check_repo.py 45.2 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
import inspect
import os
import re
19
import sys
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
Yih-Dar's avatar
Yih-Dar committed
27
from transformers.models.auto.configuration_auto import CONFIG_MAPPING_NAMES
28
29
30
31
from transformers.models.auto.feature_extraction_auto import FEATURE_EXTRACTOR_MAPPING_NAMES
from transformers.models.auto.image_processing_auto import IMAGE_PROCESSOR_MAPPING_NAMES
from transformers.models.auto.processing_auto import PROCESSOR_MAPPING_NAMES
from transformers.models.auto.tokenization_auto import TOKENIZER_MAPPING_NAMES
32
from transformers.utils import ENV_VARS_TRUE_VALUES, direct_transformers_import
33

34
35
36
37
38

# 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"
39
PATH_TO_DOC = "docs/source/en"
40

41
42
# Update this list with models that are supposed to be private.
PRIVATE_MODELS = [
Jongjyh's avatar
Jongjyh committed
43
    "AltRobertaModel",
44
    "DPRSpanPredictor",
Daniel Stancl's avatar
Daniel Stancl committed
45
    "LongT5Stack",
Li-Huai (Allan) Lin's avatar
Li-Huai (Allan) Lin committed
46
    "RealmBertModel",
47
    "T5Stack",
48
    "MT5Stack",
49
    "SwitchTransformersStack",
50
    "TFDPRSpanPredictor",
51
52
    "MaskFormerSwinModel",
    "MaskFormerSwinPreTrainedModel",
53
54
    "BridgeTowerTextModel",
    "BridgeTowerVisionModel",
55
56
]

