isSSKrSSKrSSKrSSKJrJr SSKJr SSKJ r J r SSK J r J r JrJrJr SSKrSSKJr SSKJrJrJrJr S S KJrJrJr \ (aSS KJr SS KJ r SS K!J"r" SSK#J$r$ \"5(aSSK%J&r& \RN"\(5r)Sr*Sr+\RX"SS55r-"SS\5r."SS\.\5r/"SS\/5r0g)N)ABCabstractmethod) OrderedDict)IterableMapping) TYPE_CHECKINGAnyCallableOptionalUnion)version) TensorTypeis_torch_availableis_vision_availablelogging)ParameterFormat compute_effective_axis_dimension"compute_serialized_parameters_size)PretrainedConfigFeatureExtractionMixinImageProcessingMixinPreTrainedTokenizerBase)Image lc`\rSrSr%Sr\\S'\\S'\\S'Sr \ \\S'Sr \ \\S'S r g) PatchingSpec0a Data class that holds patching specifications. Args: o: Module / object where the op to patch is located name: Name of the op to monkey patch custom_op: Custom op that patches the original op orig_op: Original op that is being patched op_wrapper: Wrapper (optional) that wraps both the original and custom ops. It is useful for ops that are class or static methods for instance. oname custom_opNorig_op op_wrapper) __name__ __module__ __qualname____firstlineno____doc__r __annotations__strr r&r r'__static_attributes__r(h/home/dmtnaga/Documents/work/airagagent/rag_env/lib/python3.13/site-packages/transformers/onnx/config.pyr!r!0s7  F I"&GXh &%)J")r1r!ct\rSrSrSrSrSrSr\R"S5r \ "SSS S .05\ "S SS S .05\ "SSS S .05\ "SS S .SS S .SS S .S .5\ "SSS S .05\ "SSS S .05\ "SS S005\ "SS S .SS S .S.5\ "SS S .SS S .S.5\ "SSSSSS.05\ "SSSS .05\ "SS S005\ "SSS S .05\ "SSS S .05\ "SSS S .05S.r SCSSS\ S\\\4Sjjr\SDSSS\ SS4Sjj5r\\S\\ \\\ 444Sj55r\S\\ \\\ 444Sj5r\S\\\ \44S j5r\S\4S!j5r\S\4S"j5r\S\4S#j5r\S\4S$j5r\S\ 4S%j5r!\S\"4S&j5r#\$S'\S\"4S(j5r%SES)\S*\S+\S,\4S-jjr&SFS)\S.\S/\ S0\4S1jjr'SGS2\(S3S)\S4\S5\S6\"S7\\)S*\S,\S+\S.\S/\ S0\S8\S9S\\ \44S:jjr*S;\\ \4S\\ \44S<jr+S=r,S>r-\S?\ S@\.\S\/\ \44SAj5r0SBr1g)H OnnxConfigEzn Base class for ONNX exportable model describing metadata on how to export the model through the ONNX format. rz1.8logitsbatchsequencerrlast_hidden_state)r8 pred_boxes pred_masksr)r8r=) start_logits end_logits num_labelsheightwidth)rrrdecoder_sequence)z causal-lmdefaultzimage-classificationzimage-segmentationz masked-imz masked-lmmultiple-choicezobject-detectionzquestion-answeringzsemantic-segmentationz seq2seq-lmzsequence-classificationztoken-classificationz vision2seq-lmz speech2seq-lmNconfigrtaskpatching_specscnXlX R;a([USURR535eX l/UlUbUO/H`nUnUR c3[R"U[URUR5S9nUR RU5 Mb g)Nz+ is not a supported task, supported tasks: )r&) _config_tasks_to_common_outputs ValueErrorkeysrI_patching_specsr& dataclassesreplacegetattrr#r$append)selfrHrIrJspec final_specs r2__init__OnnxConfig.__init__ps 44 4&CDDaDaDfDfDhCij  !