check_copies.py 24 KB
Newer Older
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
# 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.

import argparse
import glob
18
import importlib.util
19
20
import os
import re
21
import sys
22
23

import black
24
from doc_builder.style_doc import style_docstrings_in_code
25
26
27
28
29


# All paths are set with the intent you should run this script from the root of the repo with the command
# python utils/check_copies.py
TRANSFORMERS_PATH = "src/transformers"
30
PATH_TO_DOCS = "docs/source/en"
31
REPO_PATH = "."
32

33
# Mapping for files that are full copies of others (keys are copies, values the file to keep them up to data with)
34
35
36
37
FULL_COPIES = {
    "examples/tensorflow/question-answering/utils_qa.py": "examples/pytorch/question-answering/utils_qa.py",
    "examples/flax/question-answering/utils_qa.py": "examples/pytorch/question-answering/utils_qa.py",
}
38

39

40
41
42
43
44
LOCALIZED_READMES = {
    # If the introduction or the conclusion of the list change, the prompts may need to be updated.
    "README.md": {
        "start_prompt": "🤗 Transformers currently provides the following architectures",
        "end_prompt": "1. Want to contribute a new model?",
Sylvain Gugger's avatar
Sylvain Gugger committed
45
46
47
48
        "format_model_list": (
            "**[{title}]({model_link})** (from {paper_affiliations}) released with the paper {paper_title_link} by"
            " {paper_authors}.{supplements}"
        ),
49
50
51
52
    },
    "README_zh-hans.md": {
        "start_prompt": "🤗 Transformers 目前支持如下的架构",
        "end_prompt": "1. 想要贡献新的模型?",
Sylvain Gugger's avatar
Sylvain Gugger committed
53
54
55
56
        "format_model_list": (
            "**[{title}]({model_link})** (来自 {paper_affiliations}) 伴随论文 {paper_title_link} 由 {paper_authors}"
            " 发布。{supplements}"
        ),
57
58
59
60
    },
    "README_zh-hant.md": {
        "start_prompt": "🤗 Transformers 目前支援以下的架構",
        "end_prompt": "1. 想要貢獻新的模型?",
Sylvain Gugger's avatar
Sylvain Gugger committed
61
62
63
64
        "format_model_list": (
            "**[{title}]({model_link})** (from {paper_affiliations}) released with the paper {paper_title_link} by"
            " {paper_authors}.{supplements}"
        ),
65
    },
66
67
68
    "README_ko.md": {
        "start_prompt": "🤗 Transformers는 다음 모델들을 제공합니다",
        "end_prompt": "1. 새로운 모델을 올리고 싶나요?",
Sylvain Gugger's avatar
Sylvain Gugger committed
69
        "format_model_list": (
Arthur's avatar
Arthur committed
70
71
            "**[{title}]({model_link})** ({paper_affiliations} 에서 제공)은 {paper_authors}.{supplements}의"
            " {paper_title_link}논문과 함께 발표했습니다."
Sylvain Gugger's avatar
Sylvain Gugger committed
72
        ),
73
    },
74
75
76
77
78
79
80
81
    "README_es.md": {
        "start_prompt": "🤗 Transformers actualmente proporciona las siguientes arquitecturas",
        "end_prompt": "1. ¿Quieres aportar un nuevo modelo?",
        "format_model_list": (
            "**[{title}]({model_link})** (from {paper_affiliations}) released with the paper {paper_title_link} by"
            " {paper_authors}.{supplements}"
        ),
    },
82
83
84
85
    "README_ja.md": {
        "start_prompt": "🤗Transformersは現在、以下のアーキテクチャを提供しています",
        "end_prompt": "1. 新しいモデルを投稿したいですか?",
        "format_model_list": (
86
87
            "**[{title}]({model_link})** ({paper_affiliations} から) {paper_authors}.{supplements} から公開された研究論文"
            " {paper_title_link}"
88
89
        ),
    },
90
91
92
93
    "README_hd.md": {
        "start_prompt": "🤗 ट्रांसफॉर्मर वर्तमान में निम्नलिखित आर्किटेक्चर का समर्थन करते हैं",
        "end_prompt": "1. एक नए मॉडल में योगदान देना चाहते हैं?",
        "format_model_list": (
94
95
            "**[{title}]({model_link})** ({paper_affiliations} से) {paper_authors}.{supplements} द्वारा"
            "अनुसंधान पत्र {paper_title_link} के साथ जारी किया गया"
96
97
        ),
    },
98
99
100
}