57
58
# 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.
59
IGNORE_NON_TESTED = PRIVATE_MODELS.copy() + [
60
    # models to ignore for not tested
NielsRogge's avatar
NielsRogge committed
61
    "InstructBlipQFormerModel",  # Building part of bigger (tested) model.
62
63
    "NllbMoeDecoder",
    "NllbMoeEncoder",
Jason Phang's avatar
Jason Phang committed
64
    "LlamaDecoder",  # Building part of bigger (tested) model.
NielsRogge's avatar
NielsRogge committed
65
    "Blip2QFormerModel",  # Building part of bigger (tested) model.
NielsRogge's avatar
NielsRogge committed
66
67
    "DetaEncoder",  # Building part of bigger (tested) model.
    "DetaDecoder",  # Building part of bigger (tested) model.
68
    "ErnieMForInformationExtraction",
69
70
    "GraphormerEncoder",  # Building part of bigger (tested) model.
    "GraphormerDecoderHead",  # Building part of bigger (tested) model.
NielsRogge's avatar
NielsRogge committed
71
    "CLIPSegDecoder",  # Building part of bigger (tested) model.
72
73
    "TableTransformerEncoder",  # Building part of bigger (tested) model.
    "TableTransformerDecoder",  # Building part of bigger (tested) model.
74
75
    "TimeSeriesTransformerEncoder",  # Building part of bigger (tested) model.
    "TimeSeriesTransformerDecoder",  # Building part of bigger (tested) model.
76
77
    "InformerEncoder",  # Building part of bigger (tested) model.
    "InformerDecoder",  # Building part of bigger (tested) model.
78
79
    "AutoformerEncoder",  # Building part of bigger (tested) model.
    "AutoformerDecoder",  # Building part of bigger (tested) model.
80
81
    "JukeboxVQVAE",  # Building part of bigger (tested) model.
    "JukeboxPrior",  # Building part of bigger (tested) model.
NielsRogge's avatar
NielsRogge committed
82
83
    "DeformableDetrEncoder",  # Building part of bigger (tested) model.
    "DeformableDetrDecoder",  # Building part of bigger (tested) model.
Younes Belkada's avatar
Younes Belkada committed
84
    "OPTDecoder",  # Building part of bigger (tested) model.
85
86
    "FlaxWhisperDecoder",  # Building part of bigger (tested) model.
    "FlaxWhisperEncoder",  # Building part of bigger (tested) model.
87
88
    "WhisperDecoder",  # Building part of bigger (tested) model.
    "WhisperEncoder",  # Building part of bigger (tested) model.
89
    "DecisionTransformerGPT2Model",  # Building part of bigger (tested) model.
NielsRogge's avatar
NielsRogge committed
90
    "SegformerDecodeHead",  # Building part of bigger (tested) model.
Gunjan Chhablani's avatar
Gunjan Chhablani committed
91
92
93
    "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
94
95
96
    "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
97
98
99
    "DetrEncoder",  # Building part of bigger (tested) model.
    "DetrDecoder",  # Building part of bigger (tested) model.
    "DetrDecoderWrapper",  # Building part of bigger (tested) model.
100
101
    "ConditionalDetrEncoder",  # Building part of bigger (tested) model.
    "ConditionalDetrDecoder",  # Building part of bigger (tested) model.
Suraj Patil's avatar
Suraj Patil committed
102
103
    "M2M100Encoder",  # Building part of bigger (tested) model.
    "M2M100Decoder",  # Building part of bigger (tested) model.
Chan Woo Kim's avatar
Chan Woo Kim committed
104
    "MCTCTEncoder",  # Building part of bigger (tested) model.
wangpeng's avatar
wangpeng committed
105
    "MgpstrModel",  # Building part of bigger (tested) model.
Suraj Patil's avatar
Suraj Patil committed
106
107
    "Speech2TextEncoder",  # Building part of bigger (tested) model.
    "Speech2TextDecoder",  # Building part of bigger (tested) model.
Patrick von Platen's avatar
Patrick von Platen committed
108
109
    "LEDEncoder",  # Building part of bigger (tested) model.
    "LEDDecoder",  # Building part of bigger (tested) model.
110
    "BartDecoderWrapper",  # Building part of bigger (tested) model.
111
    "BartEncoder",  # Building part of bigger (tested) model.
112
    "BertLMHeadModel",  # Needs to be setup as decoder.
113
    "BlenderbotSmallEncoder",  # Building part of bigger (tested) model.
114
    "BlenderbotSmallDecoderWrapper",  # Building part of bigger (tested) model.
115
