fTh TSrSSKrSSKJrJrJr SSKrSSKrSSKJr SSK J r J r J r SSK Jr SSKJrJrJr SS KJr SS KJrJr S S KJr \R4"\5rS \S\4SjrS,S\\\4S\4Sjjr "SS\RB5r""SS\RF5r$"SS\RB5r%"SS\RB5r&"SS\RB5r'"SS\RB5r("SS \RB5r)"S!S"\RB5r*\"S#S$\55r+\"S%S&\+55r,\"S'S(9"S)S*\+55r-/S+Qr.g)-zPyTorch EfficientNet model.N)OptionalTupleUnion)nn)BCEWithLogitsLossCrossEntropyLossMSELoss)ACT2FN)BaseModelOutputWithNoAttention(BaseModelOutputWithPoolingAndNoAttention$ImageClassifierOutputWithNoAttention)PreTrainedModel)auto_docstringlogging)EfficientNetConfigconfig num_channelscURnXR-n[U[XS- -5U-U-5nUSU-:aX2- n[U5$)z4 Round number of filters based on depth multiplier. g?) depth_divisorwidth_coefficientmaxint)rrdivisornew_dims n/var/www/auris/envauris/lib/python3.13/site-packages/transformers/models/efficientnet/modeling_efficientnet.py round_filtersr's`""G,,,L'3|k9:gEOPG|## w< kernel_sizeadjustc[U[5(aX4nUSS-USS-4nU(aUSS- USUSS- US4$USUSUSUS4$)a. Utility function to get the tuple padding value for the depthwise convolution. Args: kernel_size (`int` or `tuple`): Kernel size of the convolution layers. adjust (`bool`, *optional*, defaults to `True`): Adjusts padding value to apply to right and bottom sides of the input. rrr) isinstancer)r!r"corrects r correct_padr&6s~+s##"0 1~"KNa$78G  Q GAJNGAJGG GAJ GAJ??r cn^\rSrSrSrS\4U4SjjrS\RS\R4Sjr Sr U=r $) EfficientNetEmbeddingsJzD A module that corresponds to the stem module of the original work. rc |>[TU]5 [US5Ul[R "SS9Ul[R"URURSSSSS9Ul [R"URURURS 9Ul [URUlg) N )rrrrpaddingr rvalidFr!strider-bias)epsmomentum)super__init__rout_dimr ZeroPad2dr-Conv2dr convolution BatchNorm2dbatch_norm_epsbatch_norm_momentum batchnormr hidden_act activationselfr __class__s rr5EfficientNetEmbeddings.__init__Os $VR0 ||L9 99   1QPW^c  &:O:OZ`ZtZtu !2!23r pixel_valuesreturncURU5nURU5nURU5nURU5nU$N)r-r9r=r?)rArDfeaturess rforwardEfficientNetEmbeddings.forwardZsA<< -##H->>(+??8,r )r?r=r9r6r-) __name__ __module__ __qualname____firstlineno____doc__rr5torchTensorrI__static_attributes__ __classcell__rBs@rr(r(Js5 41 4ELLU\\r r(c:^\rSrSrSU4SjjrSrU=r$)EfficientNetDepthwiseConv2dcc 8>X-n [T U]UU UUUUUUUS9 g)N) in_channels out_channelsr!r0r-dilationgroupsr1 padding_mode)r4r5) rArYdepth_multiplierr!r0r-r[r1r]rZrBs rr5$EfficientNetDepthwiseConv2d.__init__ds:#5  #%#%  r )rr rrrTzeros)rKrLrMrNr5rRrSrTs@rrVrVcs$   r rVcz^\rSrSrSrS\S\S\S\4U4SjjrS\RS \R4S jr S r U=r $) EfficientNetExpansionLayer}zW This corresponds to the expansion phase of each block in the original implementation. rin_dimr6r0c>[TU]5 [R"UUSSSS9Ul[R "X1R S9Ul[URUl g)NrsameFrYrZr!r-r1) num_featuresr2) r4r5rr8 expand_convr:r; expand_bnr r> expand_act)rArrer6r0rBs rr5#EfficientNetExpansionLayer.