101
102
103
104
105
106
# 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(TRANSFORMERS_PATH, "__init__.py"),
    submodule_search_locations=[TRANSFORMERS_PATH],
)
107
108
109
transformers_module = importlib.util.module_from_spec(spec)
spec.loader.exec_module(transformers_module)
transformers_module = sys.modules["transformers"]
110
111


112
def _should_continue(line, indent):
113
    return line.startswith(indent) or len(line) <= 1 or re.search(r"^\s*\)(\s*->.*:|:)\s*$", line) is not None
114
115


116
def find_code_in_transformers(object_name):
Patrick von Platen's avatar
Patrick von Platen committed
117
    """Find and return the code source code of `object_name`."""
118
119
120
121
122
123
124
    parts = object_name.split(".")
    i = 0

    # First let's find the module where our object lives.
    module = parts[i]
    while i < len(parts) and not os.path.isfile(os.path.join(TRANSFORMERS_PATH, f"{module}.py")):
        i += 1
125
126
        if i < len(parts):
            module = os.path.join(module, parts[i])
127
128
129
130
131
    if i >= len(parts):
        raise ValueError(
            f"`object_name` should begin with the name of a module of transformers but got {object_name}."
        )

132
    with open(os.path.join(TRANSFORMERS_PATH, f"{module}.py"), "r", encoding="utf-8", newline="\n") as f:
133
134
135
136
137
138
        lines = f.readlines()

    # Now let's find the class / func in the code!
    indent = ""
    line_index = 0
    for name in parts[i + 1 :]:
139
        while (
140
            line_index < len(lines) and re.search(rf"^{indent}(class|def)\s+{name}(\(|\:)", lines[line_index]) is None
141
        ):
142
143
144
145
146
147
148
149
150
            line_index += 1
        indent += "    "
        line_index += 1

    if line_index >= len(lines):
        raise ValueError(f" {object_name} does not match any function or class in {module}.")

    # We found the beginning of the class / func, now let's find the end (when the indent diminishes).
    start_index = line_index
151
    while line_index < len(lines) and _should_continue(lines[line_index], indent):
152
153
154
155
156
157
158
159
160
161
        line_index += 1
    # Clean up empty lines at the end (if any).
    while len(lines[line_index - 1]) <= 1:
        line_index -= 1

    code_lines = lines[start_index:line_index]
    return "".join(code_lines)


_re_copy_warning = re.compile(r"^(\s*)#\s*Copied from\s+transformers\.(\S+\.\S+)\s*($|\S.*$)")
162
_re_replace_pattern = re.compile(r"^\s*(\S+)->(\S+)(\s+.*|$)")
163
_re_fill_pattern = re.compile(r"<FILL\s+[^>]*>")
164
165


166
167
168
169
170
171
172
def get_indent(code):
    lines = code.split("\n")
    idx = 0
    while idx < len(lines) and len(lines[idx]) == 0:
        idx += 1
    if idx < len(lines):
        return re.search(r"^(\s*)\S", lines[idx]).groups()[0]
173
174
175
176
177
178
179
180
181
182
    return ""


def blackify(code):
    """
    Applies the black part of our `make style` command to `code`.
    """
    has_indent = len(get_indent(code)) > 0
    if has_indent:
        code = f"class Bla:\n{code}"
183
    mode = black.Mode(target_versions={black.TargetVersion.PY37}, line_length=119)
184
    result = black.format_str(code, mode=mode)
185
    result, _ = style_docstrings_in_code(result)
186
    return result[len("class Bla:\n") :] if has_indent else result
187
188


189
190
191
192
193
194
def is_copy_consistent(filename, overwrite=False):
    """
    Check if the code commented as a copy in `filename` matches the original.

    Return the differences or overwrites the content depending on `overwrite`.
    """
195
    with open(filename, "r", encoding="utf-8", newline="\n") as f:
196
        lines = f.readlines()
197
    diffs = []
198
    line_index = 0
199
    # Not a for loop cause `lines` is going to change (if `overwrite=True`).
200
201
202
203
204
205
206
207
208
    while line_index < len(lines):
        search = _re_copy_warning.search(lines[line_index])
        if search is None:
            line_index += 1
            continue

        # There is some copied code here, let's retrieve the original.
        indent, object_name, replace_pattern = search.groups()
        theoretical_code = find_code_in_transformers(object_name)
209
        theoretical_indent = get_indent(theoretical_code)
210
211
212
213
214
215
216
217
218
219
220
221

        start_index = line_index + 1 if indent == theoretical_indent else line_index + 2
        indent = theoretical_indent
        line_index = start_index

