U 9%et@szddlZddlZddlZddlmZddlmZmZmZm Z m Z m Z ddl Z ddlmZmZmZddlmZddlmZddlmZmZmZmZmZmZd d lmZmZm Z e!e"Z#erdd l$m%Z%dd l&m'Z'erddl(Z(erddl)Z*dd l+m,Z,dZ-er&ddl.Z.ddl/m-Z-ddl0m1Z1e j2e j2e3e3e j2e dddZ4dddZ5GdddeZ6eeGddde Z7dS)N)Iterable) TYPE_CHECKINGDictListOptionalTupleUnion) SquadExample SquadFeatures"squad_convert_examples_to_features) ModelCard)PreTrainedTokenizer)PaddingStrategyadd_end_docstringsis_tf_availableis_tokenizers_availableis_torch_availablelogging)PIPELINE_INIT_ARGSArgumentHandler ChunkPipeline)TFPreTrainedModel)PreTrainedModel)-TF_MODEL_FOR_QUESTION_ANSWERING_MAPPING_NAMES)Dataset)*MODEL_FOR_QUESTION_ANSWERING_MAPPING_NAMES)startendtopkmax_answer_lenundesired_tokensreturncCs|jdkr|d}|jdkr$|d}tt|dt|d}tt||d}|}|dkrtt|g}nDt||krt | }n*t | |d|} | t ||  }t ||j dd\} } t | |t | |@} | | } | | } |d| | f} | | | fS)aG Take the output of any `ModelForQuestionAnswering` and will generate probabilities for each span to be the actual answer. In addition, it filters out some unwanted/impossible cases like answer len being greater than max_answer_len or answer end position being before the starting position. The method supports output the k-best answer through the topk argument. Args: start (`np.ndarray`): Individual start probabilities for each token. end (`np.ndarray`): Individual end probabilities for each token. topk (`int`): Indicates how many possible answer span(s) to extract from the model output. max_answer_len (`int`): Maximum size of the answer to extract from the model's output. undesired_tokens (`np.ndarray`): Mask determining tokens that can be part of the answer rNr)ndimnpmatmul expand_dimsZtrilZtriuflattenZargmaxlenZargsortZ argpartitionZ unravel_indexshapeisinnonzero)rrr r!r"outer candidatesZ scores_flatZidx_sortidxstartsendsZ desired_spansscoresr4h/var/www/html/Darija-Ai-API/env/lib/python3.8/site-packages/transformers/pipelines/question_answering.py decode_spans.s&    r6@BFc Cstt|d}|dk r$||@}|dk} t| d|}t| d|}t||jddd}||}t||jddd}||}|rt||d|d}d|d<|d<t |||||\} } } | | | |fS) ai Takes the raw output of any `ModelForQuestionAnswering` and first normalizes its outputs and then uses `decode_spans()` to generate probabilities for each span to be the actual answer. Args: start (`np.ndarray`): Individual start logits for each token. end (`np.ndarray`): Individual end logits for each token. p_mask (`np.ndarray`): A mask with 1 for values that cannot be in the answer attention_mask (`np.ndarray`): The attention mask generated by the tokenizer min_null_score(`float`): The minimum null (empty) answer score seen so far. topk (`int`): Indicates how many possible answer span(s) to extract from the model output. handle_impossible_answer(`bool`): Whether to allow null (empty) answers max_answer_len (`int`): Maximum size of the answer to extract from the model's output. rNggr$T)ZaxisZkeepdims)rr) r&absarraywhereexpmaxsumminitemr6) rrp_maskattention_maskmin_null_scoretop_khandle_impossible_answerr!r"Zundesired_tokens_maskr1r2r3r4r4r5select_starts_ends`s  rFc@s eZdZdZddZddZdS) QuestionAnsweringArgumentHandlera& QuestionAnsweringPipeline requires the user to provide multiple arguments (i.e. question & context) to be mapped to internal [`SquadExample`]. QuestionAnsweringArgumentHandler manages all the possible to create a [`SquadExample`] from the command-line supplied arguments. cCst|tr|St|trdD]b}||kr2tdq||dkrPtd|dqt||trt||dkrtd|dqtjf|St|ddS)NquestioncontextzFYou need to provide a dictionary with keys {question:..., context:...