>:j"SSKJr SSKrSSKrSSKrSSKrSSKrSSKrSSKrSSK r SSK J r J r SSK Jr SSKJr SSKJr SSKJrJr SSKrSSKrSSKJr SSKrSSKJrJr SS KJ r SS KJ!r!J"r" SS KJ#r#J$r$J%r% SS K&J'r'J(r( SS K)J*r+ SSK,J-r- SSK.J/r/ SSK0J1r1J2r2 SSK3J4r4J5r5J6r6 SSK7J8r8J9r9 SSK:J;r;JJ?r? SSK@JArAJBrBJCrCJDrDJErE SSKFJGrG \+R"\I5rJ"SS\"R\?\ 5rLg)) annotationsN)ABCabstractmethod) OrderedDict)Queue)Path)AnyLiteral)CardDataHfApi)version)Tensornn)PreTrainedModelis_datasets_availableis_torch_npu_available)get_class_from_dynamic_moduleget_relative_import_files)logging) __version__) BaseEvaluator)format_modalityinfer_batch_modality)Modality PairInput SingleInput)BaseModelCardDatagenerate_model_card)ModuleRouter Transformer)PeftAdapterMixin)get_device_nameimport_from_string load_dir_pathload_file_pathsave_to_hub_args_decorator)ORIGINAL_TRANSFORMER_MODELSc^\rSrSr%Sr\rSrS\S'0r S\S'Sr S \S 'SASSSSSS SS SSSSSS S .SBU4Sjjjjr SCSjr SDSjr SESjr\SFSj5rSGSjrSHSjrSISjrSJSjrSJSjrSKSjr\SLSj5r\SMSj5rSNSjrSASOSjjrSPSjrSQSjrSRSSSjjrSTS jrSRSSS!jjr SUS"jr!SVSWS#jjr"\#SXSYS$jj5r$SZS[S%jjr%S\S]S&jjr&\'S\S]S'jj5r(S\S]S(jjr)S^S)jr*S_S`S*jjr+SaSbS+jjr,ScS,jr-SASdS-jjr.SASeS.jjr/\SfS/j5r0SASgS0jjr1\ShS1j5r2S2r3\SiS3j5r4\SjS4j5r5\5RlSCS5j5r5\SjS6j5r7\SIS7j5r8\8RlSCS8j5r8\SkS9j5r9\SfS:j5r:\:RlSASlS;jj5r:\SmS<j5r;\SHS=j5r<\SHS>j5r=SnS?jr>S@r?U=r@$)o BaseModel/a Base class for SentenceTransformer, SparseEncoder, and CrossEncoder models. This class provides common functionality for: - Model loading (from Hub, local paths, or creating new models) - Model saving (to disk and Hub) - Device management - Module architecture (sequential composition) - Configuration management - Tokenizer/processor access All models inherit from nn.Sequential and are composed of a sequence of modules that are called sequentially in the forward pass. N str | None default_huggingface_organizationzdict[str, str | None]_default_promptssentence_transformers_model_idstr _model_card_model_id_placeholderFtorch)modulesdevicepromptsdefault_prompt_name cache_foldertrust_remote_coderevisionlocal_files_onlytoken model_kwargsprocessor_kwargs config_kwargsmodel_card_databackendc >[UR5nU(aURU5 UUlXPlXplU=(d UR U S9UlSUl0Ul SUl URRUl XlUc[R "S5nUc"[#5n[$R'SU35 US:Xa-[(R*R-S5b SSKJn U"5 U(a[R2R5U5(dz[R6S :XaS U;dUR9S 5S :a[;S US 35eUR<b/S U;a)UR?5[@;aUR<S U3nU(aURCUU UUUU U U U S9 uo lUbJ[EU[F5(d5[G[IU5VVs/sHunn[KU5U4PM snn5nUb1[MU5S:Xa"[OS[QU5RS35e[RTU]U5 [WUSRY5S5nUb@URZn[]UR_55S SHnURaU5 M URaU5 SUl1URe5 URRgU5 gs snnf)aG Initialize a BaseModel instance. Args: model_name_or_path (str, optional): If a filepath on disk, loads the model from that path. Otherwise, tries to download a pre-trained model. If that fails, tries to construct a model from the Hugging Face Hub with that name. Defaults to None. modules (list[nn.Module], optional): A list of torch modules that are called sequentially. Can be used to create custom models from scratch. Defaults to None. device (str, optional): Device (like ``"cuda"``, ``"cpu"``, ``"mps"``, ``"npu"``) that should be used for computation. If None, checks if a GPU can be used. Defaults to None. prompts (dict[str, str], optional): A dictionary with prompts for the model. The key is the prompt name, the value is the prompt text. The prompt text will be prepended before any text during inference. For example: ``{"query": "query: ", "passage": "passage: "}``. If a model has saved prompts, you can override them by passing your own, or pass ``{"query": "", "document": ""}`` to disable them. Defaults to None. default_prompt_name (str, optional): The name of the prompt that should be used by default. If not set, no prompt will be applied. Defaults to None. cache_folder (str, optional): Path to store models. Can also be set by the ``SENTENCE_TRANSFORMERS_HOME`` environment variable. Defaults to None. trust_remote_code (bool, optional): Whether to allow for custom models defined on the Hub in their own modeling files. Only set to ``True`` for repositories you trust and in which you have read the code, as it will execute code present on the Hub on your local machine. Defaults to False. revision (str, optional): The