__init__sX 99   WBWBWX !2!23r hidden_statesrEclURU5nURU5nURU5nU$rG)rjrkrlrArns rrI"EfficientNetExpansionLayer.forwards4((7 }5  6 r )rlrkrj)rKrLrMrNrOrrr5rP FloatTensorrQrIrRrSrTs@rrcrc}sM 41 43 4 4VY 4U%6%65<<r rcc ~^\rSrSrSrS\S\S\S\S\4 U4SjjrS \ RS \ R4S jr S r U=r$) EfficientNetDepthwiseLayerzc This corresponds to the depthwise convolution phase of each block in the original implementation. rrer0r!adjust_paddingcF>[TU]5 X0lURS:XaSOSn[XES9n[R "US9Ul[X$X6SS9Ul[R"X!RURS9Ul [URUlg) Nrr.rg)r"r,Fr/rir2r3)r4r5r0r&rr7depthwise_conv_padrVdepthwise_convr:r;r<depthwise_normr r> depthwise_act) rArrer0r!rvconv_padr-rBs rr5#EfficientNetDepthwiseLayer.__init__s  "kkQ.7FkA"$,,w"?9 FSX !nn%:%:VE_E_ $F$5$56r rnrEcURS:XaURU5nURU5nURU5nUR U5nU$)Nr)r0ryrzr{r|rps rrI"EfficientNetDepthwiseLayer.forwardsT ;;!  33MBM++M: ++M: **=9 r )r|rzryr{r0rKrLrMrNrOrrboolr5rPrrrQrIrRrSrTs@rrtrts_7"77 7  7  7, U%6%6 5<<  r rtc ~^\rSrSrSrS S\S\S\S\4U4SjjjrS\ RS \ R4S jr S r U=r$) EfficientNetSqueezeExciteLayerzd This corresponds to the Squeeze and Excitement phase of each block in the original implementation. rre expand_dimexpandc>[TU]5 U(aUOUUl[S[ X!R -55Ul[R"SS9Ul [R"URUR SSS9Ul [R"UR URSSS9Ul [URUl[R "5Ulg)Nr) output_sizerg)rYrZr!r-)r4r5dimrrsqueeze_expansion_ratiodim_serAdaptiveAvgPool2dsqueezer8reducerr r> act_reduceSigmoid act_expand)rArrerrrBs rr5'EfficientNetSqueezeExciteLayer.__init__s !':V!S*H*H!HIJ ++: ii   ii    !!2!23**,r rnrEcUnURU5nURU5nURU5nURU5nUR U5n[ R "X!5nU$rG)rrrrrrPmul)rArninputss rrI&EfficientNetSqueezeExciteLayer.forwardsa ]3  M2  6  M2  6  &8 r )rrrrrrr)FrrTs@rrrsR'1'3'C'Y]''* U%6%6 5<<  r rc ^\rSrSrSrS\S\S\S\S\S\4 U4S jjr S \ RS \ RS \ R4S jr SrU=r$)EfficientNetFinalBlockLayerzS This corresponds to the final phase of each block in the original implementation. rrer6r0 drop_rateid_skipc>[TU]5 US:H=(a U(+Ul[R"UUSSSS9Ul[R "X1RURS9Ul [R"US9Ul g)NrrgFrhrxp) r4r5 apply_dropoutrr8 project_convr:r;r< project_bnDropoutdropout)rArrer6r0rrrBs rr5$EfficientNetFinalBlockLayer.__init__sx #q[8[II   .. &;&;fF`F` zzI. r embeddingsrnrEcURU5nURU5nUR(aURU5nX!