&4&@NbHDJ||#(00wtvvtyy?YZ  ' ' 3 Ir1returncU"XS9$)z Instantiate a OnnxConfig for a specific model Args: config: The model's configuration to use when exporting to ONNX Returns: OnnxConfig for this model )rIr(clsrHrIs r2from_model_configOnnxConfig.from_model_configs6%%r1c[5e)z Mapping containing the axis definition of the input tensors to provide to the model Returns: For each input: its name associated to the axes symbolic name and the axis position within the tensor )NotImplementedErrorrUs r2inputsOnnxConfig.inputss "##r1c`URURn[R"U5$)z Mapping containing the axis definition of the output tensors to provide to the model Returns: For each output: its name associated to the axes symbolic name and the axis position within the tensor )rMrIcopydeepcopy)rUcommon_outputss r2outputsOnnxConfig.outputss'66tyyA}}^,,r1cB[URS5(aSS0$g)z Dictionary of keys to override in the model's config before exporting Returns: Dictionary with the keys (and their corresponding values) to override use_cacheFN)hasattrrLrbs r2values_overrideOnnxConfig.values_overrides# 4<< - -' 'r1c"[R$)zP The default batch size to use if no other indication Returns: Integer > 0 )r4default_fixed_batchrbs r2default_batch_sizeOnnxConfig.default_batch_sizes---r1c"[R$)zU The default sequence length to use if no other indication Returns: Integer > 0 )r4default_fixed_sequencerbs r2default_sequence_length"OnnxConfig.default_sequence_lengths000r1c"[R$)zW The default number of choices to use if no other indication Returns: Integer > 0 )r4default_fixed_num_choicesrbs r2default_num_choicesOnnxConfig.default_num_choicess333r1c[$)z[ Which onnx opset to use when exporting the model Returns: Integer ONNX Opset version )DEFAULT_ONNX_OPSETrbs r2default_onnx_opsetOnnxConfig.default_onnx_opsets "!r1cg)zx What absolute tolerance value to use during model conversion validation. Returns: Float absolute tolerance value. gh㈵>r(rbs r2atol_for_validationOnnxConfig.atol_for_validationsr1c~[5(a.SSKJn [R"U"55UR :$g)z The minimum PyTorch version required to export the model. Returns: `bool`: Whether the installed version of PyTorch is compatible with the model. r)get_torch_versionF)rtransformers.utilsrr parsetorch_onnx_minimum_version)rUrs r2is_torch_support_available%OnnxConfig.is_torch_support_availables1    <==!2!459X9XX Xr1num_parameterscD[U[R5[:$)z Flag indicating if the model requires using external data format Args: num_parameters: Number of parameter on the model Returns: True if model.num_parameters() * size_of(float32) >= 2Gb False otherwise )rrFloatEXTERNAL_DATA_FORMAT_SIZE_LIMIT)rs r2use_external_data_format#OnnxConfig.use_external_data_formats! /~?T?T U. / r1 batch_size num_channels image_height image_widthc/n[U5Hin[RRX4U5S-nUR [ R "URS55RS55 Mk U$)Nuint8RGB) rangenprandomrandrTr fromarrayastypeconvert)rUrrrrimages_datas r2_generate_dummy_images!OnnxConfig._generate_dummy_imagessaz"A99>>,\JSPD MM%//$++g*>?GGN O# r1 