^Njh1 PUddlZddlmZmZmZmZddlZddlm Z m Z ddl m Z ddl mZmZddlmZddlmZddlmZmZdd lmZdd lmZmZdd lmZmZed d ddded dedddddeddgZe ee!d<GddeeeZ"GddeeZ#y)N)AnyIterableSequenceType)Encoding Tokenizer)load_tokenizer) NumpyArrayDevice) OnnxProvider)OnnxOutputContext)define_cache_dir iter_batch) LateInteractionTextEmbeddingBase) OnnxTextModelTextEmbeddingWorker)DenseModelDescription ModelSourcezcolbert-ir/colbertv2.0zQText embeddings, Unimodal (text), English, 512 input tokens truncation, 2023 yearmitg)\(?)hfz model.onnx)modeldim descriptionlicense size_in_GBsources model_filez%answerdotai/answerai-colbert-small-v1`zQText embeddings, Unimodal (text), English, 512 input tokens truncation, 2024 yearz apache-2.0gp= ף?zvespa_colbert.onnxsupported_colbert_modelsc8eZdZdZdZdZdZ d'dedede de e fd Z d'd e ee fdede de ee ffd Zd'd eedede deefd ZdedeefdZd eedeefdZ d(dee ezdededede def dZedeefdZdddej6ddddfdededzdedzdeedzdeezdeedzdededzd edzde ffd! Zd)d"Z d*d ee ezded#edzde de e f d$Z dee ezde de e fd%Z!ede"e#e fd&Z$xZ%S)+Colbertz[MASK]outputis_dockwargsreturnc+K|s|jD]}|y|j |j tdt |jD]G\}}t |D]4\}}||j vs||j k(s$d|j||f<6I|xjtj|jdzc_tjj|jddd} tj| d} |xj| zc_t|j|jD]\}} || dk(yw)NzJinput_ids and attention_mask must be provided for document post-processingrr$T)ordaxiskeepdimsg-q=r#) model_output input_idsattention_mask ValueError enumerate skip_list pad_token_idnp expand_dimslinalgnormmaximumzip) selfr&r'r( embeddingitoken_sequencejtoken_idr8 norm_clampedr0s s/Users/ahmed/devFolder/Ultron/claude-voice/.venv/lib/python3.12/site-packages/fastembed/late_interaction/colbert.py_post_process_onnx_outputz!Colbert._post_process_onnx_output.s>#00 1'6+@+@+H `&/v/?/?%@!>#,^#>1XARAR5R67--ad3$=&A   2>>&2G2G#K K 99>>&"5"511t>TD::dE2L   < / -01D1DfF[F[-\) >! 344.]s BECE onnx_inputc *|r |jn |j}tj|dj tj d|d|d<tj|dj tj ddd|d<|S)Nr/r#)r,r0)DOCUMENT_MARKER_TOKEN_IDQUERY_MARKER_TOKEN_IDr5insertastypeint64)r;rDr'r( marker_tokens rB_preprocess_onnx_inputzColbert._preprocess_onnx_inputGs9?t44DD^D^ "$)) { # * *288 4aA# ;(*yy ' ( / / 91aa( #$ documentsc r|r|j|S|jtt|S)N)rN)query)_tokenize_documents_tokenize_querynextiter)r;rNr'r(s rBtokenizezColbert.tokenizeSs@  $ $y $ 9 %%Di,A%B rMrPcZ|jJ|jj|g}|SN)query_tokenizer encode_batch)r;rPencodeds rBrRzColbert._tokenize_queryZs1##///&&33UG<rMc<|jj|}|SrW) tokenizerrY)r;rNrZs rBrQzColbert._tokenize_documents_s..--i8rMFtexts batch_sizeinclude_extensionc t|dr |j|jd}t|tr|gn|}|r |j n |j }|Jt||D]z}|j|D]S} |r|t| jz }t| j} |r|t| |jz }O|| z }U|sm|t|z }||S)Nrr)hasattrrload_onnx_model isinstancestrr\rXrrYsumr0maxMIN_QUERY_LENGTHlen) r;r]r^r'r_r( token_numr\batchtokens attend_counts rB token_countzColbert.token_countcstW%);  " %eS1u&,DNN$2F2F $$$z2E#007V%:%:!;;I#&v'<'<#=L(!St7L7L%MM "\1 8!S 3 rMctS)zLists the supported models. Returns: list[DenseModelDescription]: A list of DenseModelDescription objects containing the model information. )r clss rB_list_supported_modelszColbert._list_supported_modelss ('rMN model_name cache_dirthreads providerscuda device_ids lazy_load device_idspecific_model_pathc  bt ||||fi| ||_||_|j | |_||_||_d|_|||_n |j |j d|_|j||_ tt||_ | |_|j|j|j|j |j|_d|_d|_t)|_d|_|js|j/yy)a Args: model_name (str): The name of the model to