        # Loop to check the observed code, stop when indentation diminishes or if we see a End copy comment.
        should_continue = True
        while line_index < len(lines) and should_continue:
            line_index += 1
            if line_index >= len(lines):
                break
            line = lines[line_index]
222
            should_continue = _should_continue(line, indent) and re.search(f"^{indent}# End copy", line) is None
223
224
225
226
227
228
229
230
231
        # Clean up empty lines at the end (if any).
        while len(lines[line_index - 1]) <= 1:
            line_index -= 1

        observed_code_lines = lines[start_index:line_index]
        observed_code = "".join(observed_code_lines)

        # Before comparing, use the `replace_pattern` on the original code.
        if len(replace_pattern) > 0:
232
233
234
235
236
237
            patterns = replace_pattern.replace("with", "").split(",")
            patterns = [_re_replace_pattern.search(p) for p in patterns]
            for pattern in patterns:
                if pattern is None:
                    continue
                obj1, obj2, option = pattern.groups()
238
                theoretical_code = re.sub(obj1, obj2, theoretical_code)
239
240
241
                if option.strip() == "all-casing":
                    theoretical_code = re.sub(obj1.lower(), obj2.lower(), theoretical_code)
                    theoretical_code = re.sub(obj1.upper(), obj2.upper(), theoretical_code)
242

243
244
245
246
247
            # Blackify after replacement. To be able to do that, we need the header (class or function definition)
            # from the previous line
            theoretical_code = blackify(lines[start_index - 1] + theoretical_code)
            theoretical_code = theoretical_code[len(lines[start_index - 1]) :]

248
249
        # Test for a diff and act accordingly.
        if observed_code != theoretical_code:
Rak Alexey's avatar
Rak Alexey committed
250
251
252
253
254
255
            diff_index = start_index + 1
            for observed_line, theoretical_line in zip(observed_code.split("\n"), theoretical_code.split("\n")):
                if observed_line != theoretical_line:
                    break
                diff_index += 1
            diffs.append([object_name, diff_index])
256
257
258
259
            if overwrite:
                lines = lines[:start_index] + [theoretical_code] + lines[line_index:]
                line_index = start_index + 1

260
    if overwrite and len(diffs) > 0:
261
262
        # Warn the user a file has been modified.
        print(f"Detected changes, rewriting {filename}.")
263
        with open(filename, "w", encoding="utf-8", newline="\n") as f:
264
            f.writelines(lines)
265
    return diffs
266
267
268
269
270
271


def check_copies(overwrite: bool = False):
    all_files = glob.glob(os.path.join(TRANSFORMERS_PATH, "**/*.py"), recursive=True)
    diffs = []
    for filename in all_files:
272
273
        new_diffs = is_copy_consistent(filename, overwrite)
        diffs += [f"- {filename}: copy does not match {d[0]} at line {d[1]}" for d in new_diffs]
274
275
276
    if not overwrite and len(diffs) > 0:
        diff = "\n".join(diffs)
        raise Exception(
277
            "Found the following copy inconsistencies:\n"
278
            + diff
279
            + "\nRun `make fix-copies` or `python utils/check_copies.py --fix_and_overwrite` to fix them."
280
        )
281
282
283
    check_model_list_copy(overwrite=overwrite)


284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
def check_full_copies(overwrite: bool = False):
    diffs = []
    for target, source in FULL_COPIES.items():
        with open(source, "r", encoding="utf-8") as f:
            source_code = f.read()
        with open(target, "r", encoding="utf-8") as f:
            target_code = f.read()
        if source_code != target_code:
            if overwrite:
                with open(target, "w", encoding="utf-8") as f:
                    print(f"Replacing the content of {target} by the one of {source}.")
                    f.write(source_code)
            else:
                diffs.append(f"- {target}: copy does not match {source}.")

    if not overwrite and len(diffs) > 0:
        diff = "\n".join(diffs)
        raise Exception(
            "Found the following copy inconsistencies:\n"
            + diff
            + "\nRun `make fix-copies` or `python utils/check_copies.py --fix_and_overwrite` to fix them."
        )