-nU$rG)rrrr)rArrns rrI#EfficientNetFinalBlockLayer.forwardsE))-8  6    LL7M)6Mr )rrrrrKrLrMrNrOrrfloatrr5rPrrrQrIrRrSrTs@rrrso/(/25/@C/MP/]b/mq/"%"3"3EDUDUZ_ZfZfr rc^\rSrSrSrS\S\S\S\S\S\S \S \S \4U4S jjr S \ RS\ R4Sjr SrU=r$)EfficientNetBlockiaA This corresponds to the expansion and depthwise convolution phase of each block in the original implementation. Args: config ([`EfficientNetConfig`]): Model configuration class. in_dim (`int`): Number of input channels. out_dim (`int`): Number of output channels. stride (`int`): Stride size to be used in convolution layers. expand_ratio (`int`): Expand ratio to set the output dimensions for the expansion and squeeze-excite layers. kernel_size (`int`): Kernel size for the depthwise convolution layer. drop_rate (`float`): Dropout rate to be used in the final phase of each block. id_skip (`bool`): Whether to apply dropout and sum the final hidden states with the input embeddings during the final phase of each block. Set to `True` for the first block of each stage. adjust_padding (`bool`): Whether to apply padding to only right and bottom side of the input kernel before the depthwise convolution operation, set to `True` for inputs with odd input sizes. rrer6r0 expand_ratior!rrrvc p>[T U]5 XPlURS:waSOSUlX%-n UR(a[ XXS9Ul[ UUR(aU OUUUU S9Ul[XXRS9Ul [UUR(aU OUUUUUS9Ul g)NrTF)rrer6r0)rrer0r!rv)rrerr)rrer6r0rr) r4r5rrrc expansionrtrzrsqueeze_exciter projection) rArrer6r0rr!rrrv expand_in_dimrBs rr5EfficientNetBlock.__init__!s ("//14d% - ;;7mDN9$(KK=V#)  =];; 6$(KK=V  r rnrEcUnURS:waURU5nURU5nURU5nUR X!5nU$)Nr)rrrzrr)rArnrs rrIEfficientNetBlock.forwardJsY"    ! NN=9M++M: ++M:  B r )rzrrrrrrrTs@rrrs4' "' '  '  '  ' ' ' ' ' R U%6%6 5<<  r rc v^\rSrSrSrS\4U4SjjrS S\RS\ \ S\ \ S\ 4S jjr S r U=r$) EfficientNetEncoderiWz Forward propagates the embeddings through each EfficientNet block. Args: config ([`EfficientNetConfig`]): Model configuration class. rc>^^[TT]5 UTlURTlU4Sjm[ UR 5n[ U4SjUR55nSn/n[U5Hn[XR U5n[XRU5nURUn URUn URUn [T"URU55Hmn U S:XaSOSn U S:aSOU n U S:aUOUnXAR;aSOSnURU-U- n[!UUUU U U UU US9 nUR#U5 US- nMo M [$R&"U5Tl[$R*"W[US5SS SS 9Tl[$R."UR0UR2UR4S 9TlR:Tlg) Nc\>[[R"TRU-55$rG)rmathceildepth_coefficient)repeatsrAs r round_repeats3EfficientNetEncoder.__init__..round_repeatses"tyy!7!7'!ABC Cr c34># UH nT"U5v M g7frGr`).0nrs r /EfficientNetEncoder.__init__..jsL3Kaq))3KsrTFr) rrer6r0r!rrrrvirgrhrx)r4r5rrlenrYsumnum_block_repeatsrangerrZstrides kernel_sizes expand_ratiosdepthwise_paddingdrop_connect_raterappendr ModuleListblocksr8top_convr: hidden_dimr;r<top_bnr r>top_activation)rArnum_base_blocks num_blockscurr_block_numrirer6r0r!rjrrvrblockrrBs` @rr5EfficientNetEncoder.__init__`s  !'!9!9 Df001L63K3KLL 'A"6+=+=a+@AF#F,?,?,BCG^^A&F --a0K!//2L=)A)A!)DEF"#q&$e!e$%Ev*8.sX$Fq$Fs )last_hidden_statern)rrrrtupler )rArnrrall_hidden_statesrs rrIEfficientNetEncoder.forwards 1E],$[[E!-0M##!%55!!  