a hzS@sdZddlZddlmZmZddlZddlZddlmZddlm Z m Z m Z ddl m Z ddlmZmZmZdd lmZdd lmZmZd d lmZeeZeed ddZd-eeefedddZ Gdddej!Z"Gdddej#Z$Gdddej!Z%Gdddej!Z&Gdddej!Z'Gdddej!Z(Gd d!d!ej!Z)Gd"d#d#ej!Z*eGd$d%d%eZ+eGd&d'd'e+Z,ed(d)Gd*d+d+e+Z-gd,Z.dS).zPyTorch EfficientNet model.N)OptionalUnion)nn)BCEWithLogitsLossCrossEntropyLossMSELoss)ACT2FN)BaseModelOutputWithNoAttention(BaseModelOutputWithPoolingAndNoAttention$ImageClassifierOutputWithNoAttention)PreTrainedModel)auto_docstringlogging)EfficientNetConfig)config num_channelscCsJ|j}||j9}t|t||d||}|d|krB||7}t|S)z< Round number of filters based on depth multiplier. g?)Z depth_divisorZwidth_coefficientmaxint)rrZdivisorZnew_dimrr/var/www/html/assistant/venv/lib/python3.9/site-packages/transformers/models/efficientnet/modeling_efficientnet.py round_filters's   rT) kernel_sizeadjustcCsrt|tr||f}|dd|ddf}|rR|dd|d|dd|dfS|d|d|d|dfSdS)aJ 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. rrrN) isinstancer)rrZcorrectrrr correct_pad6s $rcs:eZdZdZedfdd ZejejdddZZ S)EfficientNetEmbeddingszL A module that corresponds to the stem module of the original work. rcshtt|d|_tjdd|_tj|j|jddddd|_ tj |j|j |j d |_ t|j|_dS) N )rrrrpaddingrrvalidFrstrider"bias)epsmomentum)super__init__rout_dimr ZeroPad2dr"Conv2dr convolution BatchNorm2dbatch_norm_epsbatch_norm_momentum batchnormr hidden_act activationselfr __class__rrr*Os  zEfficientNetEmbeddings.__init__) pixel_valuesreturncCs,||}||}||}||}|SN)r"r.r2r4)r6r9featuresrrrforwardZs     zEfficientNetEmbeddings.forward) __name__ __module__ __qualname____doc__rr*torchTensorr= __classcell__rrr7rrJs rcseZdZdfdd ZZS) EfficientNetDepthwiseConv2drrrTzerosc s*||} tj|| |||||||d dS)N) in_channels out_channelsrr%r"dilationgroupsr& padding_mode)r)r*) r6rGZdepth_multiplierrr%r"rIr&rKrHr7rrr*ds z$EfficientNetDepthwiseConv2d.__init__)rrrrrTrF)r>r?r@r*rDrrr7rrEcsrEcs@eZdZdZeeeedfdd Zejej dddZ Z S)EfficientNetExpansionLayerz_ This corresponds to the expansion phase of each block in the original implementation. rin_dimr+r%csBttj||dddd|_tj||jd|_t|j |_ dS)NrsameFrGrHrr"r&) num_featuresr') r)r*rr- expand_convr/r0 expand_bnr r3 expand_act)r6rrNr+r%r7rrr*s z#EfficientNetExpansionLayer.__init__ hidden_statesr:cCs"||}||}||}|Sr;)rRrSrTr6rVrrrr=s   z"EfficientNetExpansionLayer.forward) r>r?r@rArrr*rB FloatTensorrCr=rDrrr7rrL}s rLcsBeZdZdZeeeeedfdd Zej ej dddZ Z S)EfficientNetDepthwiseLayerzk This corresponds to the depthwise convolution phase of each block in the original implementation. rrNr%radjust_paddingcsvt||_|jdkrdnd}t||d}tj|d|_t||||dd|_tj ||j |j d|_ t |j|_dS) Nrr#rO)rr!Fr$rQr'r()r)r*r%rrr,depthwise_conv_padrEdepthwise_convr/r0r1depthwise_normr r3 depthwise_act)r6rrNr%rr[Zconv_padr"r7rrr*s    z#EfficientNetDepthwiseLayer.__init__rUcCs6|jdkr||}||}||}||}|S)Nr)r%r]r^r_r`rWrrrr=s      z"EfficientNetDepthwiseLayer.forward r>r?r@rArrboolr*rBrXrCr=rDrrr7rrYsrYcsBeZdZdZd eeeedfdd Zej ej dddZ Z S) EfficientNetSqueezeExciteLayerzl This corresponds to the Squeeze and Excitement phase of each block in the original implementation. FrrN expand_dimexpandcst|r|n||_tdt||j|_tjdd|_ tj |j|jddd|_ tj |j|jddd|_ t |j|_t|_dS)Nr)Z output_sizerO)rGrHrr")r)r*dimrrZsqueeze_expansion_ratioZdim_serZAdaptiveAvgPool2dsqueezer-reducerfr r3 act_reduceZSigmoid act_expand)r6rrNrerfr7rrr*s$  z'EfficientNetSqueezeExciteLayer.