    "BlenderbotEncoder",  # Building part of bigger (tested) model.
116
    "BlenderbotDecoderWrapper",  # Building part of bigger (tested) model.
117
    "MBartEncoder",  # Building part of bigger (tested) model.
118
    "MBartDecoderWrapper",  # Building part of bigger (tested) model.
119
120
121
122
    "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.
Sanchit Gandhi's avatar
Sanchit Gandhi committed
123
    "MusicgenDecoder",  # Building part of bigger (tested) model.
StevenTang1998's avatar
StevenTang1998 committed
124
125
    "MvpDecoderWrapper",  # Building part of bigger (tested) model.
    "MvpEncoder",  # Building part of bigger (tested) model.
126
    "PegasusEncoder",  # Building part of bigger (tested) model.
127
    "PegasusDecoderWrapper",  # Building part of bigger (tested) model.
Jason Phang's avatar
Jason Phang committed
128
129
130
    "PegasusXEncoder",  # Building part of bigger (tested) model.
    "PegasusXDecoder",  # Building part of bigger (tested) model.
    "PegasusXDecoderWrapper",  # Building part of bigger (tested) model.
131
    "DPREncoder",  # Building part of bigger (tested) model.
132
    "ProphetNetDecoderWrapper",  # Building part of bigger (tested) model.
Li-Huai (Allan) Lin's avatar
Li-Huai (Allan) Lin committed
133
134
135
136
    "RealmBertModel",  # Building part of bigger (tested) model.
    "RealmReader",  # Not regular model.
    "RealmScorer",  # Not regular model.
    "RealmForOpenQA",  # Not regular model.
137
    "ReformerForMaskedLM",  # Needs to be setup as decoder.
138
    "Speech2Text2DecoderWrapper",  # Building part of bigger (tested) model.
Ratthachat (Jung)'s avatar
Ratthachat (Jung) committed
139
    "TFDPREncoder",  # Building part of bigger (tested) model.
140
141
    "TFElectraMainLayer",  # Building part of bigger (tested) model (should it be a TFPreTrainedModel ?)
    "TFRobertaForMultipleChoice",  # TODO: fix
142
    "TFRobertaPreLayerNormForMultipleChoice",  # TODO: fix
143
    "TrOCRDecoderWrapper",  # Building part of bigger (tested) model.
amyeroberts's avatar
amyeroberts committed
144
145
    "TFWhisperEncoder",  # Building part of bigger (tested) model.
    "TFWhisperDecoder",  # Building part of bigger (tested) model.
abhishek thakur's avatar
abhishek thakur committed
146
    "SeparableConv1D",  # Building part of bigger (tested) model.
147
    "FlaxBartForCausalLM",  # Building part of bigger (tested) model.
148
    "FlaxBertForCausalLM",  # Building part of bigger (tested) model. Tested implicitly through FlaxRobertaForCausalLM.
Younes Belkada's avatar
Younes Belkada committed
149
    "OPTDecoderWrapper",
150
    "TFSegformerDecodeHead",  # Not a regular model.
Jongjyh's avatar
Jongjyh committed
151
    "AltRobertaModel",  # Building part of bigger (tested) model.
Younes Belkada's avatar
Younes Belkada committed
152
    "BlipTextLMHeadModel",  # No need to test it as it is tested by BlipTextVision models
Matt's avatar
Matt committed
153
    "TFBlipTextLMHeadModel",  # No need to test it as it is tested by BlipTextVision models
154
155
    "BridgeTowerTextModel",  # No need to test it as it is tested by BridgeTowerModel model.
    "BridgeTowerVisionModel",  # No need to test it as it is tested by BridgeTowerModel model.
156
157
158
159
160
161
162
163
164
165
166
167
    "SpeechT5Decoder",  # Building part of bigger (tested) model.
    "SpeechT5DecoderWithoutPrenet",  # Building part of bigger (tested) model.
    "SpeechT5DecoderWithSpeechPrenet",  # Building part of bigger (tested) model.
    "SpeechT5DecoderWithTextPrenet",  # Building part of bigger (tested) model.
    "SpeechT5Encoder",  # Building part of bigger (tested) model.
    "SpeechT5EncoderWithoutPrenet",  # Building part of bigger (tested) model.
    "SpeechT5EncoderWithSpeechPrenet",  # Building part of bigger (tested) model.
    "SpeechT5EncoderWithTextPrenet",  # Building part of bigger (tested) model.
    "SpeechT5SpeechDecoder",  # Building part of bigger (tested) model.
    "SpeechT5SpeechEncoder",  # Building part of bigger (tested) model.
    "SpeechT5TextDecoder",  # Building part of bigger (tested) model.
    "SpeechT5TextEncoder",  # Building part of bigger (tested) model.
168
169
170
171
172
]