specific model version to use. It can be a branch name, a tag name, or a commit id, for a stored model on Hugging Face. Defaults to None. local_files_only (bool, optional): Whether to only look at local files (i.e., do not try to download the model). Defaults to False. token (bool or str, optional): Hugging Face authentication token to download private models. Defaults to None. model_kwargs (dict[str, Any], optional): Keyword arguments passed to the underlying Hugging Face Transformers model via ``AutoModel.from_pretrained``. Particularly useful options include: - ``torch_dtype``: Override the default ``torch.dtype`` and load the model under a specific dtype. Can be ``torch.float16``, ``torch.bfloat16``, ``torch.float32``, or ``"auto"`` to use the dtype from the model's ``config.json``. - ``attn_implementation``: The attention implementation to use. For example ``"eager"``, ``"sdpa"``, or ``"flash_attention_2"``. If you ``pip install kernels``, then ``"flash_attention_2"`` should work without having to install ``flash_attn``. It is frequently the fastest option. Defaults to ``"sdpa"`` when available (torch>=2.1.1). - ``device_map``: Device map for model parallelism, e.g. ``"auto"``. - ``provider``: For ``backend="onnx"``, the ONNX execution provider (e.g. ``"CUDAExecutionProvider"``). - ``file_name``: For ``backend="onnx"`` or ``"openvino"``, the filename to load (e.g. for optimized or quantized models). - ``export``: For ``backend="onnx"`` or ``"openvino"``, whether to export the model to the backend format. Also set automatically if the exported file doesn't exist. See the `PreTrainedModel.from_pretrained `_ documentation for more details. Defaults to None. processor_kwargs (dict[str, Any], optional): Keyword arguments passed to the Hugging Face Transformers processor/tokenizer via ``AutoProcessor.from_pretrained``. See the `AutoTokenizer.from_pretrained `_ documentation for more details. Defaults to None. config_kwargs (dict[str, Any], optional): Keyword arguments passed to the Hugging Face Transformers config via ``AutoConfig.from_pretrained``. See the `AutoConfig.from_pretrained `_ documentation for more details. Defaults to None. model_card_data (CardData, optional): A model card data object that contains information about the model. Used to generate a model card when saving the model. If not set, a default model card data object is created. Defaults to None. backend (str, optional): The backend to use for inference. Can be ``"torch"`` (default), ``"onnx"``, or ``"openvino"``. Defaults to ``"torch"``. )r:NSENTENCE_TRANSFORMERS_HOMEzNo device provided, using hpuoptimumr)adapt_transformers_to_gaudi\/zPath z not foundr;r7r9r8r:r<r=r>z$An empty modules list was passed to z@. Please provide at least one module, e.g. a Transformer module.F)4dictr.updater5r6r8model_card_data_classr? module_kwargs_model_card_vars_model_card_text __class____name__ model_typer@osgetenvr#loggerinfo importlibutil find_spec*optimum.habana.transformers.modeling_utilsrEpathexistssepcountFileNotFoundErrorr-lowerr( _load_modules isinstancer enumerater0len ValueErrortypesuper__init__next parametersdtypelistchildrentois_hpu_graph_enabled_validate_promptsregister_model)selfmodel_name_or_pathr3r4r5r6r7r8r9r:r;r<r=r>r?r@default_promptsrEidxmodule first_param first_dtyperPs ]/home/mande/repo/quber/.venv/lib/python3.13/site-packages/sentence_transformers/base/model.pyrhBaseModel.__init__Jsht445   " "7 +& #6 !2.o$2L2L^n2L2o! " $..11  99%ABL >$&F KK4VH= > U?y~~77 BN ^ ' ) bggnn5G&H&H$4+=#=BTBZBZ[^B_bcBc'%0B/C:(NOO55A11&,,.6QQ(,(M(M'NaPbOc%d" *.*<*<")!"3!1)!1++= + 'G'  z';'G'G!SZI["\I[+#vCHf#5I["\]G  3w<1#46tDz7J7J6KLQQ  ! 47--/6  "%++Kt}}/3 +&4 $)!   ++D15#]s;L c URR5HupUbM SURU'M URbXURUR;a>[SURS[ URR 55<S35eURR5VVs/sHup4US:wdM UPM snn=n(a1[ U5n[RSUSUS:aS OSS U35 UR(a$[RS URS 35 ggs snnf) z9Validate prompt configuration and log prompt information.NzDefault prompt name ';' not found in the configured prompts dictionary with keys .zLoaded z promptrHsz with these keys: zDefault prompt name is set to 'zs'. This prompt will be applied to all inference calls, except if a `prompt` or `prompt_name` parameter is provided.) r5itemsr6rerlkeysrdrUrV warning_once)rrkeyvaluekvnon_empty_keysns ryrpBaseModel._validate_promptss8,,,,.JC}$& S!