be a TFPreTrainedModel ?)
    "TFRobertaForMultipleChoice",  # TODO: fix
92
    "TrOCRDecoderWrapper",  # Building part of bigger (tested) model.
abhishek thakur's avatar
abhishek thakur committed
93
    "SeparableConv1D",  # Building part of bigger (tested) model.
94
    "FlaxBartForCausalLM",  # Building part of bigger (tested) model.
95
    "FlaxBertForCausalLM",  # Building part of bigger (tested) model. Tested implicitly through FlaxRobertaForCausalLM.
Younes Belkada's avatar
Younes Belkada committed
96
    "OPTDecoderWrapper",
97
98
99
100
101
]

# 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
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
    "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",
117
118
]

119
120
# 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.
121
IGNORE_NON_AUTO_CONFIGURED = PRIVATE_MODELS.copy() + [
122
    # models to ignore for model xxx mapping
NielsRogge's avatar
NielsRogge committed
123
    "DPTForDepthEstimation",
124
    "DecisionTransformerGPT2Model",
NielsRogge's avatar
NielsRogge committed
125
    "GLPNForDepthEstimation",
NielsRogge's avatar
NielsRogge committed
126
127
128
129
    "ViltForQuestionAnswering",
    "ViltForImagesAndTextClassification",
    "ViltForImageAndTextRetrieval",
    "ViltForMaskedLM",
Suraj Patil's avatar
Suraj Patil committed
130
131
132
    "XGLMEncoder",
    "XGLMDecoder",
    "XGLMDecoderWrapper",
NielsRogge's avatar
NielsRogge committed
133
134
    "PerceiverForMultimodalAutoencoding",
    "PerceiverForOpticalFlow",
NielsRogge's avatar
NielsRogge committed
135
    "SegformerDecodeHead",
Kamal Raj's avatar
Kamal Raj committed
136
    "FlaxBeitForMaskedImageModeling",
Gunjan Chhablani's avatar
Gunjan Chhablani committed
137
138
139
    "PLBartEncoder",
    "PLBartDecoder",
    "PLBartDecoderWrapper",
NielsRogge's avatar
NielsRogge committed
140
    "BeitForMaskedImageModeling",
Suraj Patil's avatar
Suraj Patil committed
141
142
    "CLIPTextModel",
    "CLIPVisionModel",
Yih-Dar's avatar
Yih-Dar committed
143
144
    "TFCLIPTextModel",
    "TFCLIPVisionModel",
Suraj Patil's avatar
Suraj Patil committed
145
146
    "FlaxCLIPTextModel",
    "FlaxCLIPVisionModel",
147
    "FlaxWav2Vec2ForCTC",
NielsRogge's avatar
NielsRogge committed
148
    "DetrForSegmentation",
149
150
    "DPRReader",
    "FlaubertForQuestionAnswering",
151
152
153
154
    "FlavaImageCodebook",
    "FlavaTextModel",
    "FlavaImageModel",
    "FlavaMultimodalModel",
155
    "GPT2DoubleHeadsModel",
Ryokan RI's avatar
Ryokan RI committed
156
    "LukeForMaskedLM",
NielsRogge's avatar
NielsRogge committed
157
158
159
    "LukeForEntityClassification",
    "LukeForEntityPairClassification",
    "LukeForEntitySpanClassification",
160
161
162
163
    "OpenAIGPTDoubleHeadsModel",
    "RagModel",
    "RagSequenceForGeneration",
    "RagTokenForGeneration",
Li-Huai (Allan) Lin's avatar
Li-Huai (Allan) Lin committed
164
165
166
167
    "RealmEmbedder",
    "RealmForOpenQA",
    "RealmScorer",
    "RealmReader",
Ratthachat (Jung)'s avatar
Ratthachat (Jung) committed
168
    "TFDPRReader",
169
170
    "TFGPT2DoubleHeadsModel",
    "TFOpenAIGPTDoubleHeadsModel",
Ratthachat (Jung)'s avatar
Ratthachat (Jung) committed
171
172
173
    "TFRagModel",
    "TFRagSequenceForGeneration",
    "TFRagTokenForGeneration",
174
    "Wav2Vec2ForCTC",
Patrick von Platen's avatar
Patrick von Platen committed
175
    "HubertForCTC",
176
177
    "SEWForCTC",
    "SEWDForCTC",
178
179
    "XLMForQuestionAnswering",
    "XLNetForQuestionAnswering",
abhishek thakur's avatar
abhishek thakur committed
180
    "SeparableConv1D",
Gunjan Chhablani's avatar
Gunjan Chhablani committed
181
182
183
184
    "VisualBertForRegionToPhraseAlignment",
    "VisualBertForVisualReasoning",
    "VisualBertForQuestionAnswering",
    "VisualBertForMultipleChoice",
Will Rice's avatar
Will Rice committed
185
    "TFWav2Vec2ForCTC",
Will Rice's avatar
Will Rice committed
186
    "TFHubertForCTC",
187
    "MaskFormerForInstanceSegmentation",
188
189
]