# 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
173
174
175
176
177
178
179
180
181
182
183
184
185
    "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",
Matt's avatar
Matt committed
186
    "models/vision_text_dual_encoder/test_modeling_tf_vision_text_dual_encoder.py",
Yih-Dar's avatar
Yih-Dar committed
187
188
    "models/vision_text_dual_encoder/test_modeling_flax_vision_text_dual_encoder.py",
    "models/decision_transformer/test_modeling_decision_transformer.py",
189
190
]

191
192
# 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.
193
IGNORE_NON_AUTO_CONFIGURED = PRIVATE_MODELS.copy() + [
194
    # models to ignore for model xxx mapping
195
196
    "AlignTextModel",
    "AlignVisionModel",
197
198
199
200
    "ClapTextModel",
    "ClapTextModelWithProjection",
    "ClapAudioModel",
    "ClapAudioModelWithProjection",
NielsRogge's avatar
NielsRogge committed
201
202
203
    "Blip2ForConditionalGeneration",
    "Blip2QFormerModel",
    "Blip2VisionModel",
204
    "ErnieMForInformationExtraction",
205
    "GitVisionModel",
206
207
    "GraphormerModel",
    "GraphormerForGraphClassification",
Younes Belkada's avatar
Younes Belkada committed
208
209
210
211
212
213
    "BlipForConditionalGeneration",
    "BlipForImageTextRetrieval",
    "BlipForQuestionAnswering",
    "BlipVisionModel",
    "BlipTextLMHeadModel",
    "BlipTextModel",
Matt's avatar
Matt committed
214
215
216
217
218
219
    "TFBlipForConditionalGeneration",
    "TFBlipForImageTextRetrieval",
    "TFBlipForQuestionAnswering",
    "TFBlipVisionModel",
    "TFBlipTextLMHeadModel",
    "TFBlipTextModel",
NielsRogge's avatar
NielsRogge committed
220
    "Swin2SRForImageSuperResolution",
221
222
    "BridgeTowerForImageAndTextRetrieval",
    "BridgeTowerForMaskedLM",
223
    "BridgeTowerForContrastiveLearning",
NielsRogge's avatar
NielsRogge committed
224
225
226
    "CLIPSegForImageSegmentation",
    "CLIPSegVisionModel",
    "CLIPSegTextModel",
Matt's avatar
Matt committed
227
    "EsmForProteinFolding",
228
    "GPTSanJapaneseModel",
229
    "TimeSeriesTransformerForPrediction",
230
    "InformerForPrediction",
231
    "AutoformerForPrediction",
232
233
    "JukeboxVQVAE",
    "JukeboxPrior",
Jason Phang's avatar
Jason Phang committed
234
235
236
237
238
239
    "PegasusXEncoder",
    "PegasusXDecoder",
    "PegasusXDecoderWrapper",
    "PegasusXEncoder",
    "PegasusXDecoder",
    "PegasusXDecoderWrapper",
240
    "SamModel",
NielsRogge's avatar
NielsRogge committed
241
    "DPTForDepthEstimation",
242
    "DecisionTransformerGPT2Model",
NielsRogge's avatar
NielsRogge committed
243
    "GLPNForDepthEstimation",
NielsRogge's avatar
NielsRogge committed
244
245
    "ViltForImagesAndTextClassification",
    "ViltForImageAndTextRetrieval",
246
    "ViltForTokenClassification",
NielsRogge's avatar
NielsRogge committed
247
    "ViltForMaskedLM",
Suraj Patil's avatar
Suraj Patil committed
248
249
250
    "XGLMEncoder",
    "XGLMDecoder",
    "XGLMDecoderWrapper",
NielsRogge's avatar
NielsRogge committed
251
252
    "PerceiverForMultimodalAutoencoding",
    "PerceiverForOpticalFlow",
NielsRogge's avatar
NielsRogge committed
253
    "SegformerDecodeHead",
254
    "TFSegformerDecodeHead",
Kamal Raj's avatar
Kamal Raj committed
255
    "FlaxBeitForMaskedImageModeling",
Gunjan Chhablani's avatar
Gunjan Chhablani committed
256
257
258
    "PLBartEncoder",
    "PLBartDecoder",
    "PLBartDecoderWrapper",
NielsRogge's avatar
NielsRogge committed
259
    "BeitForMaskedImageModeling",
260
261
    "ChineseCLIPTextModel",
    "ChineseCLIPVisionModel",
Suraj Patil's avatar
Suraj Patil committed
262
    "CLIPTextModel",
263
    "CLIPTextModelWithProjection",
Suraj Patil's avatar
Suraj Patil committed
264
    "CLIPVisionModel",
265
    "CLIPVisionModelWithProjection",
266
267
    "GroupViTTextModel",
    "GroupViTVisionModel",
Yih-Dar's avatar
Yih-Dar committed
268
269
    "TFCLIPTextModel",
    "TFCLIPVisionModel",
270
271
    "TFGroupViTTextModel",
    "TFGroupViTVisionModel",
Suraj Patil's avatar
Suraj Patil committed
272
273
    "FlaxCLIPTextModel",
    "FlaxCLIPVisionModel",
274
    "FlaxWav2Vec2ForCTC",
NielsRogge's avatar
NielsRogge committed
275
    "DetrForSegmentation",
Younes Belkada's avatar
Younes Belkada committed
276
277
278
    "Pix2StructVisionModel",
    "Pix2StructTextModel",
    "Pix2StructForConditionalGeneration",
279
    "ConditionalDetrForSegmentation",
280
281
    "DPRReader",
    "FlaubertForQuestionAnswering",
282
283
284
285
    "FlavaImageCodebook",
    "FlavaTextModel",
    "FlavaImageModel",
    "FlavaMultimodalModel",
286
    "GPT2DoubleHeadsModel",
287
    "GPTSw3DoubleHeadsModel",
NielsRogge's avatar
NielsRogge committed
288
289
    "InstructBlipVisionModel",
    "InstructBlipQFormerModel",
290
    "LayoutLMForQuestionAnswering",
Ryokan RI's avatar
Ryokan RI committed
291
    "LukeForMaskedLM",
NielsRogge's avatar
NielsRogge committed
292
293
294
    "LukeForEntityClassification",
    "LukeForEntityPairClassification",
    "LukeForEntitySpanClassification",
wangpeng's avatar
wangpeng committed
295
    "MgpstrModel",
296
    "OpenAIGPTDoubleHeadsModel",
297
298
299
    "OwlViTTextModel",
    "OwlViTVisionModel",
    "OwlViTForObjectDetection",
300
301
302
    "RagModel",
    "RagSequenceForGeneration",
    "RagTokenForGeneration",
Li-Huai (Allan) Lin's avatar
Li-Huai (Allan) Lin committed
303
304
305
306
    "RealmEmbedder",
    "RealmForOpenQA",
    "RealmScorer",
    "RealmReader",
Ratthachat (Jung)'s avatar
Ratthachat (Jung) committed
307
    "TFDPRReader",
308
    "TFGPT2DoubleHeadsModel",
309
    "TFLayoutLMForQuestionAnswering",
310
    "TFOpenAIGPTDoubleHeadsModel",
Ratthachat (Jung)'s avatar
Ratthachat (Jung) committed
311
312
313
    "TFRagModel",
    "TFRagSequenceForGeneration",
    "TFRagTokenForGeneration",
314
    "Wav2Vec2ForCTC",
Patrick von Platen's avatar
Patrick von Platen committed
315
    "HubertForCTC",
316
317
    "SEWForCTC",
    "SEWDForCTC",
318
319
    "XLMForQuestionAnswering",
    "XLNetForQuestionAnswering",
abhishek thakur's avatar
abhishek thakur committed
320
    "SeparableConv1D",
Gunjan Chhablani's avatar
Gunjan Chhablani committed
321
322
323
324
    "VisualBertForRegionToPhraseAlignment",
    "VisualBertForVisualReasoning",
    "VisualBertForQuestionAnswering",
    "VisualBertForMultipleChoice",
Will Rice's avatar
Will Rice committed
325
    "TFWav2Vec2ForCTC",
Will Rice's avatar
Will Rice committed
326
    "TFHubertForCTC",
NielsRogge's avatar
NielsRogge committed
327
328
    "XCLIPVisionModel",
    "XCLIPTextModel",
Jongjyh's avatar
Jongjyh committed
329
330
331
    "AltCLIPTextModel",
    "AltCLIPVisionModel",
    "AltRobertaModel",
Zineng Tang's avatar
Zineng Tang committed
332
    "TvltForAudioVisualClassification",
333
334
335
    "SpeechT5ForSpeechToSpeech",
    "SpeechT5ForTextToSpeech",
    "SpeechT5HifiGan",
Sanchit Gandhi's avatar
Sanchit Gandhi committed
336
337
    "MusicgenModel",
    "MusicgenForConditionalGeneration",
338
339
]

