l^(h9dddlmZddlZddlZddlZddlmZddlmZddlm Z m Z m Z ddl Z ddl Z ddl mZddlmZmZmZmZmZmZddlmZdd lmZej4eZe r erdd lmZdd ZGd d ej@Z!y)) annotationsN)fnmatch)Path) TYPE_CHECKINGAnyCallable)nn) AutoConfig AutoModel AutoTokenizer MT5ConfigPretrainedConfigT5Config)is_peft_available)find_adapter_config_file PeftConfigcdfd }|S)Ncttjt|z dt|z fi|S)NT)exist_ok)osmakedirsr)save_directorykwargs_save_pretrained_fn subfolders f/var/www/skyplay_api_hub/venv/lib/python3.12/site-packages/sentence_transformers/models/Transformer.pywrapperz)_save_pretrained_wrapper..wrappers5 D(94tD"4#7)#CNvNN)rz str | PathreturnNone)rrrs`` r_save_pretrained_wrapperr#sO NrcneZdZUdZdZded< d dfd Z ddZ ddZddZ dd Z dd Z dd Z dd Z dd ZddZddZddZd dZ d! d"dZd#dZd!d$dZed%dZxZS)& TransformeraHugging Face AutoModel to generate token embeddings. Loads the correct class, e.g. BERT / RoBERTa etc. Args: model_name_or_path: Hugging Face models name (https://huggingface.co/models) max_seq_length: Truncate any inputs longer than max_seq_length model_args: Keyword arguments passed to the Hugging Face Transformers model tokenizer_args: Keyword arguments passed to the Hugging Face Transformers tokenizer config_args: Keyword arguments passed to the Hugging Face Transformers config cache_dir: Cache dir for Hugging Face Transformers to store/load models do_lower_case: If true, lowercases the input (independent if the model is cased or not) tokenizer_name_or_path: Name or path of the tokenizer. When None, then model_name_or_path is used backend: Backend used for model inference. Can be `torch`, `onnx`, or `openvino`. Default is `torch`. Tbool save_in_rootc t |ddg|_||_| |_|i}|i}|i}|j ||| |\} } |j || || | fi|| d|vr||d<tj||n|fd|i||_ |t|jdrtt|jjdrTt|jdr>t|jjj|jj}||_|:|jj"j$|jj_yy)Nmax_seq_length do_lower_casemodel_max_length cache_dirconfigmax_position_embeddings)super__init__ config_keysr*backend _load_config _load_modelr from_pretrained tokenizerhasattr auto_modelr-minr.r+r) __class____name__tokenizer_class) selfmodel_name_or_pathr) model_argstokenizer_args config_argsr,r*tokenizer_name_or_pathr2r- is_peft_modelr:s rr0zTransformer.__init__9sk ,o>*  J  !N  K $ 1 12DiQXZe f +VYeZde  %*?!$T__%;%;%S%SUYUcUcUtUt!u, ! -59^^5M5M5V5VDOO " " 2 .rc @t|||jd|jd|jdd Dts td|dk7r t dd d lm}|j|fi|d |id fStj|fi|d |idfS) aLoads the transformers or PEFT configuration Args: model_name_or_path (str): The model name on Hugging Face (e.g. 'sentence-transformers/all-MiniLM-L6-v2') or the path to a local model directory. cache_dir (str | None): The cache directory to store the model configuration. backend (str): The backend used for model inference. Can be `torch`, `onnx`, or `openvino`. config_args (dict[str, Any]): Keyword arguments passed to the Hugging Face Transformers config. Returns: tuple[PretrainedConfig, bool]: The model configuration and a boolean indicating whether the model is a PEFT model. tokenrevisionlocal_files_onlyF)r,rErFrGzgLoading a PEFT model requires installing the `peft` package. You can install it via `pip install peft`.torchaPEFT models can currently only be loaded with the `torch` backend. To use other backends, load the model with `backend="torch"`, call `model[0].auto_model.merge_and_unload()`, save that model with `model.save_pretrained()` and then load the model with the desired backend.rrr,T) rgetr Exception ValueErrorpeftrr5r )r=r>r,r2rArs rr3zTransformer._load_configis %"#!oog.