U 4Af @sddlZddlZddlZddlZddlmZmZddlmZddl m Z ddl m Z ddl mZddlmZmZmZmZmZmZmZmZddlZdd lmZdd lmZdd lmZm Z m!Z!m"Z"d d l#m$Z$e"%e&Z'e (j)Z*ee+j,j,a-t-j,j,a.eGdddZ/dddddddddddd Z0e1e2dddZ3e1e4ddd Z5e1ee1d!d"d#Z6de1e4e1d$d%d&Z7de1e1eee1efeee1efe2e1d(d)d*Z8dee1ej9fe1eee1efeee1efe2d+d,d-Z:e1e/e/ee1e1fd.d/d0Z;d1d2Zej?d=ej@d>ZAd?d@ZBdee1ej9fe/e/ee1e2ee1dBdCdDZCdeee1ej9feee1eee1ej9fdEdFdGZDde1eee1ee1eeeeffdHdIdJZEej?dKej@d>ZFde1eee1eeeeffe1dLdMdNZGdOdPZHe?dQZIde1eee1ee1ee1fdHdRdSZJdeee1dTdUdVZKdee1ej9feee1e2dWdXdYZLde/e/eee1e2dZd[d\ZMe/e/d]d^d_ZNd`dadbgdcgddgdegdfgdggdhgdiZOe/e/ee1ee1fdjdkdlZPdmZQdee1ej9fe/e/eee1ej9feee1dndodpZRdqdrZSde1e/e2eee1ee1dsdtduZTedvdwdxZUGdydzdze$ZVde1ee1eeeeee1ed{d|d}ZWe1e2d~ddZXddZYdS)N)ArgumentParser Namespace) dataclass)date)chain)Path)AnyCallableDictListOptionalPatternTupleUnion)auto)model_type_to_module_name)is_flax_availableis_tf_availableis_torch_availablelogging)BaseTransformersCLICommandc@seZdZUdZeed<eed<dZeeed<dZeeed<dZ eeed<dZ eeed<dZ eeed <dZ eeed <dZ eeed <dZeeed <dZeeed <ddZdS) ModelPatternsa Holds the basic information about a new model for the add-new-model-like command. Args: model_name (`str`): The model name. checkpoint (`str`): The checkpoint to use for doc examples. model_type (`str`, *optional*): The model type, the identifier used internally in the library like `bert` or `xlm-roberta`. Will default to `model_name` lowercased with spaces replaced with minuses (-). model_lower_cased (`str`, *optional*): The lowercased version of the model name, to use for the module name or function names. Will default to `model_name` lowercased with spaces and minuses replaced with underscores. model_camel_cased (`str`, *optional*): The camel-cased version of the model name, to use for the class names. Will default to `model_name` camel-cased (with spaces and minuses both considered as word separators. model_upper_cased (`str`, *optional*): The uppercased version of the model name, to use for the constant names. Will default to `model_name` uppercased with spaces and minuses replaced with underscores. config_class (`str`, *optional*): The tokenizer class associated with this model. Will default to `"{model_camel_cased}Config"`. tokenizer_class (`str`, *optional*): The tokenizer class associated with this model (leave to `None` for models that don't use a tokenizer). image_processor_class (`str`, *optional*): The image processor class associated with this model (leave to `None` for models that don't use an image processor). feature_extractor_class (`str`, *optional*): The feature extractor class associated with this model (leave to `None` for models that don't use a feature extractor). processor_class (`str`, *optional*): The processor class associated with this model (leave to `None` for models that don't use a processor). model_name checkpointN model_typemodel_lower_casedmodel_camel_casedmodel_upper_cased config_classtokenizer_classimage_processor_classfeature_extractor_classprocessor_classcCs|jdkr|jdd|_|jdkrD|jdddd|_|jdkr|jd}ttdd|D}dd|D}d ||_|j dkr|j dddd|_ |j dkr|jd|_ dS) N -_cSsg|]}|dqS)r&)split.0wr,L/tmp/pip-unpacked-wheel-zw5xktn0/transformers/commands/add_new_model_like.py `sz/ModelPatterns.__post_init__..cSs$g|]}|d|ddqS)rrN)upperr)r,r,r-r.bsConfig) rrlowerreplacerrr(listrjoinrr/r )selfwordsr,r,r- __post_init__Xs       zModelPatterns.