190
191
192
193
194
195
# 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"),
196
        ("data2vec-vision", "data2vec"),
197
198
199
200
    ]
)


201
202
203
204
205
206
207
208
209
# 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()


210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
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)}."
        )


230
231
232
# 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
233
    """Get the model modules inside the transformers library."""
234
235
236
237
238
239
240
241
    _ignore_modules = [
        "modeling_auto",
        "modeling_encoder_decoder",
        "modeling_marian",
        "modeling_mmbt",
        "modeling_outputs",
        "modeling_retribert",
        "modeling_utils",
Sylvain Gugger's avatar
Sylvain Gugger committed
242
        "modeling_flax_auto",
243
        "modeling_flax_encoder_decoder",
Stas Bekman's avatar
Stas Bekman committed
244
        "modeling_flax_utils",
245
        "modeling_speech_encoder_decoder",
246
        "modeling_flax_speech_encoder_decoder",
247
        "modeling_flax_vision_encoder_decoder",
248
249
        "modeling_transfo_xl_utilities",
        "modeling_tf_auto",
250
        "modeling_tf_encoder_decoder",
251
252
253
254
        "modeling_tf_outputs",
        "modeling_tf_pytorch_utils",
        "modeling_tf_utils",
        "modeling_tf_transfo_xl_utilities",
255
        "modeling_tf_vision_encoder_decoder",
256
        "modeling_vision_encoder_decoder",
257
258
    ]
    modules = []
Sylvain Gugger's avatar
Sylvain Gugger committed
259
260
261
262
263
264
265
266
267
    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)
268
269
270
    return modules


271
def get_models(module, include_pretrained=False):
Patrick von Platen's avatar
Patrick von Platen committed
272
    """Get the objects in module that are models."""
273
    models = []
274
    model_classes = (transformers.PreTrainedModel, transformers.TFPreTrainedModel, transformers.FlaxPreTrainedModel)
275
    for attr_name in dir(module):
276
        if not include_pretrained and ("Pretrained" in attr_name or "PreTrained" in attr_name):
277
278
279
280
281
282
283
            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


284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
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)}.")


314
315
316
# 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
317
318
319
320
321
322
    """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.
    """

323
324
325
    _ignore_files = [
        "test_modeling_common",
        "test_modeling_encoder_decoder",
326
        "test_modeling_flax_encoder_decoder",
327
        "test_modeling_flax_speech_encoder_decoder",
328
329
        "test_modeling_marian",
        "test_modeling_tf_common",
330
        "test_modeling_tf_encoder_decoder",
331
332
    ]
    test_files = []
Yih-Dar's avatar
Yih-Dar committed
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
    # 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)