340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
# DO NOT edit this list!
# (The corresponding pytorch objects should never be in the main `__init__`, but it's too late to remove)
OBJECT_TO_SKIP_IN_MAIN_INIT_CHECK = [
    "FlaxBertLayer",
    "FlaxBigBirdLayer",
    "FlaxRoFormerLayer",
    "TFBertLayer",
    "TFLxmertEncoder",
    "TFLxmertXLayer",
    "TFMPNetLayer",
    "TFMobileBertLayer",
    "TFSegformerLayer",
    "TFViTMAELayer",
]

355
356
357
358
359
360
# 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"),
361
        ("data2vec-vision", "data2vec"),
NielsRogge's avatar
NielsRogge committed
362
        ("donut-swin", "donut"),
363
364
365
366
    ]
)


367
# This is to make sure the transformers module imported is the one in the repo.
368
transformers = direct_transformers_import(PATH_TO_TRANSFORMERS)
369
370


371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
def check_missing_backends():
    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 repo consistency 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 repo consistency 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."
            )


395
396
397
398
399
400
401
402
403
404
405
406
407
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("__")]

408
    missing_models = sorted(set(_models).difference(models))
409
410
411
412
413
414
    if missing_models:
        raise Exception(
            f"The following models should be included in {models_dir}/__init__.py: {','.join(missing_models)}."
        )


415
416
417
# 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
418
    """Get the model modules inside the transformers library."""
419
420
421
422
423
424
425
426
    _ignore_modules = [
        "modeling_auto",
        "modeling_encoder_decoder",
        "modeling_marian",
        "modeling_mmbt",
        "modeling_outputs",
        "modeling_retribert",
        "modeling_utils",
Sylvain Gugger's avatar
Sylvain Gugger committed
427
        "modeling_flax_auto",
428
        "modeling_flax_encoder_decoder",
Stas Bekman's avatar
Stas Bekman committed
429
        "modeling_flax_utils",
430
        "modeling_speech_encoder_decoder",
431
        "modeling_flax_speech_encoder_decoder",
432
        "modeling_flax_vision_encoder_decoder",
amyeroberts's avatar
amyeroberts committed
433
        "modeling_timm_backbone",
434
435
        "modeling_transfo_xl_utilities",
        "modeling_tf_auto",
436
        "modeling_tf_encoder_decoder",
437
438
439
440
        "modeling_tf_outputs",
        "modeling_tf_pytorch_utils",
        "modeling_tf_utils",
        "modeling_tf_transfo_xl_utilities",
441
        "modeling_tf_vision_encoder_decoder",
442
        "modeling_vision_encoder_decoder",
443
444
    ]
    modules = []
Sylvain Gugger's avatar
Sylvain Gugger committed
445
446
447
448
449
450
451
452
453
    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)
454
455
456
    return modules


457
def get_models(module, include_pretrained=False):
Patrick von Platen's avatar
Patrick von Platen committed
458
    """Get the objects in module that are models."""
459
    models = []
460
    model_classes = (transformers.PreTrainedModel, transformers.TFPreTrainedModel, transformers.FlaxPreTrainedModel)
461
    for attr_name in dir(module):
462
        if not include_pretrained and ("Pretrained" in attr_name or "PreTrained" in attr_name):
463
464
465
466
467
468
469
            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


470
471
472
473
474
475
476
477
478
479
480
481
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
482
483
    if model.endswith("Prenet"):
        return True
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
    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)}.")


502
503
504
# 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
505
506
507
508
509
510
    """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.
    """

511
512
513
    _ignore_files = [
        "test_modeling_common",
        "test_modeling_encoder_decoder",
514
        "test_modeling_flax_encoder_decoder",
515
        "test_modeling_flax_speech_encoder_decoder",
516
517
        "test_modeling_marian",
        "test_modeling_tf_common",
518
        "test_modeling_tf_encoder_decoder",
519
520
    ]
    test_files = []
Yih-Dar's avatar
Yih-Dar committed
521
522
523
524
525
526
527
528
529
530
531
532
533
    # 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]
534
                if "test_modeling" in filename and os.path.splitext(filename)[0] not in _ignore_files:
Yih-Dar's avatar
Yih-Dar committed
535
536
537
                    file = os.path.join(*path.split(os.sep)[1:])
                    test_files.append(file)

538
539
540
541
542
543
    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
544
    """Parse the content of test_file to detect what's in all_model_classes"""
Sylvain Gugger's avatar
Sylvain Gugger committed
545
    # 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
546
    with open(os.path.join(PATH_TO_TESTS, test_file), "r", encoding="utf-8", newline="\n") as f:
547
        content = f.read()
Sylvain Gugger's avatar
Sylvain Gugger committed
548
    all_models = re.findall(r"all_model_classes\s+=\s+\(\s*\(([^\)]*)\)", content)
549
550
    # 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
551
    if len(all_models) > 0:
552
        model_tested = []
Sylvain Gugger's avatar
Sylvain Gugger committed
553
554
555
556
557
        for entry in all_models:
            for line in entry.split(","):
                name = line.strip()
                if len(name) > 0:
                    model_tested.append(name)
558
559
560
561
        return model_tested


def check_models_are_tested(module, test_file):
Patrick von Platen's avatar
Patrick von Platen committed
562
    """Check models defined in module are tested in test_file."""
563
    # XxxPreTrainedModel are not tested
564
565
566
    defined_models = get_models(module)
    tested_models = find_tested_models(test_file)
    if tested_models is None:
567
        if test_file.replace(os.path.sep, "/") in TEST_FILES_WITH_NO_COMMON_TESTS:
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
            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
587
    """Check all models are properly tested."""
588
589
590
591
    modules = get_model_modules()
    test_files = get_model_test_files()
    failures = []
    for module in modules:
592
593
        test_file = [file for file in test_files if f"test_{module.__name__.split('.')[-1]}.py" in file]
        if len(test_file) == 0:
594
            failures.append(f"{module.__name__} does not have its corresponding test file {test_file}.")
595
596
597
598
        elif len(test_file) > 1:
            failures.append(f"{module.__name__} has several test files: {test_file}.")
        else:
            test_file = test_file[0]
599
600
601
            new_failures = check_models_are_tested(module, test_file)
            if new_failures is not None:
                failures += new_failures
602
603
604
605
    if len(failures) > 0:
        raise Exception(f"There were {len(failures)} failures:\n" + "\n".join(failures))