__post_init__)__name__ __module__ __qualname____doc__str__annotations__rr rrrr r!r"r#r$r8r,r,r,r-r*s  rz[CONFIG_CLASS]z[TOKENIZER_CLASS]z[IMAGE_PROCESSOR_CLASS]z[FEATURE_EXTRACTOR_CLASS]z[PROCESSOR_CLASS]z [CHECKPOINT]z [MODEL_TYPE]z[MODEL_UPPER_CASED]z[MODEL_CAMELCASED]z[MODEL_LOWER_CASED]z [MODEL_NAME]) r r!r"r#r$rrrrrr)linereturncCst|dkp|S)z4 Determines whether a line is empty or not. r)lenisspacer?r,r,r- is_empty_lineysrDcCs(td|}|dkrdSt|dS)z@ Returns the number of spaces that start a line indent. z^(\s*)(?:\S|$)Nr)researchrAgroups)r?rFr,r,r- find_indents rH)contentr@cCsg}g}|d}ddddg}|D]}t|dk}|rRt|dkrR|dd }t|st|dkr|r||kr|||d|g}q|d||g}q"||q"t|dkr|d||S) z Parse the content of a module in the list of objects it defines. Args: content (`str`): The content to parse Returns: `List[str]`: The list of objects defined in the module.  )]}"""rrz # Copied from)r(rA startswithrDrHappendr5)rIobjectscurrent_objectlines end_markersr?is_valid_objectr,r,r-parse_module_contents&       rV)rI indent_levelr@cCsg}|d}ddddg}t|D]\}}|dkrh|dkrht|sht||krhtd|dt|d t||krt|sqt|dk}t|s|d st||kr|r||kr||d |S||q"t|dkrd |Sd S) aEReturn the first block in `content` with the indent level `indent_level`. The first line in `content` should be indented at `indent_level` level, otherwise an error will be thrown. This method will immediately stop the search when a (non-empty) line with indent level less than `indent_level` is encountered. Args: content (`str`): The content to parse indent_level (`int`, *optional*, default to 0): The indent level of the blocks to search for Returns: `str`: The first block in `content` with the indent level `indent_level`. rJrKrLrMrNrzNWhen `indent_level > 0`, the first line in `content` should have indent level z. Got z instead.:N) r( enumeraterDrH ValueErrorrAendswithlstriprPr5)rIrWrRrSrTidxr?rUr,r,r- extract_blocks2  $      r^F)textrI add_after add_before exact_matchr@cs|dkr|dkrtd|dk r0|dk r0td|dkr<|n|fdd}g}|dD]F}||r|dk rz|||||dk r||q\||q\d|S)a A utility to add some content inside a given text. Args: text (`str`): The text in which we want to insert some content. content (`str`): The content to add. add_after (`str` or `Pattern`): The pattern to test on a line of `text`, the new content is added after the first instance matching it. add_before (`str` or `Pattern`): The pattern to test on a line of `text`, the new content is added before the first instance matching it. exact_match (`bool`, *optional*, defaults to `False`): A line is considered a match with `add_after` or `add_before` if it matches exactly when `exact_match=True`, otherwise, if `add_after`/`add_before` is present in the line. The arguments `add_after` and `add_before` are mutually exclusive, and one exactly needs to be provided. Returns: `str`: The text with the new content added if a match was found. Nz3You need to pass either `add_after` or `add_before`z/You can't pass both `add_after` or `add_before`cs0ttr|dk Sr$|kS|kSdSN) isinstancer rFrCrbpatternr,r-this_is_the_lines  z-add_content_to_text..this_is_the_linerJ)rZr(rPr5)r_rIr`rarbrg new_linesr?r,rer-add_content_to_texts     ri) file_namerIr`rarbc Cs\t|ddd}|}W5QRXt|||||d}t|ddd}||W5QRXdS)a A utility to add some content inside a given file. Args: file_name (`str` or `os.PathLike`): The name of the file in which we want to insert some content. content (`str`): The content to add. add_after (`str` or `Pattern`): The pattern to test on a line of `text`, the new content is added after the first instance