$4!,1CU!K   %&}'! w (-:--.@eKe[degkk k))*<a aW`achhhrc |dk(ri}|r dD]}||vs|j|||<t|tr|j|||fi|nDt|tr|j |||fi|nt j|f||d||_|r|j|||fi||yy|dk(r|j|||fi|y|dk(r|j|||fi|ytd|d) aLoads the transformers or PEFT model into the `auto_model` attribute Args: model_name_or_path (str): The model name on Hugging Face (e.g. 'sentence-transformers/all-MiniLM-L6-v2') or the path to a local model directory. config ("PeftConfig" | PretrainedConfig): The model configuration. cache_dir (str | None): The cache directory to store the model configuration. backend (str): The backend used for model inference. Can be `torch`, `onnx`, or `openvino`. is_peft_model (bool): Whether the model is a PEFT model. model_args (dict[str, Any]): Keyword arguments passed to the Hugging Face Transformers model. rH)rFr-r,onnxopenvinozUnsupported backend 'z6'. `backend` should be `torch`, `onnx`, or `openvino`.N) pop isinstancer_load_t5_modelr _load_mt5_modelr r5r8_load_peft_model_load_onnx_model_load_openvino_modelrK) r=r>r-r,r2rCr?adapter_only_kwargsadapter_only_kwargs rr4zTransformer._load_models<( g #% *6e&)Z7BL..QcBd+,>?e&(+###$6 XZXFI.$$$%7YjY"+";";&#/5#NX#%%%&8&)qzq]pq   !D ! !"4fi V: V  " %D % %&8&) Zz Z4WI=stu urc \ddlm}|j|j|f||d||_y)Nr) PeftModelrN)rLr[r5r8)r=r>r-r,r?r[s rrUzTransformer._load_peft_models5"3)33 OO/ 8>) Wa rc  t|tst|tr td ddlm}ddlm}t|}|j}d} d} |j||||| | \} }| r|jddd |vrh|d } t| tsXt| js td t| d 5} t!j"| |d <dddni|d <|j$|f||| d ||_t)|j&j*|j,|j&_| r|j/||| yy#t$r tdwxYw#1swYxYw)Nz8T5 models are not yet supported by the OpenVINO backend.r)OVModelForFeatureExtraction)OV_XML_FILE_NAMEzUsing the OpenVINO backend requires installing Optimum and OpenVINO. You can install them with pip: `pip install optimum[openvino]`.OpenVINOz openvino*.xml file_name ov_configzXov_config should be a dictionary or a path to a .json file containing an OpenVINO configzutf-8)encodingr-r,export)rRrr rK optimum.intelr]optimum.intel.openvinor^ModuleNotFoundErrorrJrexists_backend_should_exportrQdictopenjsonloadr5r8r#_save_pretrainedr2_backend_warn_to_save)r=r>r-r,r?r]r^ load_pathis_local backend_nametarget_file_globrdrafs rrWz Transformer._load_openvino_models fh ':fi+HWX X  A ?+, ##%! *"88 x-=?OQ]   NN; - * $";/Ii.I--/$r)g6;!.2iilJ{+;;')J{ #8c7R7b7b 8  8  8 ,DDOODdDdfjfrfr+s(   & &'98\ R [# R  8;;s E"E:"E7:Fc  ddl}ddlm}m}|j d|jd|d<t|}|j} d} d} |j|| ||| | \} }| r|j dd|j|f||| d||_ t|jj|j|j_| r|j!|| | yy#t$r t dwxYw) Nr)ONNX_WEIGHTS_NAMEORTModelForFeatureExtractionzUsing the ONNX backend requires installing Optimum and ONNX Runtime. You can install them with pip: `pip install optimum[onnxruntime]` or `pip install optimum[onnxruntime-gpu]`providerONNXz*.onnxr`rc) onnxruntimeoptimum.onnxruntimervrwrgrJrQget_available_providersrrhrir5r8r#rnr2ro) r=r>r-r,r?ortrvrwrprqrrrsrds rrVzTransformer._load_onnx_models,  % [", C@PR^   NN; -9e8T8d8d 9  9  9 ,DDOODdDdfjfrfr+s(   & &'98\ R K# <  s C,,Dc &|jdd}|r||fS|jd|}|jdd} | rt| |jnt|j} | r%t| |j|jn#t|j|j} | r| d|nd|} |r<|j | D cgc]!