606
def get_all_auto_configured_models():
Patrick von Platen's avatar
Patrick von Platen committed
607
    """Return the list of all models in at least one auto class."""
608
    result = set()  # To avoid duplicates we concatenate all model classes in a set.
609
610
    if is_torch_available():
        for attr_name in dir(transformers.models.auto.modeling_auto):
611
            if attr_name.startswith("MODEL_") and attr_name.endswith("MAPPING_NAMES"):
612
613
614
                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):
615
            if attr_name.startswith("TF_MODEL_") and attr_name.endswith("MAPPING_NAMES"):
616
617
618
                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):
619
            if attr_name.startswith("FLAX_MODEL_") and attr_name.endswith("MAPPING_NAMES"):
620
                result = result | set(get_values(getattr(transformers.models.auto.modeling_flax_auto, attr_name)))
621
    return list(result)
622
623


624
625
626
627
628
629
630
631
632
633
634
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


635
def check_models_are_auto_configured(module, all_auto_models):
Patrick von Platen's avatar
Patrick von Platen committed
636
    """Check models defined in module are each in an auto class."""
637
638
639
    defined_models = get_models(module)
    failures = []
    for model_name, _ in defined_models:
640
        if model_name not in all_auto_models and not ignore_unautoclassed(model_name):
641
642
643
644
645
646
647
648
649
            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
650
    """Check all models are each in an auto class."""
651
    check_missing_backends()
652
653
654
655
656
657
658
659
660
661
662
    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))


663
664
def check_all_auto_object_names_being_defined():
    """Check all names defined in auto (name) mappings exist in the library."""
665
    check_missing_backends()
666

667
    failures = []
668
    mappings_to_check = {
669
670
671
672
673
674
        "TOKENIZER_MAPPING_NAMES": TOKENIZER_MAPPING_NAMES,
        "IMAGE_PROCESSOR_MAPPING_NAMES": IMAGE_PROCESSOR_MAPPING_NAMES,
        "FEATURE_EXTRACTOR_MAPPING_NAMES": FEATURE_EXTRACTOR_MAPPING_NAMES,
        "PROCESSOR_MAPPING_NAMES": PROCESSOR_MAPPING_NAMES,
    }

675
676
    # Each auto modeling files contains multiple mappings. Let's get them in a dynamic way.
    for module_name in ["modeling_auto", "modeling_tf_auto", "modeling_flax_auto"]:
677
678
679
        module = getattr(transformers.models.auto, module_name, None)
        if module is None:
            continue
680
681
682
683
684
        # all mappings in a single auto modeling file
        mapping_names = [x for x in dir(module) if x.endswith("_MAPPING_NAMES")]
        mappings_to_check.update({name: getattr(module, name) for name in mapping_names})

    for name, mapping in mappings_to_check.items():
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
        for model_type, class_names in mapping.items():
            if not isinstance(class_names, tuple):
                class_names = (class_names,)
                for class_name in class_names:
                    if class_name is None:
                        continue
                    # dummy object is accepted
                    if not hasattr(transformers, class_name):
                        # If the class name is in a model name mapping, let's not check if there is a definition in any modeling
                        # module, if it's a private model defined in this file.
                        if name.endswith("MODEL_MAPPING_NAMES") and is_a_private_model(class_name):
                            continue
                        failures.append(
                            f"`{class_name}` appears in the mapping `{name}` but it is not defined in the library."
                        )
    if len(failures) > 0:
        raise Exception(f"There were {len(failures)} failures:\n" + "\n".join(failures))


Yih-Dar's avatar
Yih-Dar committed
704
705
def check_all_auto_mapping_names_in_config_mapping_names():
    """Check all keys defined in auto mappings (mappings of names) appear in `CONFIG_MAPPING_NAMES`."""
706
    check_missing_backends()
Yih-Dar's avatar
Yih-Dar committed
707

708
    failures = []
Yih-Dar's avatar
Yih-Dar committed
709
    # `TOKENIZER_PROCESSOR_MAPPING_NAMES` and `AutoTokenizer` is special, and don't need to follow the rule.
710
    mappings_to_check = {
Yih-Dar's avatar
Yih-Dar committed
711
712
713
714
715
        "IMAGE_PROCESSOR_MAPPING_NAMES": IMAGE_PROCESSOR_MAPPING_NAMES,
        "FEATURE_EXTRACTOR_MAPPING_NAMES": FEATURE_EXTRACTOR_MAPPING_NAMES,
        "PROCESSOR_MAPPING_NAMES": PROCESSOR_MAPPING_NAMES,
    }

716
717
    # Each auto modeling files contains multiple mappings. Let's get them in a dynamic way.
    for module_name in ["modeling_auto", "modeling_tf_auto", "modeling_flax_auto"]:
718
719
720
        module = getattr(transformers.models.auto, module_name, None)
        if module is None:
            continue
721
722
723
724
725
        # all mappings in a single auto modeling file
        mapping_names = [x for x in dir(module) if x.endswith("_MAPPING_NAMES")]
        mappings_to_check.update({name: getattr(module, name) for name in mapping_names})

    for name, mapping in mappings_to_check.items():
Yih-Dar's avatar
Yih-Dar committed
726
727
728
729
730
731
732
733
734
735
        for model_type, class_names in mapping.items():
            if model_type not in CONFIG_MAPPING_NAMES:
                failures.append(
                    f"`{model_type}` appears in the mapping `{name}` but it is not defined in the keys of "
                    "`CONFIG_MAPPING_NAMES`."
                )
    if len(failures) > 0:
        raise Exception(f"There were {len(failures)} failures:\n" + "\n".join(failures))