/  # # /D4L4LTXT`T`4`'(@(@'AB((,T\\->->-@(A'DAG  -1LL,>,>,@L,@DAAGa,@L L> LN#A KK'!G1q5Cb+AASTbScd e  # #   1$2J2J1KLEF  $Ms : E Ec TUcJUbURUnU$UR b&URR UR S5nU$Ub[RS5 U$![a5 [SUS[URR 55<S35ef=f)zResolve a prompt from a prompt name or the default prompt name. Args: prompt: An explicit prompt string, or None. prompt_name: A key into ``self.prompts``, or None. Returns: The resolved prompt string, or None if no prompt applies. Nz Prompt name 'r}r~zvProvide either a `prompt`, a `prompt_name`, or neither, but not both. Ignoring the `prompt_name` in favor of `prompt`.) r5KeyErrorrerlrr6getrUwarning)rrprompt prompt_names ry_resolve_promptBaseModel._resolve_prompts >&!\\+6F ))5))$*B*BDI   $ NNC   $' }4optuyvBvBvGvGvIqJpMMNOs A((?B'cUR$)zReturn the backend used for inference, which can be one of "torch", "onnx", or "openvino". Returns: str: The backend used for inference. )r@rrs ry get_backendBaseModel.get_backend s ||c$[USSS/5$)zmReturn the list of modalities supported by this model, e.g. ``["text"]`` or ``["text", "image", "message"]``.r modalitiestextgetattrrs ryrBaseModel.modalities(stAw vh77rc^URmUT;ag[U[5(aST;a[U4SjU55$g)a[Check if the model supports the given modality. A modality is supported if: 1. It is directly listed in :attr:`modalities` (including tuple modalities that are explicitly listed), or 2. It is a tuple of modalities (e.g. ``("image", "text")``) where each part is individually supported and the model also supports ``"message"`` format, which is used to combine multiple modalities into a single input. Args: modality: A single modality string (e.g. ``"text"``, ``"image"``) or a tuple of modality strings (e.g. ``("image", "text")``). Returns: bool: Whether the model supports the given modality. Example:: >>> from sentence_transformers import SentenceTransformer >>> model = SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2") >>> model.supports("text") True >>> model.supports("image") False Tmessagec3,># UH oT;v M g7fN).0part supporteds ry %BaseModel.supports..Ls>XTy(XsF)rrbtupleall)rrmodalityrs @rysupportsBaseModel.supports-sA6OO y  h & &9 +A>X>> >rc2[UR55n[5nU(aUR5up4[ U[ 5(aIUR R5H+nUR[UR555 M- UR(a-X0R;aURURU5 [US5(aURUR5 U(aM[U5$)ae Get the keyword arguments specific to this model for inference methods like `encode` or `predict`. Example: >>> from sentence_transformers import SentenceTransformer, SparseEncoder >>> SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2").get_model_kwargs() [] >>> SentenceTransformer("jinaai/jina-embeddings-v4", trust_remote_code=True).get_model_kwargs() ['task', 'truncate_dim'] >>> SparseEncoder("opensearch-project/opensearch-neural-sparse-encoding-doc-v3-distill").get_model_kwargs() ['task'] Returns: list[str]: A list of keyword arguments for the forward pass. forward_kwargs) rlnamed_childrensetpoprbr sub_modulesvaluesextendrMrKhasattrr)rrr3r module_namerv route_moduless ryget_model_kwargsBaseModel.get_model_kwargsOs"t**,-")++- K&&))%+%7%7%>%>%@MNN4 (D(D(F#GH&A!!k5G5G&G%%d&8&8&EFv/00%%f&;&;<gN##rcUR$)a' .. deprecated:: Use the :attr:`max_seq_length` property instead. Returns the maximal sequence length that the first module of the model accepts. Longer inputs will be truncated. Returns: Optional[int]: The maximal sequence length that the model accepts, or None if it is not defined. max_seq_lengthrs ryget_max_seq_lengthBaseModel.get_max_seq_lengthms"""rcXUR[[UR55$)z4Returns the first module of this sequential embedder)_modulesriiterrs ry _first_moduleBaseModel._first_modulezs}}T$t}}"5677rcXUR[[UR55$)z3Returns the last module of this sequential embedder)rrireversedrs ry _last_moduleBaseModel._last_module~s}}T(4=="9:;;rc|USn[U[5(a UR$[U[5(aURR 5HNn[ [UR55S5n[U[5(aUR(aMN g [UR5$g)zCheck if the first module (Transformer or Router containing Transformers) supports flattened text-only inputs. For Router models, this returns True only if ALL routes support flattening, since we cannot know at this point which route will be used. rNF) rbr!can_flatten_inputsr rrrirrmbool)rr input_moduleroutefirst_in_routes ry_can_flatten_inputsBaseModel._can_flatten_inputss Aw lK 0 022 2 lF + +%1188:!