}`z` cannot be Nonerz` cannot be emptyz2 argument needs to be of type (SquadExample, dict)) isinstancer dictKeyError ValueErrorstrr*QuestionAnsweringPipeline create_sample)selfr@kr4r4r5 normalizes     z*QuestionAnsweringArgumentHandler.normalizecs|dk rlt|dkrlt|dkr*|d}n>t|dkr`dd|Dthkr`|d|ddg}nt|}ndkrd}ndkrd}nd krzd krztd trtd trڇfd d d D}ntd tr:td tr:td td krtd dd td d D}n>td trptd trpd d dg}ntdntdtdk rtjtfntjf}t||r|St|t r|g}n$t|t rt|}ntdt |D]\}}| |||<q|S)Nrrr cSsh|] }t|qSr4)type).0elr4r4r5 sz.rHXdatarIrJcsg|]}|ddqS)rJrHr4)rWQkwargsr4r5 sz=QuestionAnsweringArgumentHandler.__call__..z2Questions and contexts don't have the same lengthscSsg|]\}}||dqS)rHr4)rWr\Cr4r4r5r_szArguments can't be understoodzUnknown arguments zInvalid arguments ) r*rPlistrLrOziprtypes GeneratorTyperMr enumeraterU)rSargsr^inputsZgenerator_typesir@r4r]r5__call__s@             z)QuestionAnsweringArgumentHandler.__call__N)__name__ __module__ __qualname____doc__rUrir4r4r4r5rGsrGc seZdZdZdZdZd#edeee ee e dfdd Z e ee e e fee e e feee efd d d Zd$d dZfddZd%ddZddZd&ddZdeeeeeeefdddZe eeee ee effd d!d"ZZS)'rQa Question Answering pipeline using any `ModelForQuestionAnswering`. See the [question answering examples](../task_summary#question-answering) for more information. Example: ```python >>> from transformers import pipeline >>> oracle = pipeline(model="deepset/roberta-base-squad2") >>> oracle(question="Where do I live?", context="My name is Wolfgang and I live in Berlin") {'score': 0.9191, 'start': 34, 'end': 40, 'answer': 'Berlin'} ``` Learn more about the basics of using a pipeline in the [pipeline tutorial](../pipeline_tutorial) This question answering pipeline can currently be loaded from [`pipeline`] using the following task identifier: `"question-answering"`. The models that this pipeline can use are models that have been fine-tuned on a question answering task. See the up-to-date list of available models on [huggingface.co/models](https://huggingface.co/models?filter=question-answering). zquestion,contextFN)rrmodel tokenizer modelcard frameworktaskc sBtjf|||||d|t|_||jdkr8tntdS)Nrotf)super__init__rG _args_parserZcheck_model_typersrr)rSrprqrrrsrtr^ __class__r4r5rws  z"QuestionAnsweringPipeline.__init__)rIrJr#cCs4t|trddt||DStd||dddSdS)aC QuestionAnsweringPipeline leverages the [`SquadExample`] internally. This helper method encapsulate all the logic for converting question(s) and context(s) to [`SquadExample`]. We currently support extractive question answering. Arguments: question (`str` or `List[str]`): The question(s) asked. context (`str` or `List[str]`): The context(s) in which we will look for the answer. Returns: One or a list of [`SquadExample`]: The corresponding [`SquadExample`] grouping question and context. c Ss"g|]\}}td||dddqS)N)r )rWqcr4r4r5r_(sz;QuestionAnsweringPipeline.create_sample..N)rLrarbr rHr4r4r5rRs z'QuestionAnsweringPipeline.create_samplec Ksi} |dk r|| d<|dk r$|| d<|dk r4|| d<|dk rD|| d<i} |dk rh|dkrhtdt|}|dk r|dkrtd|d|| d <|dk r|dkrtd ||dk r|| d <|dk r|| d <| dk r| | d <| i| fS)Npadding doc_stridemax_question_len max_seq_lenz/topk parameter is deprecated, use top_k insteadrz$top_k parameter should be >= 1 (got )rDz-max_answer_len parameter should be >= 1 (got r!rEalign_to_words)warningswarn UserWarningrO) rSr}r rDr~r!rrrErr^Zpreprocess_paramsZpostprocess_paramsr4r4r5_sanitize_parameters,s6 z.QuestionAnsweringPipeline._sanitize_parameterscsJ|j||}t|ttfr:t|dkr:tj|df|Stj|f|S)aX Answer the question(s) given as inputs by using the context(s). Args: args ([`SquadExample`] or a list of [`SquadExample`]): One or several [`SquadExample`] containing the question and context. X ([`SquadExample`] or a list of [`SquadExample`], *optional*): One or several [`SquadExample`] containing the question and context (will be treated the same way as if passed as the first positional argument). data ([`SquadExample`] or a list of [`SquadExample`], *optional*): One or several [`SquadExample`] containing the question and context (will be treated the same way as if passed as the first positional argument). question (`str` or `List[str]`): One or several question(s) (must be used in conjunction with the `context` argument). context (`str` or `List[str]`): One or several context(s) associated with the question(s) (must be used in conjunction with the `question` argument). topk (`int`, *optional*, defaults to 1): The number of answers to return (will be chosen by order of likelihood). Note that we return less than topk answers if there are not enough options available within the context. doc_stride (`int`, *optional*, defaults to 128): If the context is too long to fit with the question for the model, it will be split in several chunks with some overlap. This argument controls the size of that overlap. max_answer_len (`int`, *optional*, defaults to 15): The maximum length of predicted answers (e.g., only answers with a shorter length are considered). max_seq_len (`int`, *optional*, defaults to 384): The maximum length of the total sentence (context + question) in tokens of each chunk passed to the model. The context will be split in several chunks (using `doc_stride` as overlap) if needed. max_question_len (`int`, *optional*, defaults to 64): The maximum length of the question after tokenization. It will be truncated if needed. handle_impossible_answer (`bool`, *optional*, defaults to `False`): Whether or not we accept impossible as an answer. align_to_words (`bool`, *optional*, defaults to `True`): Attempts to align the answer to real words. Improves quality on space separated langages. Might hurt on non-space-separated languages (like Japanese or Chinese) Return: A `dict` or a list of `dict`: Each result comes as a dictionary with the following keys: - **score** (`float`) -- The probability associated to the answer. - **start** (`int`) -- The character start index of the answer (in the tokenized version of the input). - **end** (`int`) -- The character end index of the answer (in the tokenized version of the input). - **answer** (`str`) -- The answer to the question. rr)rxrLratupler*rvri)rSrfr^examplesryr4r5riWs0 z"QuestionAnsweringPipeline.__call__ do_not_pad@c#st|tr$td|d|dddd}|dkr:t|jjd}|dkrPt|dd}||krntd|d|d|jjst|g|j|||t j d d d }n>|jj d k|jr|j n|j r|j n|j |rd nd ||ddddd td}fddt|D}g}t|D]} d| } dkr8d| nd} dkrRd| nd} |jjdk rtt| |jjkd} | D]}d|| |<q|| }|t| | | || diddddgddd ddqt|D]\}}i}i}|jjddg}|jD]\}}||kr|jdkr^t|}|jtjkrLt|tj }t!