} | j |j#}} nctj|jd|jdd|jd d }|Dcgc]}t|| s|}}| |v}|s`d|vr\| |v}|rVt|d kDr4d|vr0tjd |d |jd|d| d |j|d<||d<|| }t|j}t|d kDr8|d|d<t|jddg|ddj|d<|rQtjd|d|jd|d|r"tjd|d|d|dd||fScc} wcc}w)a Determines whether the model should be exported to the backend, or if it can be loaded directly. Also update the `file_name` and `subfolder` model_args if necessary. These are the cases: 1. If export is set in model_args, just return export 2. If `/` exists; set export to False 3. If `/` exists; set export to False and set subfolder to the backend (e.g. "onnx") 4. If `` contains a folder, add those folders to the subfolder and set the file_name to the last part We will warn if: 1. The expected file does not exist in the model directory given the optional file_name and subfolder. If there are valid files for this backend, but they're don't align with file_name, then we give a useful warning. 2. Multiple files are found in the model directory that match the target file name and the user did not specify the desired file name via `model_kwargs={"file_name": ""}` Args: load_path: The model repository or directory, as a Path instance is_local: Whether the model is local or remote, i.e. whether load_path is a local directory model_args: The model_args dictionary. Notable keys are "export", "file_name", and "subfolder" target_file_name: The expected file name in the model directory, e.g. "model.onnx" or "openvino_model.xml" target_file_glob: The glob pattern to match the target file name, e.g. "*.onnx" or "openvino*.xml" backend_name: The human-readable name of the backend for use in warnings, e.g. "ONNX" or "OpenVINO" Returns: Tuple[bool, dict[str, Any]]: A tuple of the export boolean and the updated model_args dictionary. rdNr`rz/**/z**/modelrFrE) repo_typerFrEz Multiple z files found in z: z, defaulting to zW. Please specify the desired file name via `model_kwargs={"file_name": ""}`.zNo z found in z. Exporting the model to .z#If you intended to load one of the  zN files, please specify the desired file name via `model_kwargs={"file_name": "rz"}`.)rQrIras_posixr2glob relative_tohuggingface_hublist_repo_filesrlenloggerwarningparts)r=rprqr?target_file_namersrrrdr`rprimary_full_pathsecondary_full_path glob_patternpathmodel_file_names all_filesfname model_foundfile_name_partss rriz"Transformer._backend_should_export0sN$/ :% %NN;0@A NN;5 ENDI6??ATXYbTcTlTlTn DLL) 4 = = ?dllI.779  @I)D)9(:;PSTdSeNf  S\SaSabnSop4 0 0 ; D D Fp p'77""$!# D9 nnWd3 I 4=]%|@\] ] (+;; {*<-1AAK'(1,J1NNN#L>1A)BTBTBVAYY[\l[mm}R~UUrs+/,, ;'*3 ;' >$_Fy///  ! #&5b&9J{ #&*:>>+r+J&b_]`^`Ma&b&k&k&mJ{ #  NNi]*Y-?-?-A,DD]^j]kklm  9:J9K1\N[^^nop^q]rrwy z!!Y q ^s &J J&Jcdd|d}|r |d|dz }n |d|dz }tj|y)NzSaving the exported zA model is heavily recommended to avoid having to export it again.z# Do so with `model.save_pretrained(z)`.z Do so with `model.push_to_hub(z, create_pr=True)`.)rr)r=r>rqrrto_logs rroz!Transformer._backend_warn_to_savesN' ~5vw  ;r-r,r?rs rrSzTransformer._load_t5_models8/=HM98.88  '- FP rc Vddlm}dg|_|j|f||d||_y)rr)MT5EncoderModelrrNN)rrrr5r8)r=r>r-r,r?rs rrTzTransformer._load_mt5_models80>I]:9/99  '- FP rcld|jd|jjjdS)Nz Transformer(z) with Transformer model: r)get_config_dictr8r:r;r=s r__repr__zTransformer.