%d5>>+;&>nFgFgFg ;  001 1rc[U5n[R"XRS9nUSUS-S-USSS2'XS- US- S-S2USSS2'U$)aInterleave a largest-to-smallest sorted index array so that each consecutive batch contains a mix of long and short inputs. When text-only inputs are flattened via flash attention, there is no padding. Total tokens per batch equals the sum of actual lengths. Grouping all long inputs together creates peak-memory batches, while grouping all short inputs together under-utilises the GPU. Interleaving (largest, smallest, 2nd largest, 2nd smallest, ...) balances the total token count across batches for more uniform memory usage. )rkNrHr)rdnpemptyrk) sorted_idxr interleaveds ry_interleave_sorted_indices$BaseModel._interleave_sorted_indicessl  Ohhq(8(89 &~!a%A6 ADqD&1uA!|b'@A ADqDrc([U[5(a [U5$[U[[45(a:U(a3[US[ 5(a SUS;d[ SU55$[U[ 5(aSU;a[RUS5$SU;aMUSn[U[5(a [U5$[U[5(a[ SU55$g[ SUR555$[U[5(a1U(a*[US[ 5(a[ S U55$[U[R[R45(a;[U[R5(aUR5$UR$SS KJn [X5(aURSURS-$g![$a gf=f) zEstimate the "size" of an input sample for length-based batch sorting. The exact value doesn't matter, it's only used to group similarly sized inputs together so that padding waste is minimised within each batch. rrolec3L# UHn[RU5v M g7frr* _input_length)rrs ryr*BaseModel._input_length..sFv!922155v"$arraycontentc 3# UH@n[U[5(dM[[UR SS555v MB g7f)rr|N)rbrJrdr0r)ritems ryrrs7l'$U_`dfjUk=s3txx';#<=='s A +A rHc3L# UHn[RU5v M g7frr)rrs ryrrsK?ay..q11?rc3L# UHn[RU5v M g7frr)rmsgs ryrrsFvy..s33vr)Image)rbr0rdrrlrJsumr*rrrndarrayr2rnelementsize PIL.Imager ImportError)samplerPILImages ryrBaseModel._input_lengths fc " "v;  fudm , ,z&)T::vPQ?RFvFFF fd # #&  ..vg??F" +gs++w<'gt,,l'lllK6==?KK K fd # #:fQi3N3NFvFF F frzz5<<8 9 9(265<<(H(H6??$ Yfkk Y  3&++{{1~ A66,    s 4H HHc zUR5Hup40n[U[5(aUnOx/nURbURR U/5nUR 5VVs0sH0upxXv;d$[ US5(dMXtR;dM.Xx_M2 nnnU"U40UD6nM U$s snnf)z.Forward pass through all modules in the model.r)rrbr rMrrrr) rrinputkwargsrrvrMmodule_kwarg_keysrrs ryforwardBaseModel.forwards#'#6#6#8 KM&&)) & $&!%%1(,(:(:(>(>{B(O%'-lln!&4 /GFDT4UZ]avavZvCJ&4! 52M2E$9 !s.B7 B7B7c ^U(d0$Sn[UTRS9nUbTR U5(dSR STR55nS[ U5S[T5RSU3n[U[5(a2[U4SjU55(aUS S R U5S 3- n[U5eTS R"U4S U0UD6nU$![[4a Nf=f![ad U(aCUS:Xa=UVs/sH*n[U[5(aX(S -4USS-OX(-PM, Os snfnnTS R"U40UD6nU$[a U(aCUS:Xa=UVs/sH*n[U[5(aX(S -4USS-OX(-PM, Os snfnnTS R"U40UD6nU$![a TS RU5nU$f=ff=f)a Preprocesses the inputs for the model. Args: inputs (list[SingleInput | PairInput]): A list of inputs to be preprocessed. Each input can be a string, dict, tuple, PIL Image, numpy array, torch Tensor, or other supported modality. If a single input is provided, it must be wrapped in a list. prompt (str, optional): A prompt string to prepend to text inputs. Defaults to None. If the model supports the ``message`` modality, the prompt will be added as a system message to the input messages instead of being prepended to text. Returns: dict[str, Tensor | Any]: A dictionary of tensors with the preprocessed inputs. N)supported_modalities, c38# UHn[U5v M g7fr)r)rms ryr'BaseModel.preprocess.. s!No/!"4"4osz Modality 'z' is not supported by this z model. Supported modalities: c3@># UHoTR;v M g7fr)r)rrrrs ryrrs2`W_t4??3JW_sz This model supports z and zR individually, but not in the same input. Please process each modality separately.rrrrH)rrre TypeErrorrjoinrrfrQrbrr preprocessAttributeErrortokenize) rrinputsrrrrr preprocessedinps ` ryrBaseModel.preprocesssO(I  +FYH   h(?(? !Ndoo!NNI_X677RSWX\S]SfSfRgh))2 5 (E**s2`W_2`/`/`,W\\(-C,DEZZW% % 87--fNVNvNLCI&   * @(f,ntuntgj*S%:P:P6F?,s12w6V\Vbbntuu7--f??L 8(f,ntuntgj*S%:P:P6F?,s12w6V\Vbbntuu 8#Aw//A&A  8#Aw//7  8  8sYC1D1DDH"1EH4H1G?HG!!H?HHHc T[RS5 UR"SSU0UD6$)zI .. deprecated:: `tokenize` is deprecated. Use `preprocess` instead. zEThe `tokenize` method is deprecated, please use `preprocess` instead.rr)rUrr)rrtextsrs ryrBaseModel.tokenize)s* cd6e6v66rc([[4n[5(a SSKJn X#4- n[ X5(ag[ U[R5(a+URS:aURRS;agg![ a Nhf=f)z Check if the input represents a single example or a batch of examples. Args: inputs: The input to check. Returns: bool: True if the input is a single example, False if it is a batch. r)ColumnFrH)UOT) rlrrdatasetsr rrbrrndimrkkind)rrr list_typesr s ryis_singular_inputBaseModel.is_singular_input2sE] " " +i'  f ) ) fbjj ) )fkkQ.