__init__rUcCsF|}||}||}||}||}||}t||}|Sr;)rhrirjrfrkrBmul)r6rVinputsrrrr=s      z&EfficientNetSqueezeExciteLayer.forward)Frarrr7rrcsrccsHeZdZdZeeeeeedfdd Ze j e j e j dddZ Z S)EfficientNetFinalBlockLayerz[ This corresponds to the final phase of each block in the original implementation. rrNr+r% drop_rateid_skipcsXt|dko| |_tj||dddd|_tj||j|jd|_ tj |d|_ dS)NrrOFrPr\p) r)r* apply_dropoutrr- project_convr/r0r1 project_bnDropoutdropout)r6rrNr+r%rprqr7rrr*s  z$EfficientNetFinalBlockLayer.__init__) embeddingsrVr:cCs0||}||}|jr,||}||}|Sr;)rurvrtrx)r6ryrVrrrr=s    z#EfficientNetFinalBlockLayer.forwardr>r?r@rArrfloatrbr*rBrXrCr=rDrrr7rrns rnc sJeZdZdZeeeeeeeeed fdd Ze j e j dddZ Z S)EfficientNetBlocka 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. ) rrNr+r% expand_ratiorrprqr[c st||_|jdk|_||} |jr.round_repeatsc3s|]}|VqdSr;r).0n)rrr jz/EfficientNetEncoder.__init__..rr) rrNr+r%rr}rprqr[irOFrPr\)r)r*rrlenrGsumZnum_block_repeatsrangerrHstridesZ kernel_sizesZ expand_ratiosZdepthwise_paddingZdrop_connect_rater|appendrZ ModuleListblocksr-top_convr/ hidden_dimr0r1top_bnr r3top_activation)r6rZnum_base_blocksZ num_blocksZcurr_block_numrirNr+r%rr}jrqr[rpblockr7)rr6rr*`sZ            zEfficientNetEncoder.__init__FT)rVoutput_hidden_states return_dictr:cCst|r |fnd}|jD]}||}|r||f7}q||}||}||}|shtdd||fDSt||dS)Ncss|]}|dur|VqdSr;r)rvrrrrrz.EfficientNetEncoder.forward..)last_hidden_staterV)rrrrtupler )r6rVrrZall_hidden_statesrrrrr=s     zEfficientNetEncoder.forward)FT) r>r?r@rArr*rBrXrrbr r=rDrrr7rrWs:rc@s2eZdZUeed<dZdZgZej dddZ dS)EfficientNetPreTrainedModelr efficientnetr9)modulecCsFt|tjtjtjfrB|jjjd|jj d|j durB|j j dS)zInitialize the weightsg)meanZstdN) rrLinearr-r/weightdataZnormal_rZinitializer_ranger&Zzero_)r6rrrr _init_weightss z)EfficientNetPreTrainedModel._init_weightsN) r>r?r@r__annotations__Zbase_model_prefixZmain_input_nameZ_no_split_modulesrModulerrrrrrs rcsReZdZedfdd Zedeejee ee e e e fdddZ ZS) EfficientNetModelrcs~t|||_t||_t||_|jdkrDtj |j dd|_ n.|jdkrbtj |j dd|_ nt d|j|dS)NrT)Z ceil_moderz2config.pooling must be one of ['mean', 'max'] got )r)r*rrryrencoderZ pooling_typerZ AvgPool2drpoolerZ MaxPool2d ValueErrorZpooling post_initr5r7rrr*s     zEfficientNetModel.__init__N)r9rrr:cCs|dur |n|jj}|dur |n|jj}|dur8td||}|j|||d}|d}||}||jdd}|s||f|ddSt |||j dS)Nz You have to specify pixel_valuesrrrrr)r pooler_outputrV) rruse_return_dictrryrrZreshapeshaper rV)r6r9rrZembedding_outputZencoder_outputsr pooled_outputrrrr=s*  zEfficientNetModel.forward)NNN)r>r?r@rr*rrrBrXrbrrr r=rDrrr7rrs rz EfficientNet Model with an image classification head on top (a linear layer on top of the pooled features), e.g. for ImageNet. )Z custom_introc sTeZdZfddZedeejeejee ee e e e fdddZ ZS)"EfficientNetForImageClassificationcsdt||j|_||_t||_tj|jd|_ |jdkrNt |j |jnt |_ |dS)Nrrr)r)r* num_labelsrrrrrwZ dropout_raterxrrZIdentity classifierrr5r7rrr*s  $z+EfficientNetForImageClassification.__init__N)r9labelsrrr:c Csv|dur |n|jj}|j|||d}|r.|jn|d}||}||}d}|dur6|jjdur|jdkrxd|j_n4|jdkr|jt j ks|jt j krd|j_nd|j_|jjdkrt } |jdkr| | | }n | ||}nN|jjdkrt} | |d|j|d}n|jjdkr6t} | ||}|sf|f|dd} |durb|f| S| St|||jd S) a 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). NrrZ regressionZsingle_label_classificationZmulti_label_classificationr)losslogitsrV)rrrrrxrZ problem_typerZdtyperBlongrrrhrviewrr rV) r6r9rrroutputsrrrZloss_fctoutputrrrr=s@      "     z*EfficientNetForImageClassification.forward)NNNN)r>r?r@r*rrrBrXZ LongTensorrbrrr r=rDrrr7rrs  r)rrr)T)/rArtypingrrrBZtorch.utils.checkpointrZtorch.nnrrrZ activationsr Zmodeling_outputsr r r Zmodeling_utilsr utilsrrZconfiguration_efficientnetrZ get_loggerr>loggerrrrrbrrrr-rErLrYrcrnr|rrrr__all__rrrrs>     ''!QZ8E