350
351
352
353
354
355
    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
356
    """Parse the content of test_file to detect what's in all_model_classes"""
Sylvain Gugger's avatar
Sylvain Gugger committed
357
    # 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
358
    with open(os.path.join(PATH_TO_TESTS, test_file), "r", encoding="utf-8", newline="\n") as f:
359
        content = f.read()
Sylvain Gugger's avatar
Sylvain Gugger committed
360
    all_models = re.findall(r"all_model_classes\s+=\s+\(\s*\(([^\)]*)\)", content)
361
362
    # 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
363
    if len(all_models) > 0:
364
        model_tested = []
Sylvain Gugger's avatar
Sylvain Gugger committed
365
366
367
368
369
        for entry in all_models:
            for line in entry.split(","):
                name = line.strip()
                if len(name) > 0:
                    model_tested.append(name)
370
371
372
373
        return model_tested


def check_models_are_tested(module, test_file):
Patrick von Platen's avatar
Patrick von Platen committed
374
    """Check models defined in module are tested in test_file."""
375
    # XxxPreTrainedModel are not tested
376
377
378
    defined_models = get_models(module)
    tested_models = find_tested_models(test_file)
    if tested_models is None:
379
        if test_file.replace(os.path.sep, "/") in TEST_FILES_WITH_NO_COMMON_TESTS:
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
            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
399
    """Check all models are properly tested."""
400
401
402
403
    modules = get_model_modules()
    test_files = get_model_test_files()
    failures = []
    for module in modules:
404
405
        test_file = [file for file in test_files if f"test_{module.__name__.split('.')[-1]}.py" in file]
        if len(test_file) == 0:
406
            failures.append(f"{module.__name__} does not have its corresponding test file {test_file}.")
407
408
409
410
        elif len(test_file) > 1:
            failures.append(f"{module.__name__} has several test files: {test_file}.")
        else:
            test_file = test_file[0]
411
412
413
            new_failures = check_models_are_tested(module, test_file)
            if new_failures is not None:
                failures += new_failures
414
415
416
417
    if len(failures) > 0:
        raise Exception(f"There were {len(failures)} failures:\n" + "\n".join(failures))


418
def get_all_auto_configured_models():
Patrick von Platen's avatar
Patrick von Platen committed
419
    """Return the list of all models in at least one auto class."""
420
    result = set()  # To avoid duplicates we concatenate all model classes in a set.
421
422
    if is_torch_available():
        for attr_name in dir(transformers.models.auto.modeling_auto):
423
            if attr_name.startswith("MODEL_") and attr_name.endswith("MAPPING_NAMES"):
424
425
426
                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):
427
            if attr_name.startswith("TF_MODEL_") and attr_name.endswith("MAPPING_NAMES"):
428
429
430
                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):
431
            if attr_name.startswith("FLAX_MODEL_") and attr_name.endswith("MAPPING_NAMES"):
432
                result = result | set(get_values(getattr(transformers.models.auto.modeling_flax_auto, attr_name)))
433
    return [cls for cls in result]
434
435


436
437
438
439
440
441
442
443
444
445
446
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


447
def check_models_are_auto_configured(module, all_auto_models):
Patrick von Platen's avatar
Patrick von Platen committed
448
    """Check models defined in module are each in an auto class."""
449
450
451
    defined_models = get_models(module)
    failures = []
    for model_name, _ in defined_models:
452
        if model_name not in all_auto_models and not ignore_unautoclassed(model_name):
453
454
455
456
457
458
459
460
461
            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
462
    """Check all models are each in an auto class."""
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
    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."
            )
484
485
486
487
488
489
490
491
492
493
494
    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
495
496
497
498
_re_decorator = re.compile(r"^\s*@(\S+)\s+$")


def check_decorator_order(filename):
Patrick von Platen's avatar
Patrick von Platen committed
499
    """Check that in the test file `filename` the slow decorator is always last."""
500
    with open(filename, "r", encoding="utf-8", newline="\n") as f:
Sylvain Gugger's avatar
Sylvain Gugger committed
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
        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
517
    """Check that in all test files, the slow decorator is always last."""
Sylvain Gugger's avatar
Sylvain Gugger committed
518
519
520
521
522
523
524
525
526
527
528
529
530
    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(
            f"The parameterized decorator (and its variants) should always be first, but this is not the case in the following files:\n{msg}"
        )