>6<(a*XR>;aUR>U =n(aUUS'URAU5 GMG [ [R RUS5SSS9n[R"XhS S9 SSS5 U(aURCXU5 gg!,(df  GN=f![$a U R#U 5 GNf=f!,(df  N]=f)O Saves a model and its configuration files to a directory, so that it can be loaded again. Args: path (str): Path on disk where the model will be saved. model_name (str, optional): Optional model name. create_model_card (bool, optional): If True, create a README.md with basic information about this model. train_datasets (List[str], optional): Optional list with the names of the datasets used to train the model. safe_serialization (bool, optional): If True, save the model using safetensors. If False, save the model the traditional (but unsafe) PyTorch way. NTexist_okzSaving model to !config_sentence_transformers.jsonwutf8encodingr)indent sort_keysr save_in_rootr|_)safe_serializationztransformers_modules.r~r)runamer[rfr modules.json)r)"rSmakedirsrUrV_get_model_configopenr[rjsondumprcrrrr0rfrQsaver __module__ startswithsysr3__file__rr!shutilcopyrsplitbasenamerMappend_create_model_card)rrr[ model_namecreate_model_cardtrain_datasetsr modules_configconfigfOutrur!rv model_path class_ref class_file dest_file needed_file module_configrMs ryr(BaseModel.saveJs& <  D4( &tf-.'') "'',,t%HI3Y_ `dh IIf1 =a#4==1IC!]]40FqWV^<<ATATWW\\$3 WW\\$C3fAVAV0VW KK T 2 ( J NV //I##$;<< [[3<< !,Z0@0E0EF  J2$=Z#HK $Z 0D4E4J4J KIKK Y7$I )s3B78$v,:O:O9PQ (k4<+@+@*AB $'rww?O?OPZ?[enoM!!d.@.@&@W[WiWijnWoFomFo*7 h'  ! !- 0O2R"'',,t^4cF Kt IIn1 5L   # #Dn E aa `" ( J' (8L Ks*-M MN MM=<M= NcUR[[R[RS.URUR S.$)N)sentence_transformers transformerspytorch)rRrr5r6)rRrrBr2r5r6rs ryr$BaseModel._get_model_configs?//)4 , 8 8 ,, ||#'#;#;  rc*URUUUUUS9 g)rr3r4r5r N)r()rrr[r3r4r5r s rysave_pretrainedBaseModel.save_pretraineds$& !/)1  rcURR(afURSURRSUR S3SURRSURRS35nU$)z.Update the default model ID in the model card.zmodel = z("")r?model_idreplacerPrQr1)rr model_cards ry_update_default_model_id"BaseModel._update_default_model_idsy    ( (#++4>>2232d6[6[5\\]^4>>2232d6J6J6S6S5TTUVJrcU(aK[U5nUR5(d+URR(dX RlXRlUR (a8SURR ;aUR nURU5nO [U5n[[R R#US5SSS9nUR%U5 SSS5 g![a/ [RS[R"5S35 gf=f!,(df  g=f) aA Create an automatic model card and store it in the specified path. Args: path (str): The path where the model card will be stored. model_name (Optional[str], optional): The name of the model. Defaults to None. train_datasets (Optional[List[str]], optional): Deprecated argument, ignored. Returns: None generated_from_trainerz#Error while generating model card: zConsider opening an issue on https://github.com/huggingface/sentence-transformers/issues with this traceback. Skipping model card creation.N README.mdrrr)rr\r?rKsave_dirrOtagsrNr ExceptionrUerror traceback format_excr%rSr[rwrite)rrr[r3r5r9rMr8s ryr2BaseModel._create_model_cards  j)J$$&&t/C/C/L/L0:$$-)-%  %=TEYEYE^E^%^..J66zBJ 06 "'',,t[13 HD JJz "I H  :9;O;O;Q:R44   I Hs- C<!D8<6D54D58 Ec 4[RS5 U(afSU;a$[RSUSUS35 USU3nO>;R;R:STW%& "*  **,-9B#5#! +J,,.'%15.'5E_d'O2AD1H. &&@#1'9 !..# '#1=F'9D'% / !/2  KK=j>O>O=PQ R$$ $$$$;/.s !A F55 Gc UUUUUUUU S.n [USUUUUS9n U c9[RSUSURS35 UR"U40U D6$UR UUUUUS9n XR:Xa9[RSURSUS 35 UR "U40U D6$[RS U S US URS 35 UR"U40U DS U 0D6$)NrIr"r;r7r9r:zNo modules.json found for z, initializing a new z model.zLoading z model from r~z Converting z model z to rR)r&rUrVrR_load_default_modules_get_model_type_load_config_modules_load_converted_modules) rrrsr;r7r9r8r:r<r=r> load_kwargsmodules_json_pathmodel_type_being_loadeds ryraBaseModel._load_moduless@( !2 0( 0*  +  %-    $ KK45G4HH]^b^m^m]nnuv w--.