736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
def check_all_auto_mappings_importable():
    """Check all auto mappings could be imported."""
    check_missing_backends()

    failures = []
    mappings_to_check = {}
    # Each auto modeling files contains multiple mappings. Let's get them in a dynamic way.
    for module_name in ["modeling_auto", "modeling_tf_auto", "modeling_flax_auto"]:
        module = getattr(transformers.models.auto, module_name, None)
        if module is None:
            continue
        # all mappings in a single auto modeling file
        mapping_names = [x for x in dir(module) if x.endswith("_MAPPING_NAMES")]
        mappings_to_check.update({name: getattr(module, name) for name in mapping_names})

    for name, _ in mappings_to_check.items():
        name = name.replace("_MAPPING_NAMES", "_MAPPING")
        if not hasattr(transformers, name):
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
            failures.append(f"`{name}`")
    if len(failures) > 0:
        raise Exception(f"There were {len(failures)} failures:\n" + "\n".join(failures))


def check_objects_being_equally_in_main_init():
    """Check if an object is in the main __init__ if its counterpart in PyTorch is."""
    attrs = dir(transformers)

    failures = []
    for attr in attrs:
        obj = getattr(transformers, attr)
        if hasattr(obj, "__module__"):
            module_path = obj.__module__
            module_name = module_path.split(".")[-1]
            module_dir = ".".join(module_path.split(".")[:-1])
            if (
                module_name.startswith("modeling_")
                and not module_name.startswith("modeling_tf_")
                and not module_name.startswith("modeling_flax_")
            ):
                parent_module = sys.modules[module_dir]

                frameworks = []
                if is_tf_available():
                    frameworks.append("TF")
                if is_flax_available():
                    frameworks.append("Flax")

                for framework in frameworks:
                    other_module_path = module_path.replace("modeling_", f"modeling_{framework.lower()}_")
                    if os.path.isfile("src/" + other_module_path.replace(".", "/") + ".py"):
                        other_module_name = module_name.replace("modeling_", f"modeling_{framework.lower()}_")
                        other_module = getattr(parent_module, other_module_name)
                        if hasattr(other_module, f"{framework}{attr}"):
                            if not hasattr(transformers, f"{framework}{attr}"):
                                if f"{framework}{attr}" not in OBJECT_TO_SKIP_IN_MAIN_INIT_CHECK:
                                    failures.append(f"{framework}{attr}")
                        if hasattr(other_module, f"{framework}_{attr}"):
                            if not hasattr(transformers, f"{framework}_{attr}"):
                                if f"{framework}_{attr}" not in OBJECT_TO_SKIP_IN_MAIN_INIT_CHECK:
                                    failures.append(f"{framework}_{attr}")
796
797
798
799
    if len(failures) > 0:
        raise Exception(f"There were {len(failures)} failures:\n" + "\n".join(failures))


Sylvain Gugger's avatar
Sylvain Gugger committed
800
801
802
803
_re_decorator = re.compile(r"^\s*@(\S+)\s+$")


def check_decorator_order(filename):
Patrick von Platen's avatar
Patrick von Platen committed
804
    """Check that in the test file `filename` the slow decorator is always last."""
805
    with open(filename, "r", encoding="utf-8", newline="\n") as f:
Sylvain Gugger's avatar
Sylvain Gugger committed
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
        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
822
    """Check that in all test files, the slow decorator is always last."""
Sylvain Gugger's avatar
Sylvain Gugger committed
823
824
825
826
827
828
829
830
831
    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
832
833
            "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
834
835
836
        )


837
def find_all_documented_objects():
Patrick von Platen's avatar
Patrick von Platen committed
838
    """Parse the content of all doc files to detect which classes and functions it documents"""
839
840
    documented_obj = []
    for doc_file in Path(PATH_TO_DOC).glob("**/*.rst"):
Julien Plu's avatar
Julien Plu committed
841
        with open(doc_file, "r", encoding="utf-8", newline="\n") as f:
842
843
844
            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]
845
    for doc_file in Path(PATH_TO_DOC).glob("**/*.md"):
Sylvain Gugger's avatar
Sylvain Gugger committed
846
847
        with open(doc_file, "r", encoding="utf-8", newline="\n") as f:
            content = f.read()
848
        raw_doc_objs = re.findall(r"\[\[autodoc\]\]\s+(\S+)\s+", content)
Sylvain Gugger's avatar
Sylvain Gugger committed
849
        documented_obj += [obj.split(".")[-1] for obj in raw_doc_objs]
850
851
852
853
854
855
    return documented_obj