__repr__s7d22455OPTP_P_PiPiPrPrOsstuurc l|jDcic] \}}|dvr||}}}|jd i||ddi}|d}||d<trddlm}t |j|r|jj jrr|jd} |d} tj| |jj j| j} tj| | fd |d<|jjjr d |vr|d |d <|Scc}}w)z#Returns token_embeddings, cls_token) input_idsattention_masktoken_type_ids inputs_embeds return_dictTrtoken_embeddings)PeftModelForFeatureExtractionr)devicer)dim hidden_statesall_layer_embeddingsr")itemsr8rrLrrRactive_peft_configis_prompt_learningsizerHonesnum_virtual_tokensrcatr-output_hidden_states) r=featuresrkeyvaluetrans_featuresoutputsrr batch_sizerprefix_attention_masks rforwardzTransformer.forwards7'nn. UXX J  "$//ONOfO$O"1:'7#$   :4??,IJOO66II-2215 !)*:!;(-  B B U U^l^s^s)%.3YY8M~7^de-f)* ?? ! ! 6 6?g;U/6/GH+ ,; sD0cB|jjjSN)r8r- hidden_sizers rget_word_embedding_dimensionz(Transformer.get_word_embedding_dimensions%%111rc i}t|dtr|g}nt|dtr\g}g|d<|D]L}tt |j \}}|j ||dj |N|g}n7gg} }|D]*} |j | d| j | d,|| g}|D cgc])} | D cgc]} t| jc} +}} } |jr-|D cgc] } | D cgc]} | jc} "}} } |j|j||dd|jd|Scc} wcc} } wcc} wcc} } w)z-Tokenizes a text and maps tokens to token-idsr text_keysr longest_firstpt)padding truncationreturn_tensors max_length) rRstrrjnextiterrappendstripr*lowerupdater6r)) r=textsroutput to_tokenizelookuptext_keytextbatch1batch2 text_tuplecolss rtokenizezTransformer.tokenizesr eAh $ 'K a$ 'K"$F;  5!%d6<<>&:!;$""4({#**84 5'-KFF# -  jm, jm, -"6*KALL41A 4L L   ?JKs3!AGGI3KKK DNN*#..    5L4Ks0 E$ E .E$ E/E* 'E/E$*E/c\|jDcic]}||j|c}Scc}wr)r1__dict__)r=rs rrzTransformer.get_config_dicts*373C3CDCT]]3''DDDs)c>|jj|||jj|ttj j |dd5}tj|j|ddddy#1swYyxYw)N)safe_serializationsentence_bert_config.jsonw)indent) r8save_pretrainedr6rkrrjoinrldumpr)r= output_pathrfOuts rsavezTransformer.savesw '' HZ'[ &&{3 "'',,{,GH# N >RV IId**,d1 = > > >s #'BBcdD]C}tjj||}tjj|sCnt 5}t j |}ddddvrd|dvr|djdd|vrd|dvr|djdd|vrd|dvr|djd|dd|i|S#1swYpxYw)N)rzsentence_roberta_config.jsonzsentence_distilbert_config.jsonzsentence_camembert_config.jsonzsentence_albert_config.jsonz sentence_xlm-roberta_config.jsonzsentence_xlnet_config.jsonr?trust_remote_coder@rAr>r")rrrrhrkrlrmrQ)cls input_path config_namesbert_config_pathfInr-s rrmzTransformer.loads K!# Z E ww~~/0 # $ $YYs^F $ 6 !&9VL=Q&Q < $ $%8 9 v %*=HXAY*Y # $ ( ()< = F "':f]>S'S = ! % %&9 :;j;F;; $ $s CC")NNNNNFNrH)r>rr)z int | Noner?dict[str, Any] | Noner@rrArr, str | Noner*r&rBrr2rr r!) r>rr,rr2rrAdict[str, Any]r z*tuple[PeftConfig | PretrainedConfig, bool]) r>rr-zPeftConfig | PretrainedConfigr,rr2rrCr&r r!)r>rr-rr,rr r!)r>rr-rr,rr r!)rprrqr&r?rrrrsrrrrr ztuple[bool, dict[str, Any]])r>rrqrrrrr r!)r r)rdict[str, torch.Tensor]r r)r int)T)rz.list[str] | list[dict] | list[tuple[str, str]]rz str | boolr r)r r)rrrr&r r!)rrr r%)r; __module__ __qualname____doc__r'__annotations__r0r3r4rUrWrVrirorSrTrrrrrr classmethodrm __classcell__)r:s@rr%r%s-.L$ &*,004-1 $#&*.W.W#.W* .W . .W + .W.W.W!$.W.W .W`(i"%(i2<(iGJ(iYg(i 3(iT-v-v.-v -v  -v  -v -v^ 7S"%7S/?7SLO7S 7Sr,S"%,S/?,SLO,S ,S\c"c"c"# c"  c"  c"c" %c"J  vB2\`&C&NX& &PE><r sg"  // dd=B   8 $&(B<"))B