308
def get_model_list(filename, start_prompt, end_prompt):
Patrick von Platen's avatar
Patrick von Platen committed
309
    """Extracts the model list from the README."""
310
    with open(os.path.join(REPO_PATH, filename), "r", encoding="utf-8", newline="\n") as f:
311
312
313
        lines = f.readlines()
    # Find the start of the list.
    start_index = 0
314
    while not lines[start_index].startswith(start_prompt):
315
316
317
318
319
320
321
        start_index += 1
    start_index += 1

    result = []
    current_line = ""
    end_index = start_index

322
    while not lines[end_index].startswith(end_prompt):
323
324
325
326
327
328
329
330
331
332
333
334
335
        if lines[end_index].startswith("1."):
            if len(current_line) > 1:
                result.append(current_line)
            current_line = lines[end_index]
        elif len(lines[end_index]) > 1:
            current_line = f"{current_line[:-1]} {lines[end_index].lstrip()}"
        end_index += 1
    if len(current_line) > 1:
        result.append(current_line)

    return "".join(result)


336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
def convert_to_localized_md(model_list, localized_model_list, format_str):
    """Convert `model_list` to each localized README."""

    def _rep(match):
        title, model_link, paper_affiliations, paper_title_link, paper_authors, supplements = match.groups()
        return format_str.format(
            title=title,
            model_link=model_link,
            paper_affiliations=paper_affiliations,
            paper_title_link=paper_title_link,
            paper_authors=paper_authors,
            supplements=" " + supplements.strip() if len(supplements) != 0 else "",
        )

    # This regex captures metadata from an English model description, including model title, model link,
    # affiliations of the paper, title of the paper, authors of the paper, and supplemental data (see DistilBERT for example).
    _re_capture_meta = re.compile(
        r"\*\*\[([^\]]*)\]\(([^\)]*)\)\*\* \(from ([^)]*)\)[^\[]*([^\)]*\)).*?by (.*?[A-Za-z\*]{2,}?)\. (.*)$"
    )
355
356
    # This regex is used to synchronize link.
    _re_capture_title_link = re.compile(r"\*\*\[([^\]]*)\]\(([^\)]*)\)\*\*")
357
358
359
360
361
362
363
364
365
366
367
368

    if len(localized_model_list) == 0:
        localized_model_index = {}
    else:
        try:
            localized_model_index = {
                re.search(r"\*\*\[([^\]]*)", line).groups()[0]: line
                for line in localized_model_list.strip().split("\n")
            }
        except AttributeError:
            raise AttributeError("A model name in localized READMEs cannot be recognized.")

369
370
371
372
373
374
    model_keys = [re.search(r"\*\*\[([^\]]*)", line).groups()[0] for line in model_list.strip().split("\n")]

    # We exclude keys in localized README not in the main one.
    readmes_match = not any([k not in model_keys for k in localized_model_index])
    localized_model_index = {k: v for k, v in localized_model_index.items() if k in model_keys}

375
    for model in model_list.strip().split("\n"):
376
        title, model_link = _re_capture_title_link.search(model).groups()
377
        if title not in localized_model_index:
378
            readmes_match = False
379
380
            # Add an anchor white space behind a model description string for regex.
            # If metadata cannot be captured, the English version will be directly copied.
381
            localized_model_index[title] = _re_capture_meta.sub(_rep, model + " ")
382
383
384
385
386
        elif _re_fill_pattern.search(localized_model_index[title]) is not None:
            update = _re_capture_meta.sub(_rep, model + " ")
            if update != localized_model_index[title]:
                readmes_match = False
                localized_model_index[title] = update
387
388
389
390
391
        else:
            # Synchronize link
            localized_model_index[title] = _re_capture_title_link.sub(
                f"**[{title}]({model_link})**", localized_model_index[title], count=1
            )
392
393
394

    sorted_index = sorted(localized_model_index.items(), key=lambda x: x[0].lower())

395
    return readmes_match, "\n".join(map(lambda x: x[1], sorted_index)) + "\n"
396
397


Sylvain Gugger's avatar
Sylvain Gugger committed
398
def convert_readme_to_index(model_list):
399
    model_list = model_list.replace("https://huggingface.co/docs/transformers/main/", "")
Sylvain Gugger's avatar
Sylvain Gugger committed
400
401
402
    return model_list.replace("https://huggingface.co/docs/transformers/", "")