sampling_rate time_duration frequencyc /n[U5Hdn[R"SU[X2-5SS9nUR S[R "S[R -U-U-5-5 Mf U$)NrF)endpointg?r)rrlinspaceintrTsinpi)rUrrrr audio_datarts r2_generate_dummy_audio OnnxConfig._generate_dummy_audiosm z"A A}c-2O.P[`aA   cBFF1ruu9y+@1+D$EE F #r1 preprocessor)rrr seq_length num_choicesis_pair framework tokenizerrcSSKJn SSKJn SSKJn [ UU5(aU b [S5eU b2[R"S[5 [RS5 U n[ UU5(Ga7[U[RSS 9nUR!U5n[U[R"US 9nUR$b%['UR$5S:a UR$OS nS R)U/5U-/U-nUR*S :Xa[U[R,SS 9nUU-nU"UUS 9nUR/5H6unn[1S['U5U5Vs/sH nUUUU-PM snUU'M8 [3UR5US95$[3U"UUS95$[ X5(aUR6SS:wa2[SUR8R:SUR6S35e[U[RS9nUR=X'X5n[3U"UUS95$[ X5(aMUR6SS:Xa:[U[RS9nUR=X'X5n[3U"UUS95$[ X5(aMUR6SS:Xa:[U[RS9nUR?X*X5n[3U"UUS95$[S5es snf)a Generate inputs to provide to the ONNX exporter for the specific framework Args: preprocessor: ([`PreTrainedTokenizerBase`], [`FeatureExtractionMixin`], or [`ImageProcessingMixin`]): The preprocessor associated with this model configuration. batch_size (`int`, *optional*, defaults to -1): The batch size to export the model for (-1 means dynamic axis). num_choices (`int`, *optional*, defaults to -1): The number of candidate answers provided for multiple choice task (-1 means dynamic axis). seq_length (`int`, *optional*, defaults to -1): The sequence length to export the model for (-1 means dynamic axis). is_pair (`bool`, *optional*, defaults to `False`): Indicate if the input is a pair (sentence 1, sentence 2) framework (`TensorType`, *optional*, defaults to `None`): The framework (PyTorch or TensorFlow) that the tokenizer will generate tensors for. num_channels (`int`, *optional*, defaults to 3): The number of channels of the generated images. image_width (`int`, *optional*, defaults to 40): The width of the generated images. image_height (`int`, *optional*, defaults to 40): The height of the generated images. sampling_rate (`int`, *optional* defaults to 22050) The sampling rate for audio data generation. time_duration (`float`, *optional* defaults to 5.0) Total seconds of sampling for audio data generation. frequency (`int`, *optional* defaults to 220) The desired natural frequency of generated audio. Returns: Mapping[str, Tensor] holding the kwargs to provide to the model's forward function rrrrzPYou cannot provide both a tokenizer and a preprocessor to generate dummy inputs.ztThe `tokenizer` argument is deprecated and will be removed in version 5 of Transformers. Use `preprocessor` instead.zROverwriting the `preprocessor` argument with `tokenizer` to generate dummy inputs.r)fixed_dimensionnum_token_to_add0 rG) text_pair) tensor_type)return_tensors pixel_valuesz*The `preprocessor` is an image processor (zC) and expects `model_input_names[0]` to be "pixel_values", but got )r)rrinput_featuresz\Unable to generate dummy inputs for the model. Please provide