@PKP P"&"6"6 %- #7#  #oo 5 KK(4??"3<@R?SSTU V,,-?O;O O k"9!:'BTAUUYZ^ZiZiYjjklm++,>r+rZqrrrc g)a Args: model_name_or_path (str): The name or path of the pre-trained model. token (Optional[Union[bool, str]]): The token to use for the model. cache_folder (Optional[str]): The folder to cache the model. revision (Optional[str], optional): The revision of the model. Defaults to None. trust_remote_code (bool, optional): Whether to trust remote code. Defaults to False. local_files_only (bool, optional): Whether to use only local files. Defaults to False. model_kwargs (Optional[Dict[str, Any]], optional): Additional keyword arguments for the model. Defaults to None. processor_kwargs (Optional[Dict[str, Any]], optional): Additional keyword arguments for the processor/tokenizer. Defaults to None. config_kwargs (Optional[Dict[str, Any]], optional): Additional keyword arguments for the config. Defaults to None. Returns: List[nn.Module]: A list containing the transformer model and the pooling model. Nr) rrrsr;r7r9r8r:r<r=r>s ryr}BaseModel._load_default_modulessrc [USUUUUS9n U b[U SS9n [R"U 5n SSS5 SW ;agSU S;a^[R "U SS5[R "[ 5:a&[RSU SSS [ S 35 URU 5 [US UUUUS9n U b)[U SS9n U R5Ul SSS5 [US UUUUS9n[USS9n [R"U 5nSSS5 [5n[5nWGHnUS nURUXXG5n[R "UR5n[#UR$5S:XGa![R "UR&5n[)UR$5n1SkU-nSS1U:*nU(dU(a[*R,"UUSUUUUUUUU UR.S9 nSSSS.nU(a=U(d6UR15H"unnUU;dMUR3U5UU'M$ SS1nUH$nUU;dM UU;dMUR3U5 M& U"U40UD6nOI[5UUSUUUUS9n URU 5nO&URUUSUUUUUUUU UR.S9 nUUUS'UR7S/5UUS'GM Ucd[9U5n![#U!R:5S:a@[9U5R:Sn"[#U"5S:Xa[=S U"55(aU"nU(dUR>RAXS!9 UU4$!,(df  GNa=f!,(df  GN=f![a GNf=f!,(df  GN=f)"a Loads a full model using the modules.json file. Args: model_name_or_path (str): The name or path of the pre-trained model. token (Optional[Union[bool, str]]): The token to use for the model. cache_folder (Optional[str]): The folder to cache the model. revision (Optional[str], optional): The revision of the model. Defaults to None. trust_remote_code (bool, optional): Whether to trust remote code. Defaults to False. local_files_only (bool, optional): Whether to use only local files. Defaults to False. model_kwargs (Optional[Dict[str, Any]], optional): Additional keyword arguments for the model. Defaults to None. processor_kwargs (Optional[Dict[str, Any]], optional): Additional keyword arguments for the processor/tokenizer. Defaults to None. config_kwargs (Optional[Dict[str, Any]], optional): Additional keyword arguments for the config. Defaults to None. Returns: Tuple[OrderedDict[str, nn.Module], OrderedDict[str, Any]]: An ordered dictionary containing the modules of the model and their kwargs. rr|NrrrrAz:This model was created with Sentence Transformers version z, but you're using version zD. Consider updating to the latest version to avoid potential issues.rRr"rfrH> model_args config_argstokenizer_argsr<r>r[) subfolderr;r7r9r:r8r<r=r>r@rrr)r<r=r>modality_configmodule_output_name)rsrr;r7r9r:r!rr(c3*# UH oS;v M g7f)0123456789abcdefNr)rcs ryr1BaseModel._load_config_modules..s8mZlUV>P9PZls)r9)!r&r%r&loadr parserrUr_parse_model_configreadrOrUr_load_module_class_from_refinspect signaturerdrjrhrr!_load_init_kwargsr@rrr%rrpartsrr?set_base_model)#rrrsr;r7r9r8r:r<r=r>&config_sentence_transformers_json_pathfIn model_configmodel_card_pathrr6r3rMr>r: module_classload_signaturer init_paramsuses_old_namesuses_new_names init_kwargsnew_to_old_name_mappingnew_nameold_nameoptional_paramsrv local_path path_partsrevision_path_parts# ryrBaseModel._load_config_moduless?<2@  /%- 2 . 2 =U"VWZaZgZghsZttPQ]^kQlnERFQGG00;}=AB  $ $\ 2)  %-    & /F;s,/HHJD)< +  %-  #f 5!YYs^N6-# +M%f-I#'#C#C-($L %..|/@/@AN>,,-2#--l.C.CD ")"6"67 !PS^!^"0/!Bk!Q!^"-"?"?*"/"7#%1!))9*;%1)9&3 $ #K$)5,<)6/+ &n2I2O2O2Q.Hh';68C8Q H 53R (9:N&OO$3#{2x{7R'OOH5%4**<L LF "/+="/"7#%1!))9 "J*..z:F &**&+F3!-%%5&7!-%5"/ LL+ .4GM&) *3@3D3DXr3RM-/ 0s,v  /0J:##$)%)*;%<%B%B2%F")*b0S8mZl8m5m5m1H  / /0B / V %%aON4<;  6 5sGN  N1N+N1 O N N.)N1.N11 N?>N? OcURS05R5H4up#X R;dURUbM&X0RU'M6 URcURSS5Ulgg)aParse model configuration and merge saved prompts/defaults with user-provided values. User-provided prompts and default_prompt_name take precedence over saved config values. Saved prompts are only used for keys not already present in ``self.prompts``, or where the current value is ``None`` (i.e. a default placeholder, not yet filled by the user or config). Empty string ``""`` is treated as an intentional user-provided value and will not be overwritten. r5Nr6)rrr5r6)rrrr prompt_texts ryrBaseModel._parse_model_configsu)5(8(8B(G(M(M(O $K,,.