|d||<n:|jdkrt"#|}|jt"j kr|$}|%d||<n|||<q|t|dk}||d||VqdS)NrIrJir z`doc_stride` (z ) is larger than `max_seq_len` (rF)rrqZmax_seq_lengthr~Zmax_query_lengthZpadding_strategyZ is_trainingZ tqdm_enabledrightZ only_secondZ only_firstT) textZ text_pairr}Z truncation max_lengthZstrideZreturn_token_type_idsZreturn_overflowing_tokensZreturn_offsets_mappingZreturn_special_tokens_mask input_idscs$g|]}fdd|DqS)csg|]}r|dkndqS)rrr4)rWtok)question_firstr4r5r_szCQuestionAnsweringPipeline.preprocess...)Z sequence_ids)rWZspan_idZencoded_inputsrr4r5r_sz8QuestionAnsweringPipeline.preprocess..rBtoken_type_idsr)rrBrrAencoding cls_indextoken_to_orig_mapZ example_index unique_idZ paragraph_lenZtoken_is_max_contexttokensZstart_positionZ end_positionZ is_impossibleZqas_idrAruptr)exampleis_last)&rLrMr r?rqZmodel_max_lengthrOis_fastr rZ MAX_LENGTH padding_sideZ question_text context_textr*rangeZ cls_token_idr&r-r:appendr remodel_input_names__dict__itemsrsruZconstantZdtypeZint64castZint32r(torchtensorlongZ unsqueeze)rSrr}r~rrfeaturesZ num_spansrAZspan_idxZinput_ids_span_idxZattention_mask_span_idxZtoken_type_ids_span_idxZ cls_indicesrZsubmaskrhfeatureZfw_argsZothersrrTvrrr4rr5 preprocesss             z$QuestionAnsweringPipeline.preprocesscsd}fdd|jjD}|jdkr0|jjn|jj}dt|j krTd|d<|jf|}t |t r|d|d|d S|dd \}}|||d SdS) Nrcsi|]}||qSr4r4)rWrTrgr4r5 sz6QuestionAnsweringPipeline._forward..rZ use_cacheFZ start_logitsZ end_logits)rrrr ) rqrrsrpforwardcallinspect signature parameterskeysrLrM)rSrgrZ model_inputsZ model_forwardoutputrrr4rr5_forwards  z"QuestionAnsweringPipeline._forwardrr8Tc Csd}g}|D]}|d} |d} |d} |d} |dddk rN|dnd} t| | | | ||||\}}}}|jjst| j}t|||D]z\}}}|d}| | t |||kdd t |||kdd  d | j ||||d d qq t|jjd k}|d}|jjdkrP|d|jjk}nd}|r^d nd}t|||D]T\}}}||}||}||||||\}}| | ||| j||d qnq |r| |dddd t|ddddd|}t|d kr|dS|S)Nr7rrrrArBrrr$ r)scorerranswerrrleftrrncSs|dS)Nrr4)xr4r4r5_z7QuestionAnsweringPipeline.postprocess..T)keyreverse)getnumpyrFrqrr&r:Zchar_to_word_offsetrbrr@r;joinZ doc_tokensboolrZ pad_token_idr> get_indicesrsortedr*)rSZ model_outputsrDrEr!rrCZanswersrZstart_end_rrArBr1r2r3Z char_to_wordserrrencoffsetsequence_index start_index end_indexr4r4r5 postprocesssj      z%QuestionAnsweringPipeline.postprocessztokenizers.Encoding)rrrrrr#c Cs|rtz<||}||}|j||dd}|j||dd} Wqtk rp|j|d}|j|d} YqXn|j|d}|j|d} || fS)N)rrr)Z token_to_wordZ word_to_chars Exceptionoffsets) rSrrrrrZ start_wordZend_wordrrr4r4r5rds  z%QuestionAnsweringPipeline.get_indices)rrrr#c Csg}d}}}}t|dD]~\} } |j| } ||krJ|krxnn*||krZ|}||krn|t| }|| g7}||krq|t| 7}|t| d7}q"d|td|tt||dS)a When decoding from token probabilities, this method maps token indexes to actual word in the initial context. Args: text (`str`): The actual context to extract the answer from. start (`int`): The answer starting token index. end (`int`): The answer end token index. Returns: Dictionary like `{'answer': str, 'start': int, 'end': int}` rrr)rrr)resplitrqtokenizer*rr=r?) rSrrrwordsZ token_idxZchar_start_idxZ char_end_idxZ chars_idxrhwordtokenr4r4r5span_to_answervs$      z(QuestionAnsweringPipeline.span_to_answer)NNrn) NNNNNNNNN)rNrN)rFr8T)rjrkrlrmZdefault_input_namesrErrrr rPrw staticmethodrr rRrrirrrintrrrrr __classcell__r4r4ryr5rQsX + 5 u V rQ)r7rFr8)8rrcrcollections.abcrtypingrrrrrrrr&r[r r r rrr Ztokenization_utilsrutilsrrrrrrbaserrrZ get_loggerrjloggerZmodeling_tf_utilsrZmodeling_utilsrZ tokenizersZ tensorflowruZmodels.auto.modeling_tf_autorrrZtorch.utils.dataZmodels.auto.modeling_autorZndarrayrr6rFrGrQr4r4r4r5sN            7 6J