a tokenizer or a preprocessor.) feature_extraction_utilsrimage_processing_utilsrtokenization_utils_baser isinstancerNwarningswarn FutureWarningloggerwarningrr4rqnum_special_tokens_to_addru unk_tokenlenjoinrIryitemsrdictconvert_to_tensorsmodel_input_names __class__r)rr)rUrrrrrrrrrrrrrrrr token_to_add input_token dummy_inputtokenized_inputkvis r2generate_dummy_inputs OnnxConfig.generate_dummy_inputss` FAE l$; < <AVop p  MM+  NNo p$L l$; < <9J,J,J]^J(AA'JL9J,M,M`lJ !**63|?U?U;VYZ;Z&&  88[M2Z?@:MKyy--?1U1Uhi *K7 ".{k"R+113DAqJOPQSVWXSY[fJg)hJgQ!AK*@Jg)hOA&4O>>9>UVV [KL L  ; ;--a0NB @AWAWA`A`@abMMYMkMklmMnLoq :*V`VtVtuJ55jP\jK K RS S  = =,B`B`abBcguBu9*V`VtVtuJ55jP\jK K RS S | < <A_A_`aAbfvAv:*V`VtVtuJ44ZP]iK [KL Ln 5*is,Lreference_model_inputscU$)a Generate inputs for ONNX Runtime using the reference model inputs. Override this to run inference with seq2seq models which have the encoder and decoder exported as separate ONNX files. Args: reference_model_inputs ([`Mapping[str, Tensor]`): Reference inputs for the model. Returns: `Mapping[str, Tensor]`: The mapping holding the kwargs to provide to the model's forward function r()rUrs r2!generate_dummy_inputs_onnxruntime,OnnxConfig.generate_dummy_inputs_onnxruntimes &%r1cURHXnURc UROURUR5n[URUR U5 MZ gN)rPr'r%setattrr#r$)rUrVr%s r2 patch_opsOnnxConfig.patch_opssJ((D*.//*AtW[WeWeGfI DFFDIIy 1)r1cURHXnURc UROURUR5n[URUR U5 MZ gr)rPr'r&rr#r$)rUrVr&s r2 restore_opsOnnxConfig.restore_opssJ((D&*oo&=dll4??SWS_S_C`G DFFDIIw /)r1r$fieldcSSKJn [URU55VVs0sH upEUSU3U_M snn$s snnf)aK Flatten any potential nested structure expanding the name of the field with the index of the element within the structure. Args: name: The name of the nested structure field: The structure to, potentially, be flattened Returns: (dict[str, Any]): Outputs with flattened structure and key mapping this new structure. r)chain.) itertoolsr enumerate from_iterable)r]r$rridxitems r2"flatten_output_collection_property-OnnxConfig.flatten_output_collection_propertysC $7@ATATUZA[7\]7\)#4&#%7\]]]s<)rLrPrI)rFNrF)rrD(r)r"V@) rrFNrDrrrrrN)2r)r*r+r,r-rqruryr rrrrMr/r listr!rX classmethodr^propertyrrrrcrir rnrrrvrzr~floatrboolr staticmethodrrrr rrrrrrrrr0r(r1r2r4r4Es !!(u!5 (J,G!HI 3Z5PQR +X7z7R,S T)%*5")j9")j9  !(J,G!HI (J,G!HI&1g,'?@'%*5")j9  *$+ ;")j9  "-hG Yafm8n-o!p!8=O-P"QR#.1g,/G#H +X7z7R,S T$hG 0K%LM$hG 0K%LM? Fqu4(4034QYZ^_kZlQm4$ &'9 & &Ua & &$WS#X%6 67$$-gc3h&7!78-- '#s(*;!<  .C..1114S44"C""U D      "fh14HK_bmp  25 NS fi  *.""9=vghvv v  v  vJ'vvvvvvv56v c vp &PSUXPXHY &^efiknfn^o &2 0 ^c^(3-^TXY\^aYaTb^^r1r4c^\rSrSrSSSS\S\\\S\4U4Sjjjr \ SSSS\SS4S jj5r \ S\ \\ \\444U4S jj5r\ S\\ \\44S j5r\ S\4S j5r\ S\4S j5rS SSS\S\S\S\\S\ \\44 U4SjjjrS!S\ \\ \\44S\S\4SjjrSrS\S\\S\\\44U4SjjrSrU=r$)"OnnxConfigWithPastirHrrIrJuse_pastc.