use. cache_dir (str, optional): The path to the cache directory. Can be set using the `FASTEMBED_CACHE_PATH` env variable. Defaults to `fastembed_cache` in the system's temp directory. threads (int, optional): The number of threads single onnxruntime session can use. Defaults to None. providers (Optional[Sequence[OnnxProvider]], optional): The list of onnxruntime providers to use. Mutually exclusive with the `cuda` and `device_ids` arguments. Defaults to None. cuda (Union[bool, Device], optional): Whether to use cuda for inference. Mutually exclusive with `providers` Defaults to Device.AUTO. device_ids (Optional[list[int]], optional): The list of device ids to use for data parallel processing in workers. Should be used with `cuda` equals to `True`, `Device.AUTO` or `Device.CUDA`, mutually exclusive with `providers`. Defaults to None. lazy_load (bool, optional): Whether to load the model during class initialization or on demand. Should be set to True when using multiple-gpu and parallel encoding. Defaults to False. device_id (Optional[int], optional): The device id to use for loading the model in the worker process. specific_model_path (Optional[str], optional): The specific path to the onnx model dir if it should be imported from somewhere else Raises: ValueError: If the model_name is not in the format / e.g. BAAI/bge-base-en. Nr)local_files_onlyrz)super__init__rurx_select_exposed_session_options_extra_session_optionsrwrvry_get_model_descriptionmodel_descriptionrdrrs_specific_model_pathdownload_model_local_files_only _model_dir mask_token_idr4setr3rXrb) r;rrrsrtrurvrwrxryrzr( __class__s rBr~zColbert.__init__sH YB6B""&*&J&J6&R#% &*  &DN __ (!__Q/DN!%! 1, data-parallel encoding will be used, recommended for offline encoding of large datasets. If 0, use all available cores. If None, don't use data-parallel processing, use default onnxruntime threading instead. Returns: List of embeddings, one per document ) rrrsrNr^rrurvrwr|rzrN) _embed_documentsrrrdrsrurvrwrrr)r;rNr^rr(s rBembedz Colbert.embeds},)4((  $..)!nn!33 $ 9 9"&"="=    sA;B=B>Bc+Kt|tr|g}t|dr |j|j |D]/}|j |j |gddEd{1y7w)NrF)r')rcrdrarrbrC onnx_embed)r;rPr(texts rB query_embedzColbert.query_embedsv eS !GEtW%);  "D55u5e6   sA)A5+A3,A5ctSrW)ColbertEmbeddingWorkerros rB_get_worker_classzColbert._get_worker_class!s%%rM)T)iTF)r)N)N)&__name__ __module__ __qualname__rGrFrgrr boolrrr rCdictrdrLlistrrUrRrQintrm classmethodrrqr AUTOrr r~rbrrrrr __classcell__)rs@rBr"r"(s J9=5'5155HK5 * 54AE sJ/ 9= PS  c:o   $s) T C TXYaTb ST(^ T#Y4>"' Xc]"      @(t,A'B((!%"37$kk'+ $*.E#E#:E#t E# L)D0 E# Vm E#I$E#E#:E#!4ZE#E#N @# # #&# # * #  # *  # J x}!4  Q[H\ &$'::'F"G&&rMr"c$eZdZdedededefdZy)rrrrsr(r)c td||dd|S)Nr#)rrrsrtr)r")r;rrrsr(s rBinit_embeddingz%ColbertEmbeddingWorker.init_embedding's' !    rMN)rrrrdrr"rrrMrBrr&s$    PW rMr)$rtypingrrrrnumpyr5 tokenizersrr#fastembed.common.preprocessor_utilsr fastembed.common.typesr r fastembed.commonr fastembed.common.onnx_modelr fastembed.common.utilsrr:fastembed.late_interaction.late_interaction_embedding_baserfastembed.text.onnx_text_modelrr"fastembed.common.model_descriptionrrr r__annotations__r"rrrMrBrs 00*>5)9?NQ& g785 gFG'9$45,{&. j0I{&| 0< rM