matching it. add_before (`str` or `Pattern`): The pattern to test on a line of `text`, the new content is added before the first instance matching it. exact_match (`bool`, *optional*, defaults to `False`): A line is considered a match with `add_after` or `add_before` if it matches exactly when `exact_match=True`, otherwise, if `add_after`/`add_before` is present in the line. The arguments `add_after` and `add_before` are mutually exclusive, and one exactly needs to be provided. rutf-8encoding)r`rarbr+N)openreadriwrite)rjrIr`rarbfZ old_content new_contentr,r,r-add_content_to_filesrt)r_old_model_patternsnew_model_patternsr@c Csdg}dD]*}t||dk r t||dk r ||q |j|j|jfkrR|d|j|jkrj|dntd|jdd|}|j|jkr|j}t |d |dk rt|d d |}n |d | d ddg|D]}| t||t |}qg}t D]@\}}||kr|t||t||f| |t||}qdd|D}tt|t|krt|dfSt|}dd|D}|d|fS)az Replace all patterns present in a given text. Args: text (`str`): The text to treat. old_model_patterns (`ModelPatterns`): The patterns for the old model. new_model_patterns (`ModelPatterns`): The patterns for the new model. Returns: `Tuple(str, str)`: A tuple of with the treated text and the replacement actually done in it. r )r!r"r#r$Nrrz(\s*)model_type = ""z\1model_type = "[MODEL_TYPE]"z_[A-Z_]*[^A-Z_]z([A-Z_]*)([^a-zA-Z_])z[MODEL_UPPER_CASED]\1\2rrrrcSsg|] \}}|qSr,r,r*oldnewr,r,r-r.}sz*replace_model_patterns..r0cSsg|]\}}|d|qS)z->r,rxr,r,r-r.s,)getattrrPrrrrEsubrrrFextendr3ATTRIBUTE_TO_PLACEHOLDERitemsrAsetsimplify_replacementsr5) r_rurvZattributes_to_checkattrZold_model_value replacements placeholderZold_replacement_valuesr,r,r-replace_model_patternsDs@        rcCst|dkr|S|jdddd}|t|kr||\}}|d}|t|kr||\}}||||krx||qD|d7}qD|d7}q$|S)a Simplify a list of replacement patterns to make sure there are no needless ones. For instance in the sequence "Bert->BertNew, BertConfig->BertNewConfig, bert->bert_new", the replacement "BertConfig->BertNewConfig" is implied by "Bert->BertNew" so not needed. Args: replacements (`List[Tuple[str, str]]`): List of patterns (old, new) Returns: `List[Tuple[str, str]]`: The list of patterns simplified. rcSs t|dSNr)rAxr,r,r-z'simplify_replacements..keyr)rAsortr3pop)rr]ryrzjZold_2Znew_2r,r,r-rs        r) module_filer@cCsjt|}|dj}t|d}|dkrB||dkrB|d8}q$|dkrXt|dd||dS)zA Returns the module name corresponding to a module file. r0rr transformersz is not a transformers module..N)rabsolute with_suffixpartsrArZr5)rZfull_module_path module_partsr]r,r,r-get_module_from_files    rrr r!r"r#r$)z_CHECKPOINT_FOR_DOC =z_CONFIG_FOR_DOC =z_TOKENIZER_FOR_DOC =z_IMAGE_PROCESSOR_FOR_DOC =z_FEAT_EXTRACTOR_FOR_DOC =z_PROCESSOR_FOR_DOC =z(^(?:class|def)\s+([^\s:\(]+)\s*(?:\(|\:)flagsc Cs.|tj}d}t|D]D\}}||dr>|}q^q|d|dr|}q^q|dkrj|S||}t|}td||d|}t |d}t |D]}d|||<qt |dddD]H}||}|ds|d r t||kr d||<qqqtjd d |D} | S) zRemove `target_attr` in `obj`.Nz = zdef (rJr#@cSsg|]}|dk r|qSrcr,)r*rr,r,r-r.sz%remove_attributes..) r(oslineseprYr\rOrHr^r5rArange) obj target_attrrSZ target_idxr]r?rWparsedZ num_linesZnew_objr,r,r-remove_attributess0  , rT)rrurv dest_fileadd_copied_fromattrs_to_removec Cs|dkrt||j|j}t|ddd}|}W5QRXtddtd|}t|}g} |D]} d} t D]:\} } | | krt| t || t || } | | d } qqt| rqd| }t | ||\} }tjd | tjd dk }|rD|sDt| dk rDt|d krDt|}t|d }t| d |d|d|td} tdd| } | | qdd| }|dk r|D]} t|| d}qvt|ddd}||W5QRXdS)a\ Create a new module from an existing one and adapting all function and classes names from old patterns to new ones. Args: module_file (`str` or `os.PathLike`): Path to