Sylvain Gugger's avatar
Sylvain Gugger committed
403
404
405
406
407
408
def _find_text_in_file(filename, start_prompt, end_prompt):
    """
    Find the text in `filename` between a line beginning with `start_prompt` and before `end_prompt`, removing empty
    lines.
    """
    with open(filename, "r", encoding="utf-8", newline="\n") as f:
409
        lines = f.readlines()
Sylvain Gugger's avatar
Sylvain Gugger committed
410
    # Find the start prompt.
411
    start_index = 0
Sylvain Gugger's avatar
Sylvain Gugger committed
412
    while not lines[start_index].startswith(start_prompt):
413
414
415
416
        start_index += 1
    start_index += 1

    end_index = start_index
Sylvain Gugger's avatar
Sylvain Gugger committed
417
    while not lines[end_index].startswith(end_prompt):
418
419
420
421
422
423
424
425
        end_index += 1
    end_index -= 1

    while len(lines[start_index]) <= 1:
        start_index += 1
    while len(lines[end_index]) <= 1:
        end_index -= 1
    end_index += 1
Sylvain Gugger's avatar
Sylvain Gugger committed
426
427
    return "".join(lines[start_index:end_index]), start_index, end_index, lines

428

Sylvain Gugger's avatar
Sylvain Gugger committed
429
def check_model_list_copy(overwrite=False, max_per_line=119):
Patrick von Platen's avatar
Patrick von Platen committed
430
    """Check the model lists in the README and index.rst are consistent and maybe `overwrite`."""
431
432
433
434
    # Fix potential doc links in the README
    with open(os.path.join(REPO_PATH, "README.md"), "r", encoding="utf-8", newline="\n") as f:
        readme = f.read()
    new_readme = readme.replace("https://huggingface.co/transformers", "https://huggingface.co/docs/transformers")
435
    new_readme = new_readme.replace(
436
437
438
439
440
441
442
443
444
445
446
        "https://huggingface.co/docs/main/transformers", "https://huggingface.co/docs/transformers/main"
    )
    if new_readme != readme:
        if overwrite:
            with open(os.path.join(REPO_PATH, "README.md"), "w", encoding="utf-8", newline="\n") as f:
                f.write(new_readme)
        else:
            raise ValueError(
                "The main README contains wrong links to the documentation of Transformers. Run `make fix-copies` to "
                "automatically fix them."
            )
447
448

    # If the introduction or the conclusion of the list change, the prompts may need to be updated.
Sylvain Gugger's avatar
Sylvain Gugger committed
449
450
451
452
    index_list, start_index, end_index, lines = _find_text_in_file(
        filename=os.path.join(PATH_TO_DOCS, "index.mdx"),
        start_prompt="<!--This list is updated automatically from the README",
        end_prompt="### Supported frameworks",
Sylvain Gugger's avatar
Sylvain Gugger committed
453
    )
454
455
456
457
458
459
460
461
462
463
464
465
466
    md_list = get_model_list(
        filename="README.md",
        start_prompt=LOCALIZED_READMES["README.md"]["start_prompt"],
        end_prompt=LOCALIZED_READMES["README.md"]["end_prompt"],
    )

    converted_md_lists = []
    for filename, value in LOCALIZED_READMES.items():
        _start_prompt = value["start_prompt"]
        _end_prompt = value["end_prompt"]
        _format_model_list = value["format_model_list"]

        localized_md_list = get_model_list(filename, _start_prompt, _end_prompt)
467
        readmes_match, converted_md_list = convert_to_localized_md(md_list, localized_md_list, _format_model_list)
468

469
        converted_md_lists.append((filename, readmes_match, converted_md_list, _start_prompt, _end_prompt))
470

Sylvain Gugger's avatar
Sylvain Gugger committed
471
472
    converted_md_list = convert_readme_to_index(md_list)
    if converted_md_list != index_list:
473
        if overwrite:
Sylvain Gugger's avatar
Sylvain Gugger committed
474
475
            with open(os.path.join(PATH_TO_DOCS, "index.mdx"), "w", encoding="utf-8", newline="\n") as f:
                f.writelines(lines[:start_index] + [converted_md_list] + lines[end_index:])
476
477
        else:
            raise ValueError(
Sylvain Gugger's avatar
Sylvain Gugger committed
478
                "The model list in the README changed and the list in `index.mdx` has not been updated. Run "
Sylvain Gugger's avatar
Sylvain Gugger committed
479
480
481
                "`make fix-copies` to fix this."
            )

482
    for converted_md_list in converted_md_lists:
483
        filename, readmes_match, converted_md, _start_prompt, _end_prompt = converted_md_list
484