>[TU]XUS9 X@lg)N)rIrJ)superrXr)rUrHrIrJrrs r2rXOnnxConfigWithPast.__init__s >J r1rZcU"XSS9$)z Instantiate a OnnxConfig with `use_past` attribute set to True Args: config: The underlying model's config to use when exporting to ONNX Returns: OnnxConfig with `.use_past = True` T)rIrr(r\s r2 with_pastOnnxConfigWithPast.with_pasts6t44r1c^>[TU]nUR(aURUSS9 U$)Nri direction)rrirfill_with_past_key_values_)rUrhrs r2riOnnxConfigWithPast.outputss, ==  + +Ni + Pr1cV[URS5(aSUR0$g)Nrl)rmrLrrbs r2rn"OnnxConfigWithPast.values_overrides% 4<< - -/ /r1cz[URS5(d [S5eURR$)z The number of layers attribute retrieved from the model config. Override this for model configs where the number of layers attribute is not called `num_layers`. num_layerszcould not find the number of layers attribute in the model configuration, override the num_layers property of the model OnnxConfig to solve this)rmrLAttributeErrorrrbs r2rOnnxConfigWithPast.num_layerss: t||\22 B ||&&&r1cz[URS5(d [S5eURR$)z The number of attention heads attribute retrieved from the model config. Override this for model configs where the number of attention heads attribute is not called `num_attention_heads`. num_attention_headszcould not find the number of attention heads attribute in the model configuration, override the num_attention_heads property of the model OnnxConfig to solve this)rmrLrrrbs r2r&OnnxConfigWithPast.num_attention_headss; t||%:;; V ||///r1rrrrrrc 2>[TU]XX4US9nUR(a[5(d [ S5eSSKnUSR upU S-n UURU URRUR-4n SU;a5USRn URUSURXU S9/SS 9US'/US '[UR5H7n US RUR!U 5UR!U 545 M9 U$) NrrrrACannot generate dummy past_keys inputs without PyTorch installed.r input_idsrattention_mask)dtyper)dimpast_key_values)rrrrrNtorchshaperrL hidden_sizercatonesrrrTzeros)rUrrrrr common_inputsrr9seqlenpast_key_values_lengthr mask_dtyperrs r2r(OnnxConfigWithPast.generate_dummy_inputss35 `i6  ==%'' !dee)+6<C]3^ as$/ 0$=DIc9d!F!A3f"56=DIc9d!F!A3f"56 (r1c8USXSUS3'USXSUS3'g)Nrrr1rr2r(rUflattened_outputr$rrs r2_flatten_past_key_values_,OnnxConfigWithPast._flatten_past_key_values_Bs601!63%t,-23A$63%v./r1r$rc>0nUS;a([U5HupEURX1XE5 M U$[TU] X5nU$)N)r/r)rr7rr)rUr$rr6rrrs r2r5OnnxConfigWithPast.flatten_output_collection_propertyFsT 1 1#E*../?sN+   %wI$V r1)r)rFNFrrrFN)F)r)r*r+r,r/r rr!rrXrrrrrrir rnrrrrr r7rrrr0 __classcell__rs@r2rrs7; !"!!!l!34 !  !! 51 5 5Ma 5 5gc3h&7!78'#s(*;!<  'C ' ' 0S 0 0*. (,(( (  ( J' ( c ((Vqve!(gc3h.?)?!@eMPeime27 s 8C= UYZ]_bZbUc  r1rc^\rSrSr\S\\\\\444U4Sjj5r\S\ \S44U4Sjj5r \S\ \S44U4Sjj5r SS\ SS \S \S \ S \ \S\\\44 U4S jjjrS\\\\\44S\4SjrSrSrU=r$)OnnxSeq2SeqConfigWithPastiQrZc>[[U] nUR5H8up#SU;aSOSnUR5HupRSU;aXCU'MX#U'M M: UR(aUR USS9 U$)Nencoderencoder_sequencerEr:rir )rrrirrr )rUrhr$ axes_names sequence_nameaxis_idxrs r2ri!OnnxSeq2SeqConfigWithPast.outputsRs14@ . 