$,,{2K2S,7 [))P  # # +'3'7'78Mt'TD $ ,rc 0URUUUUUUUUU S9 $)NrI)r}) rrrsr;r7r9r8r:r<r=r>rRs ryr!BaseModel._load_converted_moduless7)) %/-%-'*  rc [USUUUUS9nUcg[USS9n[R"U5nUR SS5sSSS5 $!,(df  g=f)a Retrieves the model_type from the config_sentence_transformers.json file. This is used to determine whether the model being loaded matches the current class (e.g., a SentenceTransformer model loaded with SentenceTransformer, or a SparseEncoder model loaded with SparseEncoder). When the model type doesn't match, we switch to a converted loading method to ensure compatibility. Defaults to "SentenceTransformer" if the config file is missing or has no "model_type" key, for backwards compatibility with older models. Args: model_name_or_path (str): The name or path of the pre-trained model. token (Optional[Union[bool, str]]): The token to use for the model. cache_folder (Optional[str]): The folder to cache the model. revision (Optional[str], optional): The revision of the model. Defaults to None. local_files_only (bool, optional): Whether to use only local files. Defaults to False. Returns: str: The model type, e.g. "SentenceTransformer", "SparseEncoder", or "CrossEncoder". rr|NSentenceTransformerrrrR)r&r%r&rr) rrrsr;r7r9r:rrr7s ryr~BaseModel._get_model_typesc:2@  /%- 2 . 2 9( 86 JcYYs^F::l,ABK J Js (A Ac,URS5(a [U5$U(d$[RR U5(a(U(aUR SS5OSn[ UUUUS9$[U5$![[4a Nf=f)aw Load a module class from a class reference string. Args: class_ref: The class reference string (e.g., "sentence_transformers.sentence_transformer.modules.Pooling") model_name_or_path: The model name or path trust_remote_code: Whether to trust remote code revision: The model revision model_kwargs: Additional model kwargs Returns: The module class zsentence_transformers. code_revisionN)r9r) r*r$rSr[r\rrOSErrorre)rrr:rsr8r9r<rs ryr%BaseModel._load_module_class_from_refs.    8 9 9%i0 0 /A B BGSL,,_dCY]M 4&%"/ "),, Z(  s) BBBcBUb[R"USS9 U"X5$)a Evaluate the model based on an evaluator Args: evaluator (BaseEvaluator): The evaluator used to evaluate the model. output_path (str, optional): The path where the evaluator can write the results. Defaults to None. Returns: The evaluation results. Tr)rSr#)rr evaluator output_paths ryevaluateBaseModel.evaluate)s#  " KK d 3++rcUR5H.nX LdM [US5(dMURU5 M0 g![a UR5 MPf=f)z,Enable gradient checkpointing for the model.gradient_checkpointing_enableN)r3rrr)rrgradient_checkpointing_kwargsrvs ryr'BaseModel.gradient_checkpointing_enable8sXllnF!gf6U&V&V;889VW%!;88:;sAA"!A"c:UR=nb[US5(a UR$[UR5(aK[USS5(a7[USR S5(aUSR R$[ UR55R$![aZ SSjnURUS9n[ U5nUSRs$![a [R"S5ss$f=ff=f) z Get torch.device from module, assuming that the whole module has one device. In case there are no PyTorch parameters, fall back to CPU. r4r auto_modelcURR5VVs/sH%up[R"U5(dM"X4PM' nnnU$s snnfr)__dict__rr2 is_tensor)rvrrtupless ryfind_tensor_attributes0BaseModel.device..find_tensor_attributesSsB-3__-B-B-D[-DTQXYHZ&1&-D[ \s !AA)get_members_fnrHcpu)rv nn.Modulereturnzlist[tuple[str, Tensor]]) transformers_modelrr4rdrrrirj StopIteration_named_membersr2)rrrrgen first_tuples ryr4BaseModel.deviceBs #'"9"9 9  F7SegoKpKp%,, , t}}  '$q'<"@"@WTRSWM_M_aiEjEj7%%,, , +)*11 1 + %%5K%LC +"3i "1~,,,  +||E** + +s0"B66DC1.D1 DDDDc jUc[RR5(a;[[RR 55Vs/sHnSU3PM nnOe[ 5(a;[[R R 55Vs/sHnSU3PM nnO[RS5 S/S-n[RSSR[[U5535 URS5 UR5 [R"S5nUR!5nUR!5n/nUHKnUR#UR$R&XpXE4S S 9nUR)5 UR+U5 MM XEUS .$s snfs snf) a Starts a multi-process pool to infer with several independent processes. This method is recommended if you want to predict on multiple GPUs or CPUs. It is advised to start only one process per GPU. This method works together with predict and stop_multi_process_pool. Args: target_devices (List[str], optional): PyTorch target devices, e.g. ["cuda:0", "cuda:1", ...], ["npu:0", "npu:1", ...], or ["cpu", "cpu", "cpu", "cpu"]. If target_devices is None and CUDA/NPU is available, then all available CUDA/NPU devices will be used. If target_devices is None and CUDA/NPU is not available, then 4 CPU devices will be used. Returns: Dict[str, Any]: A dictionary