o 0jkn@sddlZddlmZddlmZddlmZmZmZm Z m Z m Z ddl Z ddlmZddlmZdd lmZdd lmZmZdd lmZd d lmZddlmZmZmZddlmZm Z m!Z!m"Z"ej#edGdddeZ$dS)N)deepcopy)sleep)AnyDictListOptionalTupleUnion)logging)pipeline_requires_extra)MarkDownBatchSampler) HPIConfigPaddlePredictorOption) benchmark) BasePipeline)DocumentResult LatexResultMarkdownResult)split_original_textssplit_text_recursivetranslate_code_blocktranslate_html_blockZtranscAseZdZdZdgZdddddddddedeedeed eeeefd ee d e d ee eeefe fd e ddffddZ ddZdeddfddZdeddfddZdXddZ                              dYde eeeejeejfdee dee dee dee dee d ee d!ee d"ee d#ee eefd$ee d%ee eeeefefd&eed'eed(eed)eed*eed+eed,eed-eed.eed/eed0eed1eed2eed3e d4e d5e d6e d7e d8e def@d9d:Zd;d<Zd=d>Z ? @       A dZdBeedCedDedEedFedGedHedIedJedKedLefdMdNZdOedefdPdQZdRedefdSdTZ dUedefdVdWZ!Z"S)[PP_DocTranslation_Pipelinez% PP_ DocTranslation_Pipeline zPP-DocTranslationNF)deviceengine engine_config pp_optionuse_hpip hpi_configinitial_predictorconfigrrrr r!r"r#returnc sntjd||||||d| |d|_||_|dd|_d|_d|_|r1||| |t |_ dS)aInitializes the PP-DocTranslation pipeline. Args: config (Dict): Configuration dictionary containing various settings. device (Optional[str], optional): The device to use for prediction. Defaults to `None`. engine (Optional[str], optional): Inference engine. Defaults to `None`. engine_config (Optional[Dict[str, Any]], optional): Engine-specific config. Defaults to `None`. pp_option (Optional[PaddlePredictorOption], optional): Paddle predictor options. Defaults to `None`. use_hpip (bool, optional): Whether to use HPIP. Defaults to `False`. hpi_config (Optional[Union[Dict[str, Any], HPIConfig]], optional): HPIP configuration. Defaults to `None`. initial_predictor (bool, optional): Whether to initialize the predictor. Defaults to `False`. )rrrr r!r" pipeline_nameuse_layout_parserTN) super__init__r&r$getr'layout_parsing_pipelinechat_botinintial_visual_predictorinintial_chat_predictorrmarkdown_batch_sampler) selfr$rrrr r!r"r#kwargs __class__r(w/var/www/html/Deteccion_Ine/venv/lib/python3.10/site-packages/paddlex/inference/pipelines/pp_doctranslation/pipeline.pyr*.s&    z#PP_DocTranslation_Pipeline.__init__cCs|jdur |jdSdS)N)r,close)r1r(r(r5r6as z PP_DocTranslation_Pipeline.closecCs<|dd|_|jr|didddi}|||_dS)a Initializes the visual predictor with the given configuration. Args: config (dict): The configuration dictionary containing the necessary parameters for initializing the predictor. Returns: None r'TZ SubPipelinesZ LayoutParserZpipeline_config_errorz)config error for layout_parsing_pipeline!N)r+r'Zcreate_pipeliner,)r1r$Zlayout_parsing_configr(r(r5r.es   z4PP_DocTranslation_Pipeline.inintial_visual_predictorcCshddlm}|didddi}|||_ddlm}|didid d d i}|||_d S) a Initializes the chat predictor with the given configuration. Args: config (dict): The configuration dictionary containing the necessary parameters for initializing the predictor. Returns: None rcreate_chat_botZ SubModulesZLLM_ChatZchat_bot_config_errorzconfig error for llm chat bot!)create_prompt_engineeringZPromptEngneeringZTranslate_CommonTextZpe_config_errorz%config error for translate_pe_config!N)r8r+r-r9 translate_pe)r1r$r8chat_bot_configr9Ztranslate_pe_configr(r(r5r/ys     z2PP_DocTranslation_Pipeline.inintial_chat_predictorcOstddS)NzPP-Translation Pipeline do not support to call `predict()` directly! Please invoke `visual_predict`, `build_vector`, `chat` sequentially to obtain the result.)r error)r1argsr2r(r(r5predictsz"PP_DocTranslation_Pipeline.predictTinputuse_doc_orientation_classifyuse_doc_unwarpinguse_textline_orientationuse_seal_recognitionuse_table_recognitionuse_formula_recognitionuse_chart_recognitionuse_region_detectionlayout_threshold