the module to duplicate. old_model_patterns (`ModelPatterns`): The patterns for the old model. new_model_patterns (`ModelPatterns`): The patterns for the new model. dest_file (`str` or `os.PathLike`, *optional*): Path to the new module. add_copied_from (`bool`, *optional*, defaults to `True`): Whether or not to add `# Copied from` statements in the duplicated module. Nrkrlrmz# Copyright (\d+)\sz # Copyright r%FTz^#\s+Copied fromrrz# Copied from rz with )raz [ ]+# Copied from [^ ]* rJ)rr+)r=r3rrorprEr} CURRENT_YEARrVSPECIAL_PATTERNSrr|rPrrF MULTILINE_re_class_funcrArrGrir5rrq)rrurvrrrrrrIrQZ new_objectsrZspecial_patternrfrZold_obj replacementZhas_copied_from module_nameZold_object_namer,r,r-duplicate_modulesN *   r)files frameworksr@cs|dkrt}ig}|D]T}t|jd}d|krB||qd|krT|d<qd|krf|d<q|d<qfdd|D|S) aW Filter a list of files to only keep the ones corresponding to a list of frameworks. Args: files (`List[Union[str, os.PathLike]]`): The list of files to filter. frameworks (`List[str]`, *optional*): The list of allowed frameworks. Returns: `List[Union[str, os.PathLike]]`: The list of filtered files. Nr'modelingtfflaxptcsg|]}|kr|qSr,r,r*rrZframework_to_filer,r-r.Vsz*filter_framework_files..)get_default_frameworksrnamer(rP)rrZothersrrrr,rr-filter_framework_files8s     r)rrr@c st|td}t|d}t||d}tdddd|d}d d d d d d d d dd dd dd g}t||d}fdd|D}dd|D}|||dS)a2 Retrieves all the files associated to a model. Args: model_type (`str`): A valid model type (like "bert" or "gpt2") frameworks (`List[str]`, *optional*): If passed, will only keep the model files corresponding to the passed frameworks. Returns: `Dict[str, Union[Path, List[Path]]]`: A dictionary with the following keys: - **doc_file** -- The documentation file for the model. - **model_files** -- All the files in the model module. - **test_files** -- The test files for the model. modelsz*.pyrdocssourceen model_doc.mdZtest_modeling_z.pyZtest_modeling_tf_Ztest_modeling_flax_Ztest_tokenization_Ztest_image_processing_Ztest_feature_extraction_Ztest_processor_cs g|]}tdd|qS)testsr) REPO_PATHrrr,r-r.|sz#get_model_files..cSsg|]}|r|qSr,)existsrr,r,r-r.~s)doc_file model_filesr test_files)rTRANSFORMERS_PATHr4globrr)rrZ model_modulerrrr,rr-get_model_filesYs"          rz$^_CHECKPOINT_FOR_DOC\s+=\s+(\S*)\s*$)rrr@c Cs|dkrt|}|d}|D]}dt|kr.qt|dddZ}|}t|dk rt|d}|dd }|d d }|W5QRSW5QRXqd S) a Finds the model checkpoint used in the docstrings for a given model. Args: model_type (`str`): A valid model type (like "bert" or "gpt2") model_files (`Dict[str, Union[Path, List[Path]]`, *optional*): The files associated to `model_type`. Can be passed to speed up the function, otherwise will be computed. Returns: `str`: The checkpoint used. Nrrrkrlrmrrwr0')rr=rorp_re_checkpoint_for_docrFrGr3)rrZ module_filesfnamerrrIrr,r,r-find_base_model_checkpoints    rcCs8g}tr|dtr$|dtr4|d|S)zk Returns the list of frameworks (PyTorch, TensorFlow, Flax) that are installed in the environment. rrr)rrPrrrr,r,r-rs   rzMODEL_([A-Z_]*)MAPPING_NAMESc Cs|dkrt}trtjndtr(tjndtr6tjndd}i}|D]}g}||dkrjtd|dddt ||D}|D](}t |||}||kr| ||qt |dkrFt t|||<qF|S)a Retrieve the model classes associated to a given model. Args: model_type (`str`): A valid model type (like "bert" or "gpt2") frameworks (`List[str]`, *optional*): The frameworks to look for. Will default to `["pt", "tf", "flax"]`, passing a smaller list will restrict the classes returned. Returns: `Dict[str, List[str]]`: A dictionary with one key per framework and the list of model classes associated to that framework as values. Nrrrz You selected z, in the frameworks, but it is not installed.cSsg|]}t|dk r|qSrc)_re_model_mappingrF)r*rr,r,r-r.sz*retrieve_model_classes..r)rr auto_moduleZ modeling_autorZmodeling_tf_autorZmodeling_flax_autorZdirr|rPrAr4r) rrmodules model_classes frameworkZnew_model_classesZmodel_mappingsZmodel_mapping_nameZ model_mappingr,r,r-retrieve_model_classess&  rrc sv|tjkrt|dtj|}tjj|}|tjjkrdtjj|}|ddk rZ|dn|d}nd}tjj |d}tj j |d}tj j |d}t||d} |dd} g| dD]J} d t| kr؈d qd t| krd qd t| krdq|dkrt}fdd|D}t||d} | } t|t|| d|| | d| |||||d }|| | |dS)a Retrieves all the information from a given model_type. Args: model_type (`str`): A valid model type (like "bert" or "gpt2") frameworks (`List[str]`, *optional*): If passed, will only keep the info corresponding to the passed frameworks. Returns: `Dict`: A dictionary with the following keys: - **frameworks** (`List[str]`): The list of frameworks that back this model type. - **model_classes** (`Dict[str, List[str]]`): The model classes implemented for that model type. - **model_files** (`Dict[str, Union[Path, List[Path]]]`): The files associated with that model type. - **model_patterns** (`ModelPatterns`): The various patterns for the model.  is not a valid model type.rNrrr1r0rZ modeling_tfrZ modeling_flaxrrrcsg|]}|kr|qSr,r,rZavailable_frameworksr,r-r.sz+retrieve_info_for_model..)rr) rrrrrr r!r"r#r$)rrrmodel_patterns)rMODEL_NAMES_MAPPINGrZconfiguration_autoZCONFIG_MAPPING_NAMESZtokenization_autoZTOKENIZER_MAPPING_NAMESZimage_processing_autoZIMAGE_PROCESSOR_MAPPING_NAMESgetZfeature_extraction_autoZFEATURE_EXTRACTOR_MAPPING_NAMESZprocessing_autoZPROCESSOR_MAPPING_NAMESrr3r=rPrrr/rr)rrrr Ztokenizer_classesr!r"r#r$rrrrrrr,rr-retrieve_info_for_modelsX                 r) init_filerkeep_processingc sdkrtddifdddD}|s>|ddd gt|d krNdSd |}td |d }td}td}td|d}t|ddd} | } W5QRX| d} g} d } | t| krp| | | dk r~| | | ddk r~| | d7} t | | s0| | | dkr<| d7} q| d7} t | | }t | | |kspt | | rn| d7} qPq| | | dk r| | }|D]V}| d|dd}| d|dd}| d|dd}| d|dd}qt|d kr| || d7} q|sNt d| | dkrft d| | dkrf| | | | d7} q| d7} qt|ddd} | d| W5QRXdS) ag Removes all the import lines that don't belong to a given list of frameworks or concern tokenizers/feature extractors/image processors/processors in an init. Args: init_file (`str` or `os.PathLike`): The path to the init to treat. frameworks (`List[str]`, *optional*): If passed, this will remove all imports that are subject to a framework not in frameworks keep_processing (`bool`, *optional*, defaults to `True`): Whether or not to keep the preprocessing (tokenizer, feature extractor, image processor, processor) imports in the init. NrZtorchcs g|]}|kr||qSr,)rrrnamesr,r-r.=sz,clean_frameworks_in_init..rZ sentencepieceZ tokenizersZvisionr|z^\s*if not is_(z)_available\(\):\s*$z\s*try:z\s*else:zis_(z )_availablerkrlrmrJrz, is_Z _availabler0is_z _available, z _available,zB^\s*"(tokenization|processing|feature_extraction|image_processing)zG^\s*from .