with the target processes, an input queue, and an output queue. zcuda:znpu:z1CUDA/NPU is not available. Starting 4 CPU workersrz(Starting multi-process pool on devices: rspawnT)targetargsdaemon)routput processes)r2cuda is_availablerange device_countrnpurUrVrmapr0rn share_memorymp get_contextrProcessrP_multi_process_workerstartr1) rrtarget_devicesictx input_queue output_queuer device_idps rystart_multi_process_pool"BaseModel.start_multi_process_pool^sj&  !zz&&((7tyySR`Ia?b>cde  nnW%iik yy{  'I ~~;;{AA GGI   Q (%9UU7"Y!Vs F+F0cUSHnUR5 M USH#nUR5 UR5 M% USR5 USR5 g)z Stops all processes started with start_multi_process_pool. Args: pool (Dict[str, object]): A dictionary containing the input queue, output queue, and process list. Returns: None rrrN) terminaterclose)poolrs rystop_multi_process_pool!BaseModel.stop_multi_process_poolsak"A KKM#k"A FFH GGI# W  Xrc[S5e)N0This method should be implemented in subclasses.NotImplementedError)rrrrs ry_multi_processBaseModel._multi_processs!"TUUrc[S5e)a>Worker function for multi-process inference. Must be overridden by subclasses. This is called as the target function in each spawned process by :meth:`start_multi_process_pool`. Subclasses should implement this to read from ``input_queue``, run inference on ``target_device``, and write results to ``results_queue``. rr) target_devicemodelr results_queues ryrBaseModel._multi_process_workers""TUUrc USR$)z: Property to get the tokenizer that is used by this model r) tokenizerrs ryrBaseModel.tokenizer Aw   rcXSlg![a' [S[US5RS35Sef=f)zA Property to set the tokenizer that should be used by this model rzThe first module (z() does not have a 'tokenizer' attribute.N)rrrfrQrrrs ryrrsL   %G   $T$q']%;%;$<r%r?zCardData | Noner@$Literal['torch', 'onnx', 'openvino']rNone)rr')rr,rr,rr,)rr&)rzlist[Modality])rrrr)rz list[str])rz int | None)rztorch.nn.Module)rr)r np.ndarrayrr()rint)rdict[str, Tensor]rr*)rzlist[SingleInput | PairInput]rr,rzdict[str, Tensor | Any])rz.list[str] | list[dict] | list[tuple[str, str]]rr*)rr rr)NTNT) r[r0r3r,r4rr5list[str] | Noner rrr')rdict[str, Any])rMr0rr0)N deprecated)r[r0r3r,r5r+rr') NNNTzAdd new model.NFFN)r\r0rbr,r;r,r] bool | Noner rr^r0r_r,rrr`rr5r+rr0) NNTNNFFNNF)r\r0r;r,r]r.r rr^r,r_r,rrr`rr5r+r9r,rjrrr0)NFFNNN)rsr0r;r$r7r,r9r,r8rr:rr<r%r=r%r>r%rDtuple[list[nn.Module] | OrderedDict[str, nn.Module], dict[str, Any]])rr,rr')NFFNNNN)rsr0r;r$r7r,r9r,r8rr:rr<r%r=r%r>r%rRr,rr/)NF) rsr0r;r$r7r,r9r,r:rrr0) r:r0rsr0r8rr9r,r<r%rr)rrrr,rzdict[str, float] | float)rr%rr')rz torch.device)rr+r2dict[Literal['input', 'output', 'processes'], Any])rr0rr') rr0rr*rrrrrr')rr )rzPreTrainedModel | None)r4zint | str | torch.device | Nonerr')rztorch.dtype | None)r\r0rr0)ArQr) __qualname____firstlineno____doc__rrLr-__annotations__r.r1rhrprrpropertyrrrrrrr staticmethodrrrrrrr(r$rGrNr2r'rcrararr}rrrr~rrrr4rrrrrsetterr rrrrkrrrn__static_attributes__ __classcell__)rPs@ryr*r*/sB".37$j7/1+0,L$cL*.c2IM!)-*.#'"'#!&#'.226/3+/8?#c2&c2F c2  c2 ' c2(c2!c2 c2c2c2!c2,c20c2-c2 )!c2"6#c2$ %c2c2J,688 D$< #8<$ //b*">->> ! >@76"&"&+/#' NFNFNF NF ) NF ! NF NF`  "&"&+/#'      )  !    6\h'#'#%/'#HX'# '#R $( ##'.'+#(+/> > !>  >  > ! > > %> > !> )>  >  > F!##'%)'+#(+/#k%k%k% k% ! k% # k%%k%k%!k%)k%k%k% k%d $"'!&.226/30s0s!0s! 0s  0s  0s0s,0s00s-0s N0sd $"'!&.226/3  ! !       , 0 -  N  D $"'!&.226/3w&w&!w&! w&  w&  w&w&,w&0w&-w& Nw&r U( $"'!&.226/3!%  ! !       , 0 -   N < $!& ,C,C!,C! ,C  ,C  ,C ,C\'-'- '- '-  '- , '- '-R ,;++8260V.0V ;0Vd(V V V V V V  V V!!   !! 88'' .    rr*)M __future__rrWrr&rSr-r+rprWabcrr collectionsrmultiprocessingrpathlibrtypingr r numpyrr2torch.multiprocessingrrBhuggingface_hubr r packagingr rrrrr!transformers.dynamic_module_utilsrrtransformers.utilsrtransformers_loggingrAr%sentence_transformers.base.evaluationr#sentence_transformers.base.modalityrr)sentence_transformers.base.modality_typesrrr%sentence_transformers.base.model_cardrr"sentence_transformers.base.modulesrr r!%sentence_transformers.base.peft_mixinr"sentence_transformers.utilr#r$r%r&r'sentence_transformers.util.miscr( get_loggerrQrU Sequentialr*rrryrQs" ##! "+WWf>-?UVVXJJBH  ( ( 2G /Gr