m4  M2 ++M: X]$FXX X-++  r )rrrrrr)FT)rKrLrMrNrOrr5rPrrrrr rIrRrSrTs@rrrWs\58158t05&*  (( 'tn d^  (   r rc*\rSrSr\rSrSr/rSr Sr g)EfficientNetPreTrainedModeli efficientnetrDc [U[R[R45(akURR R SURRS9 URb%URR R5 gg[U[R5(aJURR R5 URR RS5 gg)zInitialize the weightsg)meanstdNg?) r$rLinearr8weightdatanormal_rinitializer_ranger1zero_ LayerNormfill_)rAmodules r _init_weights)EfficientNetPreTrainedModel._init_weightss fryy"))4 5 5 MM   & &CT[[5R5R & S{{&   &&('  - - KK   " " $ MM   $ $S ).r r`N) rKrLrMrNr config_classbase_model_prefixmain_input_name_no_split_modulesrrRr`r rrrs%L&$O *r rc ^\rSrSrS\4U4Sjjr\S S\\RS\\ S\\ S\ \ \ 44Sjj5rS rU=r$) EfficientNetModelirc>[TU]U5 Xl[U5Ul[ U5UlURS:Xa%[R"URSS9Ul OMURS:Xa%[R"URSS9Ul O[SUR35eUR5 g)NrT) ceil_moderz2config.pooling must be one of ['mean', 'max'] got )r4r5rr(rrencoder pooling_typer AvgPool2drpooler MaxPool2d ValueErrorpooling post_initr@s rr5EfficientNetModel.__init__s   08*62    & (,,v'8'8DIDK  E ),,v'8'8DIDKQRXR`R`Qabc c r rDrrrEcUbUOURRnUbUOURRnUc [S5eUR U5nUR UUUS9nUSnUR U5nURURSS5nU(d Xg4USS-$[UUURS9$)Nz You have to specify pixel_valuesrrrrr)r pooler_outputrn) rruse_return_dictrrrrreshapeshaper rn)rArDrrembedding_outputencoder_outputsr pooled_outputs rrIEfficientNetModel.forwards%9$D $++JjJj &1%> ? " " r rz EfficientNet Model with an image classification head on top (a linear layer on top of the pooled features), e.g. for ImageNet. ) custom_introc^\rSrSrU4Sjr\S S\\RS\\RS\\ S\\ S\ \ \ 44 Sjj5rS rU=r$) "EfficientNetForImageClassificationic>[TU]U5 URUlXl[ U5Ul[ R"URS9Ul URS:a+[ R"URUR5O[ R"5Ul UR5 g)Nrr)r4r5 num_labelsrrrrr dropout_raterrrIdentity classifierrr@s rr5+EfficientNetForImageClassification.__init__s   ++ -f5zzF$7$78 KO??]^K^"))F$5$5tGdfdododq r rDlabelsrrrEcTUbUOURRnURXUS9nU(a UROUSnUR U5nUR U5nSnUGbURR cURS:XaSURlOoURS:aNUR[R:XdUR[R:XaSURlOSURlURR S:XaI[5n URS:Xa&U "UR5UR55nOU "Xr5nOURR S:Xa=[5n U "URSUR5URS55nO,URR S:Xa[!5n U "Xr5nU(dU4USS-n UbU4U -$U $[#UUUR$S 9$) ab labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*): Labels for computing the image classification/regression loss. Indices should be in `[0, ..., config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If `config.num_labels > 1` a classification loss is computed (Cross-Entropy). Nr r regressionsingle_label_classificationmulti_label_classificationr)losslogitsrn)rr rr rr problem_typerdtyperPlongrr rrviewrrrn) rArDrrroutputsrr"r!loss_fctoutputs rrI*EfficientNetForImageClassification.forwards&1%r8sr" )) AA! .,:   H % , C @U3:.@@(RYY2 ")) 44$$N$RYY$N"))BN NbW "))W t*/**&5 35 5 p B )DB  B J er