531
def find_all_documented_objects():
Patrick von Platen's avatar
Patrick von Platen committed
532
    """Parse the content of all doc files to detect which classes and functions it documents"""
533
534
    documented_obj = []
    for doc_file in Path(PATH_TO_DOC).glob("**/*.rst"):
Julien Plu's avatar
Julien Plu committed
535
        with open(doc_file, "r", encoding="utf-8", newline="\n") as f:
536
537
538
            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
539
540
541
542
543
    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]
544
545
546
547
548
549
    return documented_obj


# One good reason for not being documented is to be deprecated. Put in this list deprecated objects.
DEPRECATED_OBJECTS = [
    "AutoModelWithLMHead",
550
    "BartPretrainedModel",
551
552
    "DataCollator",
    "DataCollatorForSOP",
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
    "GlueDataset",
    "GlueDataTrainingArguments",
    "LineByLineTextDataset",
    "LineByLineWithRefDataset",
    "LineByLineWithSOPTextDataset",
    "PretrainedBartModel",
    "PretrainedFSMTModel",
    "SingleSentenceClassificationProcessor",
    "SquadDataTrainingArguments",
    "SquadDataset",
    "SquadExample",
    "SquadFeatures",
    "SquadV1Processor",
    "SquadV2Processor",
    "TFAutoModelWithLMHead",
568
    "TFBartPretrainedModel",
569
570
    "TextDataset",
    "TextDatasetForNextSentencePrediction",
571
    "Wav2Vec2ForMaskedLM",
572
    "Wav2Vec2Tokenizer",
573
574
575
576
577
578
579
580
581
582
    "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",
583
584
    "TFTrainer",
    "TFTrainingArguments",
585
586
587
588
589
590
591
]

# 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.
592
    "CharacterTokenizer",  # Internal, should never have been in the main init.
593
    "DPRPretrainedReader",  # Like an Encoder.
Sylvain Gugger's avatar
Sylvain Gugger committed
594
    "DummyObject",  # Just picked by mistake sometimes.
595
    "MecabTokenizer",  # Internal, should never have been in the main init.
596
597
598
599
600
601
602
603
604
605
606
607
    "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.
    "cached_path",  # Internal used for downloading models.
    "convert_tf_weight_name_to_pt_weight_name",  # Internal used to convert model weights
    "logger",  # Internal logger
    "logging",  # External module
608
    "requires_backends",  # Internal function
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
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
]

# 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
660
    """Check all models are properly documented."""
661
    documented_objs = find_all_documented_objects()
662
663
664
    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)]
665
666
667
668
669
    if len(undocumented_objs) > 0:
        raise Exception(
            "The following objects are in the public init so should be documented:\n - "
            + "\n - ".join(undocumented_objs)
        )
670
    check_docstrings_are_in_md()
671
672
673
674
675
676
677
678
679
    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())
680
    model_types = [MODEL_TYPE_TO_DOC_MAPPING[m] if m in MODEL_TYPE_TO_DOC_MAPPING else m for m in model_types]
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698

    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."
        )
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739


# 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"):
        with open(file, "r") as f:
            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
740
            + "\nTo fix this run `doc-builder convert path_to_py_file` after installing `doc-builder`\n"
741
742
            "(`pip install git+https://github.com/huggingface/doc-builder`)"
        )
743
744


745
def check_repo_quality():
Patrick von Platen's avatar
Patrick von Platen committed
746
    """Check all models are properly tested and documented."""
747
748
    print("Checking all models are included.")
    check_model_list()
749
750
    print("Checking all models are public.")
    check_models_are_in_init()
751
    print("Checking all models are properly tested.")
Sylvain Gugger's avatar
Sylvain Gugger committed
752
    check_all_decorator_order()
753
    check_all_models_are_tested()
754
    print("Checking all objects are properly documented.")
755
    check_all_objects_are_documented()
756
757
    print("Checking all models are in at least one auto class.")
    check_all_models_are_auto_configured()
758
759
760
761


if __name__ == "__main__":
    check_repo_quality()