4 4 6 D2;t2C.I[M","2"2"4%+8x(,0x( #5!7 ==  + +Ni + Pr1.c >[TU]nX4nU$![aq [URS5(aJ[URS5(a/URR URR 4nU$[S5ef=f)Nencoder_layersdecoder_layerszcould not find the number of encoder and decoder layers attributes in the model configuration, override the num_layers property of the model OnnxConfig to solve this)rrrrmrLrHrI)rUrrs r2r$OnnxSeq2SeqConfigWithPast.num_layerscs +J$1J t||%56674<[TU]nX4nU$![aq [URS5(aJ[URS5(a/URR URR 4nU$[S5ef=f)Nencoder_attention_headsdecoder_attention_headszcould not find the number of attention heads for the encoder and the decoder attributes in the model configuration, override the num_attention_heads property of the model OnnxConfig to solve this)rrrrmrLrMrN)rUrrs r2r-OnnxSeq2SeqConfigWithPast.num_attention_headsss "''"= #6"L #" t||%>??GDLLZsDtDt'+||'K'KT\\MqMq&r##" %  rKrrrrrrc 8>[[U] XX4US9nUR(dUOSn[[U] XXtUS9nUR 5V V s0sH upSU 3U _M nn n [ S 0UDUD6n UR(Ga[ 5(d [S5eSSKn U SRSn U SRSnU SRSnURunnU UUURRU-4nU UUS-URRU-4n/U S 'URunn[UU5n[UU5U- nUU:aS OS n[!U5HWnU S R#U R%U5U R%U5U R%U5U R%U545 MY US :XaUOUn[!UU5H7nU S R#U R%U5U R%U545 M9 U $s sn n f) Nrrdecoder_rrrdecoder_input_idsrDrrAdecoderr()rrrrrrrrNrrrrLr rminmaxrrTr#)rUrrrrrencoder_inputsdecoder_seq_lengthdecoder_inputsr$tensorr$rr9encoder_seq_lengthnum_encoder_attention_headsnum_decoder_attention_heads encoder_shape decoder_shapenum_encoder_layersnum_decoder_layersmin_num_layersmax_num_layersremaining_side_namerrrs r2r/OnnxSeq2SeqConfigWithPast.generate_dummy_inputss`14N `iO  04}}Z!14N 9KhqO IWH\H\H^_H^ HTF+V3H^_@~@@ ===%'' !dee!+.44Q7E!.{!;!A!A!!D !./B!C!I!I!!L GKG_G_ D ')D+" ((,GG M+"Q& ((,GG M02M+ ,59__ 2  2 !35GHN !35GH>YN/ADV/V)\e >*/077 M2 M2 M2 M2 +&9I%EM=E>>:/077U9KU[[Y^M_8`a;e`sHr)r c US;a[SUS35eUS:XaSOSnURupE[XE5n[XE5U- nXE:aSOSnS n US:XaS OS n [ U5H7n S U S .XSU S3'S U S .XSU S3'S U S .XSU S3'S U S .XSU S3'M9 [ Xg5H!n US:XaS U S .n OS U S .n XUSU SUS3'M# g)Nr,r-r.rcrr/rArSpast_encoder_sequencepast_decoder_sequencez past_decoder_sequence + sequencer9r0r .decoder.key.decoder.value .encoder.key.encoder.valuer1)rNrrTrUr) rUr)r r$r_r`rarbrcrBrEr axes_infos r2r 4OnnxSeq2SeqConfigWithPast.fill_with_past_key_values_sL 1 1ST]S^^hij j$-$9 y26./D/D~U+=+RiXa26?86K2Qs~&A?FK[;\ as,7 8AHM]=^ as.9 :?FK[;\ as,7 8AHM]=^ as.9 : ' ~6A"i/ ',<= ',<= IRas!,?+@E F 7r1clUSXSUS3'USXSUS3'USXSUS3'USXSUS 3'g) NrrrhrrirrjrDrkr(r5s r2r73OnnxSeq2SeqConfigWithPast._flatten_past_key_values_sj89!63%|45:;A$63%~6789!63%|45:;A$63%~67r1r(r;)r)r*r+r,rrr/rrituplerrr rrr rr r7r0r<r=s@r2r?r?Qs gc3h&7!78  E#s(O   #U38_ # #$*. C56CC C  C J' C c CCJSGCQTVYQYIZDZ<[ShkS8??r1r?)1rfrQrabcrr collectionsrcollections.abcrrtypingrr r r r numpyr packagingr utilsrrrrrrrconfiguration_utilsrrrrrrrPILr get_loggerr)rr}r dataclassr!r4rr?r(r1r2r|s ##-@@PPhh6A=A   H %#9 ***(v^v^r P SP fW? 2W?r1