485
486
487
488
489
490
491
492
        if filename == "README.md":
            continue
        if overwrite:
            _, start_index, end_index, lines = _find_text_in_file(
                filename=os.path.join(REPO_PATH, filename), start_prompt=_start_prompt, end_prompt=_end_prompt
            )
            with open(os.path.join(REPO_PATH, filename), "w", encoding="utf-8", newline="\n") as f:
                f.writelines(lines[:start_index] + [converted_md] + lines[end_index:])
493
        elif not readmes_match:
494
495
496
497
            raise ValueError(
                f"The model list in the README changed and the list in `{filename}` has not been updated. Run "
                "`make fix-copies` to fix this."
            )
498

Sylvain Gugger's avatar
Sylvain Gugger committed
499

500
501
502
503
504
505
SPECIAL_MODEL_NAMES = {
    "Bert Generation": "BERT For Sequence Generation",
    "BigBird": "BigBird-RoBERTa",
    "Data2VecAudio": "Data2Vec",
    "Data2VecText": "Data2Vec",
    "Data2VecVision": "Data2Vec",
506
    "DonutSwin": "Swin Transformer",
507
    "Marian": "MarianMT",
508
    "MaskFormerSwin": "Swin Transformer",
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
    "OpenAI GPT-2": "GPT-2",
    "OpenAI GPT": "GPT",
    "Perceiver": "Perceiver IO",
    "ViT": "Vision Transformer (ViT)",
}

# Update this list with the models that shouldn't be in the README. This only concerns modular models or those who do
# not have an associated paper.
MODELS_NOT_IN_README = [
    "BertJapanese",
    "Encoder decoder",
    "FairSeq Machine-Translation",
    "HerBERT",
    "RetriBERT",
    "Speech Encoder decoder",
    "Speech2Text",
    "Speech2Text2",
    "Vision Encoder decoder",
    "VisionTextDualEncoder",
]


README_TEMPLATE = (
532
533
    "1. **[{model_name}](https://huggingface.co/docs/main/transformers/model_doc/{model_type})** (from "
    "<FILL INSTITUTION>) released with the paper [<FILL PAPER TITLE>](<FILL ARKIV LINK>) by <FILL AUTHORS>."
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
)


def check_readme(overwrite=False):
    info = LOCALIZED_READMES["README.md"]
    models, start_index, end_index, lines = _find_text_in_file(
        os.path.join(REPO_PATH, "README.md"),
        info["start_prompt"],
        info["end_prompt"],
    )
    models_in_readme = [re.search(r"\*\*\[([^\]]*)", line).groups()[0] for line in models.strip().split("\n")]

    model_names_mapping = transformers_module.models.auto.configuration_auto.MODEL_NAMES_MAPPING
    absents = [
        (key, name)
        for key, name in model_names_mapping.items()
        if SPECIAL_MODEL_NAMES.get(name, name) not in models_in_readme
    ]
    # Remove exceptions
    absents = [(key, name) for key, name in absents if name not in MODELS_NOT_IN_README]
    if len(absents) > 0 and not overwrite:
        print(absents)
        raise ValueError(
            "The main README doesn't contain all models, run `make fix-copies` to fill it with the missing model(s)"
            " then complete the generated entries.\nIf the model is not supposed to be in the main README, add it to"
            " the list `MODELS_NOT_IN_README` in utils/check_copies.py.\nIf it has a different name in the repo than"
            " in the README, map the correspondence in `SPECIAL_MODEL_NAMES` in utils/check_copies.py."
        )

    new_models = [README_TEMPLATE.format(model_name=name, model_type=key) for key, name in absents]

    all_models = models.strip().split("\n") + new_models
    all_models = sorted(all_models, key=lambda x: re.search(r"\*\*\[([^\]]*)", x).groups()[0].lower())
    all_models = "\n".join(all_models) + "\n"

    if all_models != models:
        if overwrite:
            print("Fixing the main README.")
            with open(os.path.join(REPO_PATH, "README.md"), "w", encoding="utf-8", newline="\n") as f:
                f.writelines(lines[:start_index] + [all_models] + lines[end_index:])
        else:
            raise ValueError("The main README model list is not properly sorted. Run `make fix-copies` to fix this.")


578
579
580
581
582
if __name__ == "__main__":
    parser = argparse.ArgumentParser()
    parser.add_argument("--fix_and_overwrite", action="store_true", help="Whether to fix inconsistencies.")
    args = parser.parse_args()

583
    check_readme(args.fix_and_overwrite)
584
    check_copies(args.fix_and_overwrite)
585
    check_full_copies(args.fix_and_overwrite)