layout_nmslayout_unclip_ratiolayout_merge_bboxes_modetext_det_limit_side_lentext_det_limit_typetext_det_threshtext_det_box_threshtext_det_unclip_ratiotext_rec_score_threshseal_det_limit_side_lenseal_det_limit_typeseal_det_threshseal_det_box_threshseal_det_unclip_ratioseal_rec_score_thresh#use_wired_table_cells_trans_to_html&use_wireless_table_cells_trans_to_htmluse_table_orientation_classify use_ocr_results_with_table_cellsuse_e2e_wired_table_rec_model use_e2e_wireless_table_rec_modelc #ks|jdkrtdddiV|jdur td||j|jj|fid|d|d|d |d |d |d |d | d| d| d| d| d|d|d|d|d|d|d|d|d|d|d|d|d|d|d |d!|d"|d#|D] }!d$|!i}"|"VqdS)%a This function takes an input image or a list of images and performs various visual prediction tasks such as document orientation classification, document unwarping, general OCR, seal recognition, and table recognition based on the provided flags. Args: input (Union[str, list[str], np.ndarray, list[np.ndarray]]): Input image path, list of image paths, numpy array of an image, or list of numpy arrays. use_doc_orientation_classify (Optional[bool]): Whether to use document orientation classification. use_doc_unwarping (Optional[bool]): Whether to use document unwarping. use_textline_orientation (Optional[bool]): Whether to use textline orientation prediction. use_seal_recognition (Optional[bool]): Whether to use seal recognition. use_table_recognition (Optional[bool]): Whether to use table recognition. use_formula_recognition (Optional[bool]): Whether to use formula recognition. use_region_detection (Optional[bool]): Whether to use region detection. layout_threshold (Optional[float]): The threshold value to filter out low-confidence predictions. Default is None. layout_nms (Optional[bool], optional): Whether to use layout-aware NMS. Defaults to `False`. layout_unclip_ratio (Optional[Union[float, Tuple[float, float]]], optional): The ratio of unclipping the bounding box. Defaults to `None`. If it's a single number, then both width and height are used. If it's a tuple of two numbers, then they are used separately for width and height respectively. If it's None, then no unclipping will be performed. layout_merge_bboxes_mode (Optional[str], optional): The mode for merging bounding boxes. Defaults to `None`. text_det_limit_side_len (Optional[int]): Maximum side length for text detection. text_det_limit_type (Optional[str]): Type of limit to apply for text detection. text_det_thresh (Optional[float]): Threshold for text detection. text_det_box_thresh (Optional[float]): Threshold for text detection boxes. text_det_unclip_ratio (Optional[float]): Ratio for unclipping text detection boxes. text_rec_score_thresh (Optional[float]): Score threshold for text recognition. seal_det_limit_side_len (Optional[int]): Maximum side length for seal detection. seal_det_limit_type (Optional[str]): Type of limit to apply for seal detection. seal_det_thresh (Optional[float]): Threshold for seal detection. seal_det_box_thresh (Optional[float]): Threshold for seal detection boxes. seal_det_unclip_ratio (Optional[float]): Ratio for unclipping seal detection boxes. seal_rec_score_thresh (Optional[float]): Score threshold for seal recognition. use_wired_table_cells_trans_to_html (bool): Whether to use wired table cells trans to HTML. use_wireless_table_cells_trans_to_html (bool): Whether to use wireless table cells trans to HTML. use_table_orientation_classify (bool): Whether to use table orientation classification. use_ocr_results_with_table_cells (bool): Whether to use OCR results processed by table cells. use_e2e_wired_table_rec_model (bool): Whether to use end-to-end wired table recognition model. use_e2e_wireless_table_rec_model (bool): Whether to use end-to-end wireless table recognition model. **kwargs (Any): Additional settings to extend functionality. Returns: dict: A dictionary containing the layout parsing result. Fz1The models for layout parser are not initialized.r=NzGThe layout parsing pipeline is not initialized, will initialize it now.rArBrCrDrErFrGrHrIrJrKrLrMrNrOrPrQrRrVrSrTrUrWrXrYrZr[r\r]r^layout_parsing_result)r'r r=r,warningr.r$r?)#r1r@rArBrCrDrErFrGrHrIrJrKrLrMrNrOrPrQrRrSrTrUrVrWrXrYrZr[r\r]r^r2r_Zvisual_predict_resr(r(r5visual_predicts Q         #z)PP_DocTranslation_Pipeline.visual_predictcCsJg}|j|D]}|jd}|jd}|d|dd}|t|q|S)NrTT) input_path page_indexmarkdown_textspage_continuation_flags)r0sampleZ instancesZ input_pathsappendr)r1r@Zmarkdown_info_listZmarkdown_sampleZmarkdown_contentrcZ markdown_infor(r(r5load_from_markdown s  z-PP_DocTranslation_Pipeline.load_from_markdownc Csg}d}tdt|dtdt|D]\}}|\}} |dkrK|r3|||d}td|ddt|d t| |||qt| |kr|d krt|t| |krf|d | 7}q|rtd|ddt|d |||| }qtd|ddt|d |r|||d}|d kr|t| ||q|d ks|d krt| |||qt d||r|||d |S)a Chunks the given markdown blocks into smaller chunks of size `chunk_size` and translates them using the given translate function. Args: md_blocks (list): A list of tuples representing each block of markdown content. Each tuple consists of a string indicating the block type ('text', 'code') and the actual content of the block. chunk_size (int): The maximum size of each chunk. translate_func (callable): A callable that accepts a string argument and returns the translated version of that string. Returns: str: A string containing all the translated chunks concatenated together with newlines between them. r:zSplit the original text into z blockszStarting translation...codezTranslating block r/z...text Ztext_with_htmlhtmlzUnknown block type: ) r infolen enumeratestriprhrrr ValueErrorjoin) r1 md_blocks chunk_sizetranslate_funcZtranslation_resultschunkidxblockZ block_typeZ block_contentr(r(r5chunk_translate/sJ      z*PP_DocTranslation_Pipeline.chunk_translatezh ori_md_info_listtarget_languagervtask_description output_format rules_strfew_shot_demo_text_contentfew_shot_demo_key_value_listglossaryllm_request_intervalr<c  +sjdurtdj| dur ddlm} | | njt|tr7|d ddur7 |g}tt s@t t| t sM| dusMJdd}| durp| D]\}}t|tred |}||d |d 7}qW|dkrdur{|nd 7|7f d d }jjdd}t|}||kr||}n td|d|d|D]"}|d}t|}|||}t|d|d|d|dVqdS)a Translate the given original text into the specified target language using the configured translation model. Args: ori_md_info_list (List[Dict]): A list of dictionaries containing information about the original markdown text to be translated. target_language (str, optional): The desired target language code. Defaults to "zh". chunk_size (int, optional): The maximum number of characters allowed per chunk when splitting long texts. Defaults to 5000. task_description (str, optional): A description of the task being performed by the translation model. Defaults to `None`. output_format (str, optional): The desired output format of the translation result. Defaults to `None`. rules_str (str, optional): Rules or guidelines for the translation model to follow. Defaults to `None`. few_shot_demo_text_content (str, optional): Demo text content for the translation model. Defaults to `None`. few_shot_demo_key_value_list (str, optional): Demo text key-value list for the translation model. Defaults to `None`. glossary (Dict, optional): A dictionary containing terms and their corresponding definitions. Defaults to `None`. llm_request_interval (float, optional): The interval in seconds between each request to the LLM. Defaults to 0.0. chat_bot_config (Dict, optional): Configuration for the chat bot used in the translation process. Defaults to `None`. **kwargs: Additional keyword arguments passed to the translation model. Yields: MarkdownResult: A dictionary containing the translation result in the target language. Nz;The LLM chat bot is not initialized,will initialize it now.rr7rrdzglossary must be a dictr:u或z:  c sjtjj|d}j|ddd}d|vr#td|dur+td|dd}|S) z Translate the given text using the configured translation model. Args: text (str): The text to be translated. Returns: str: The translated text in the target language. Z original_textlanguagerrrrr)promptcontentr:z<>zThe translation did not reach the end. This may happen if your chunk_size is too large. Please reduce chunk_size and try again.Nz#The call to the large model failed.)rr;generate_promptZgenerate_chat_resultsr+ Exceptionreplacerstrip)rlr translate r-rrrrrr1rrr(r5rws&  z.translate_funcrz:Chunk size should be greater than the base prompt length (z ), but got .rercrf)rrcrdrfre)r-r r`r/r$r:r8 isinstancelistr+concatenate_markdown_pagesfloatdictitemsrtr;rrprsrr{r)r1rrrvrrrrrrrr<r2r8Z glossary_strkvrwZbase_prompt_contentZbase_prompt_lengthZori_mdZoriginal_textsruZtarget_language_textsr(rr5rhsv #          z$PP_DocTranslation_Pipeline.translate markdown_listc Csd}d}t|dkrtd|D]X}|dd}|dd}|s^|s^|r(|dnd}|dr4|ddnd}|r>td |nd } |rHtd |nd } | sW| sW|d |d7}n||d7}n|d |d7}|}q|dd dd|d} t| S)a Concatenate Markdown content from multiple pages into a single document. Args: markdown_list (list): A list containing Markdown data for each page. Returns: tuple: A tuple containing the processed Markdown text. r:Trz$The length of markdown_list is zero.rfrrez[\u4e00-\u9fff]F rmrcNrb)rcrdrfre)rprsrematchr) r1rreZ-previous_page_last_element_paragraph_end_flagresZ'page_first_element_paragraph_start_flagZ$page_last_element_paragraph_end_flagZlast_char_of_markdownZfirst_char_of_handlerZlast_is_chinese_charZfirst_is_chinese_charZconcatenate_resultr(r(r5rsP   z5PP_DocTranslation_Pipeline.concatenate_markdown_pages word_listc Cst|dkr tdg}g}t|D]$\}}|dD]}t|}||d<||q|dD]}||q.qt||dd|dS)a Concatenate Word content from multiple pages into a single document. Args: word_list (list): A list containing Word data for each page. Returns: tuple: A tuple containing the processed Word document. rz The length of word_list is zero. word_blocksrdimagesrc)rrcr)rprsrqrrhr) r1r merged_blocksimagepage_idx page_blocksrz block_copyimg_objr(r(r5concatenate_word_pages7s$      z1PP_DocTranslation_Pipeline.concatenate_word_pageslatex_info_listc Cst|dkr tdg}g}t|D]$\}}|dD]}t|}||d<||q|dD]}||q.qt|||dddS)a  Concatenate LaTeX content from multiple pages into a single document. Args: latex_info_list (list): A list containing LaTeX data for each page. Returns: tuple: A tuple containing the processed LaTeX document. rz&The length of latex_info_list is zero. latex_blocksrdrrc)rrrc)rprsrqrrhr) r1rrZ merged_imagesrrrzrrr(r(r5concatenate_latex_pagesWs$      z2PP_DocTranslation_Pipeline.concatenate_latex_pages)r%N)NNNNNNNNNNNNNNNNNNNNNNNNFFTTFT) r|r}NNNNNNr~N)#__name__ __module__ __qualname____doc__entitiesrrstrrrboolr rr*r6rr.r/r?rnpZndarrayrrintrarir{rrtuplerrr __classcell__r(r(r3r5r%sP   3        " <    A r)%rcopyrtimertypingrrrrrr numpyrutilsr Z utils.depsr Zcommon.batch_samplerrmodelsrrZutils.benchmarkrbaserresultrrrrrrrZ time_methodsrr(r(r(r5s