# One good reason for not being documented is to be deprecated. Put in this list deprecated objects.
DEPRECATED_OBJECTS = [
    "AutoModelWithLMHead",
856
    "BartPretrainedModel",
857
858
    "DataCollator",
    "DataCollatorForSOP",
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
    "GlueDataset",
    "GlueDataTrainingArguments",
    "LineByLineTextDataset",
    "LineByLineWithRefDataset",
    "LineByLineWithSOPTextDataset",
    "PretrainedBartModel",
    "PretrainedFSMTModel",
    "SingleSentenceClassificationProcessor",
    "SquadDataTrainingArguments",
    "SquadDataset",
    "SquadExample",
    "SquadFeatures",
    "SquadV1Processor",
    "SquadV2Processor",
    "TFAutoModelWithLMHead",
874
    "TFBartPretrainedModel",
875
876
    "TextDataset",
    "TextDatasetForNextSentencePrediction",
877
    "Wav2Vec2ForMaskedLM",
878
    "Wav2Vec2Tokenizer",
879
880
881
882
883
884
885
886
887
888
    "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",
889
890
    "TFTrainer",
    "TFTrainingArguments",
891
892
893
894
895
896
897
]

# 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.
898
    "CharacterTokenizer",  # Internal, should never have been in the main init.
899
    "DPRPretrainedReader",  # Like an Encoder.
Sylvain Gugger's avatar
Sylvain Gugger committed
900
    "DummyObject",  # Just picked by mistake sometimes.
901
    "MecabTokenizer",  # Internal, should never have been in the main init.
902
903
904
905
906
907
908
909
910
911
912
    "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
913
    "requires_backends",  # Internal function
Jongjyh's avatar
Jongjyh committed
914
    "AltRobertaModel",  # Internal module
915
916
917
918
919
920
921
922
923
]

# This list should be empty. Objects in it should get their own doc page.
SHOULD_HAVE_THEIR_OWN_PAGE = [
    # Benchmarks
    "PyTorchBenchmark",
    "PyTorchBenchmarkArguments",
    "TensorFlowBenchmark",
    "TensorFlowBenchmarkArguments",
924
    "AutoBackbone",
NielsRogge's avatar
NielsRogge committed
925
926
    "BitBackbone",
    "ConvNextBackbone",
Alara Dirik's avatar
Alara Dirik committed
927
    "ConvNextV2Backbone",
928
    "DinatBackbone",
Alara Dirik's avatar
Alara Dirik committed
929
    "FocalNetBackbone",
NielsRogge's avatar
NielsRogge committed
930
    "MaskFormerSwinBackbone",
931
932
    "MaskFormerSwinConfig",
    "MaskFormerSwinModel",
NielsRogge's avatar
NielsRogge committed
933
934
    "NatBackbone",
    "ResNetBackbone",
NielsRogge's avatar
NielsRogge committed
935
    "SwinBackbone",
amyeroberts's avatar
amyeroberts committed
936
937
    "TimmBackbone",
    "TimmBackboneConfig",
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
]


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
980
    """Check all models are properly documented."""
981
    documented_objs = find_all_documented_objects()
982
983
984
    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)]
985
986
987
988
989
    if len(undocumented_objs) > 0:
        raise Exception(
            "The following objects are in the public init so should be documented:\n - "
            + "\n - ".join(undocumented_objs)
        )
990
    check_docstrings_are_in_md()
991
992
993
994
995
996
    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"
997
    model_docs = [m.stem for m in model_doc_folder.glob("*.md")]
998
999

    model_types = list(transformers.models.auto.configuration_auto.MODEL_NAMES_MAPPING.keys())
1000
    model_types = [MODEL_TYPE_TO_DOC_MAPPING[m] if m in MODEL_TYPE_TO_DOC_MAPPING else m for m in model_types]
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018

    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."
        )
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045


# 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"):
1046
        with open(file, encoding="utf-8") as f:
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
            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
1060
            + "\nTo fix this run `doc-builder convert path_to_py_file` after installing `doc-builder`\n"
1061
1062
            "(`pip install git+https://github.com/huggingface/doc-builder`)"
        )
1063
1064


1065
def check_repo_quality():
Patrick von Platen's avatar
Patrick von Platen committed
1066
    """Check all models are properly tested and documented."""
1067
1068
    print("Checking all models are included.")
    check_model_list()
1069
1070
    print("Checking all models are public.")
    check_models_are_in_init()
1071
    print("Checking all models are properly tested.")
Sylvain Gugger's avatar
Sylvain Gugger committed
1072
    check_all_decorator_order()
1073
    check_all_models_are_tested()
1074
    print("Checking all objects are properly documented.")
1075
    check_all_objects_are_documented()
1076
1077
    print("Checking all models are in at least one auto class.")
    check_all_models_are_auto_configured()
1078
1079
    print("Checking all names in auto name mappings are defined.")
    check_all_auto_object_names_being_defined()
Yih-Dar's avatar
Yih-Dar committed
1080
1081
    print("Checking all keys in auto name mappings are defined in `CONFIG_MAPPING_NAMES`.")
    check_all_auto_mapping_names_in_config_mapping_names()
1082
1083
    print("Checking all auto mappings could be imported.")
    check_all_auto_mappings_importable()
1084
1085
    print("Checking all objects are equally (across frameworks) in the main __init__.")
    check_objects_being_equally_in_main_init()
1086
1087
1088
1089


if __name__ == "__main__":
    check_repo_quality()