(tokenization|processing|feature_extraction|image_processing)r+)rr~rAr5rEcompilerorpr(rFrrDrHr3striprPrq)rrrZ to_removeZremove_patternZre_conditional_importsZre_tryZre_elseZre_is_xxx_availablerrrIrSrhr]indentr?rr,rr-clean_frameworks_in_init*sd     ,"       r)rurvrwith_processingc Csttdddd}|}W5QRX|d}d}g}d} |t|krd} t||spt||dkrpd} nR||d rd } d } n6||d rd } d } n||drd} d } | r|| dkr| |||d7}q| dk r$|dk r$| |kr$| |||d7}q.z "z",r0z, "rwr%r{, r+)rorrpr(rArDrHr\rOrrPrErFrr5r!r"r#r$r3rrq)rurvrrrrrIrSr]rhrZ new_frameworkblockrZ add_blockZprocessing_classesZprocessing_classr,r,r-add_model_to_main_initysv           r)rurvc Csn|jdks|jdkrdSttdddddd}|}W5QRX|d}d }||d sl|d 7}qT|d 7}||d s||d r||}n4g}||ds||||d 7}qd|}|d 7}d|j d|krt|j|krtqqt| |j |j }| |j|j}|d||g||d}ttdddddd}| d|W5QRXdS)z Add a tokenizer to the relevant mappings in the auto module. Args: old_model_patterns (`ModelPatterns`): The patterns for the old model. new_model_patterns (`ModelPatterns`): The patterns for the new model. Nrrztokenization_auto.pyrkrlrmrJrz* TOKENIZER_MAPPING_NAMES = OrderedDict(rz$TOKENIZER_MAPPING = _LazyAutoMappingr{z ),rwr+) r!rorrpr(rOr[rPr5rr3rq) rurvrrrIrSr]r new_blockrhr,r,r-insert_tokenizer_in_auto_modules2     rz) ("{model_type}", "{model_name}"),z+ ("{model_type}", "{config_class}"),z5 ("{model_type}", "{pretrained_archive_map}"),z6 ("{model_type}", "{feature_extractor_class}"),z4 ("{model_type}", "{image_processor_class}"),z+ ("{model_type}", "{any_pt_class}"),z+ ("{model_type}", "{any_tf_class}"),z- ("{model_type}", "{any_flax_class}"),z. ("{model_type}", "{processor_class}"),)zconfiguration_auto.pyzfeature_extraction_auto.pyzimage_processing_auto.pyzmodeling_auto.pyzmodeling_tf_auto.pyzmodeling_flax_auto.pyzprocessing_auto.py)rurvrc stD]}g}t|D]tddk rhtdd|krf|fdd|Dqdkr|d|jqdkr|jdk r|jdk r|d|jqdkr|j dk r|j dk r|d|j qd kr"|j dk r,|j dk r,|d |j q|q|D]|t d d |}}}d D]:}|d |dt ||}|d |dt ||}qR||j |j }t|||dq2qt||dS)aV Add a model to the relevant mappings in the auto module. Args: old_model_patterns (`ModelPatterns`): The patterns for the old model. new_model_patterns (`ModelPatterns`): The patterns for the new model. model_classes (`Dict[str, List[str]]`): A dictionary framework to list of model classes implemented. zany_([a-z]*)_classNrcs(g|] }dddd|qS){Zany__classrM)r3)r*clsrrfr,r-r.#sz-add_model_to_auto_classes..z{config_class}z{image_processor_class}z{feature_extractor_class}z{processor_class}rr)rrrrM)r`)AUTO_CLASSES_PATTERNSrErFrGr~rPr3r r"r#r$rr|rrtr) rurvrfilenameZ new_patternsZ full_nameZold_model_lineZnew_model_linerr,rr-add_model_to_auto_classess\      ra## Overview The {model_name} model was proposed in []() by . The abstract from the paper is the following: ** Tips: This model was contributed by [INSERT YOUR HF USERNAME HERE](https://huggingface.co/). The original code can be found [here](). )rrurvrrc Cst|ddd}|}W5QRXtddtd|}|dkrFt}|dkrdt|j|jd}| d }g}g} |D]0} | d r| d | | g} qz| | qz| d | g} d } |D]} | d s| | qt d | dk r | d |jd q| s\|j| d dkr\d} | tj|jdt| ||\}}| |q| rd} | d d}t d|d}t| ||\}}d|kr|j|jkr| |qd|kr|j|jkr| |qd|kr|j|jkr| |qd|kr&|j|jkr| |q| drHd|kr| |q| drjd|kr| |qt| ddkrd|kr| |q| |qt|ddd}|d | W5QRXdS)a Duplicate a documentation file and adapts it for a new model. Args: module_file (`str` or `os.PathLike`): Path to the doc file to duplicate. old_model_patterns (`ModelPatterns`): The patterns for the old model. new_model_patterns (`ModelPatterns`): The patterns for the new model. dest_file (`str` or `os.PathLike`, *optional*): Path to the new doc file. Will default to the a file named `{new_model_patterns.model_type}.md` in the same folder as `module_file`. frameworks (`List[str]`, *optional*): If passed, will only keep the model classes corresponding to this list of frameworks in the new doc file. rkrlrmz