U 9%e]@sdZddlmZmZmZddlZddlZddlmZddlm Z m Z m Z ddl m Z ddlmZmZmZmZdd lmZdd lmZmZmZmZmZdd lmZd d lmZee Z!dZ"dZ#d dddgZ$dZ%dZ&dgZ'd3ej(e)e*ej(dddZ+Gdddej,Z-Gdddej,Z.Gdddej,Z/Gdddej,Z0Gd d!d!ej,Z1Gd"d#d#ej,Z2Gd$d%d%ej,Z3Gd&d'd'eZ4d(Z5d)Z6ed*e5Gd+d,d,e4Z7ed-e5Gd.d/d/e4Z8ed0e5Gd1d2d2e4eZ9dS)4z PyTorch ConvNextV2 model.)OptionalTupleUnionN)nn)BCEWithLogitsLossCrossEntropyLossMSELoss)ACT2FN)BackboneOutputBaseModelOutputWithNoAttention(BaseModelOutputWithPoolingAndNoAttention$ImageClassifierOutputWithNoAttention)PreTrainedModel)add_code_sample_docstringsadd_start_docstrings%add_start_docstrings_to_model_forwardloggingreplace_return_docstrings) BackboneMixin)ConvNextV2Configrzfacebook/convnextv2-tiny-1k-224iztabby, tabby catF)input drop_probtrainingreturncCsd|dks |s|Sd|}|jdfd|jd}|tj||j|jd}||||}|S)aF Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks). Comment by Ross Wightman: This is the same as the DropConnect impl I created for EfficientNet, etc networks, however, the original name is misleading as 'Drop Connect' is a different form of dropout in a separate paper... See discussion: https://github.com/tensorflow/tpu/issues/494#issuecomment-532968956 ... I've opted for changing the layer and argument names to 'drop path' rather than mix DropConnect as a layer name and use 'survival rate' as the argument. rrr)r)dtypedevice)shapendimtorchZrandrrZfloor_div)rrrZ keep_probr Z random_tensoroutputr%q/var/www/html/Darija-Ai-API/env/lib/python3.8/site-packages/transformers/models/convnextv2/modeling_convnextv2.py drop_path@s r'csPeZdZdZd eeddfdd ZejejdddZ e d d d Z Z S) ConvNextV2DropPathzXDrop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).N)rrcst||_dSN)super__init__r)selfr __class__r%r&r+Xs zConvNextV2DropPath.__init__ hidden_statesrcCst||j|jSr))r'rrr,r0r%r%r&forward\szConvNextV2DropPath.forward)rcCs d|jS)Nzp={})formatr)r,r%r%r& extra_repr_szConvNextV2DropPath.extra_repr)N) __name__ __module__ __qualname____doc__rfloatr+r"Tensorr2strr4 __classcell__r%r%r-r&r(Usr(cs:eZdZdZedfdd ZejejdddZZ S) ConvNextV2GRNz)GRN (Global Response Normalization) layer)dimcs>tttddd||_ttddd||_dS)Nr)r*r+r Parameterr"zerosweightbias)r,r>r-r%r&r+fs zConvNextV2GRN.__init__r/cCsDtj|dddd}||jdddd}|j|||j|}|S)N)rrCT)pr>keepdim)r>rEư>)r"ZnormmeanrArB)r,r0Zglobal_featuresZ norm_featuresr%r%r&r2kszConvNextV2GRN.forward) r5r6r7r8intr+r" FloatTensorr2r<r%r%r-r&r=csr=cs6eZdZdZd fdd ZejejdddZZS) ConvNextV2LayerNormaALayerNorm that supports two data formats: channels_last (default) or channels_first. The ordering of the dimensions in the inputs. channels_last corresponds to inputs with shape (batch_size, height, width, channels) while channels_first corresponds to inputs with shape (batch_size, channels, height, width). rG channels_lastcs`ttt||_tt||_||_ ||_ |j dkrTt d|j |f|_ dS)N)rLchannels_firstzUnsupported data format: ) r*r+rr?r"ZonesrAr@rBeps data_formatNotImplementedErrornormalized_shape)r,rQrNrOr-r%r&r+{s  zConvNextV2LayerNorm.__init__)xrcCs|jdkr*tjj||j|j|j|j}n|jdkr|j }| }|j ddd}|| dj ddd}||t ||j}|j|d}|jddddf||jddddf}|S)NrLrMrT)rErC)r)rOr"rZ functionalZ layer_normrQrArBrNrr9rHpowsqrtto)r,rRZ input_dtypeusr%r%r&r2s    ,zConvNextV2LayerNorm.forward)rGrL) r5r6r7r8r+r"r:r2r<r%r%r-r&rKus rKcs4eZdZdZfddZejejdddZZ S)ConvNextV2EmbeddingszThis class is comparable to (and inspired by) the SwinEmbeddings class found in src/transformers/models/swin/modeling_swin.py. csLttj|j|jd|j|jd|_t|jdddd|_ |j|_dS)Nr kernel_sizestriderGrMrNrO) r*r+rConv2d num_channels hidden_sizesZ patch_sizepatch_embeddingsrK layernormr,configr-r%r&r+s zConvNextV2Embeddings.__init__) pixel_valuesrcCs4|jd}||jkrtd||}||}|S)NrzeMake sure that the channel dimension of the pixel values match with the one set in the configuration.)r r^ ValueErrorr`ra)r,rdr^ embeddingsr%r%r&r2s    zConvNextV2Embeddings.forward r5r6r7r8r+r"rJr:r2r<r%r%r-r&rXs rXcs6eZdZdZdfdd ZejejdddZZ S) ConvNextV2Layera5This corresponds to the `Block` class in the original implementation. There are two equivalent implementations: [DwConv, LayerNorm (channels_first), Conv, GELU,1x1 Conv]; all in (N, C, H, W) (2) [DwConv, Permute to (N, H, W, C), LayerNorm (channels_last), Linear, GELU, Linear]; Permute back The authors used (2) as they find it slightly faster in PyTorch. Args: config ([`ConvNextV2Config`]): Model configuration class. dim (`int`): Number of input channels. drop_path (`float`): Stochastic depth rate. Default: 0.0. rcsttj||dd|d|_t|dd|_t|d||_t |j |_ t d||_ td|||_|dkr|t|nt|_dS)Nrr )rZpaddinggroupsrGrNr)r*r+rr]dwconvrKraLinearpwconv1r Z hidden_actactr=grnpwconv2r(Identityr')r,rcr>r'r-r%r&r+s  zConvNextV2Layer.__init__r/cCsr|}||}|dddd}||}||}||}||}||}|dddd}|||}|S)NrrCr r)rmZpermuterarorprqrrr')r,r0rrRr%r%r&r2s      zConvNextV2Layer.forward)rrgr%r%r-r&rhs  rhcs6eZdZdZd fdd ZejejdddZZ S) ConvNextV2StageaConvNeXTV2 stage, consisting of an optional downsampling layer + multiple residual blocks. Args: config ([`ConvNextV2Config`]): Model configuration class. in_channels (`int`): Number of input channels. out_channels (`int`): Number of output channels. depth (`int`): Number of residual blocks. drop_path_rates(`List[float]`): Stochastic depth rates for each layer. rCNc st|ks|dkrBtt|dddtj|||d|_n t|_pXdg|tjfddt|D|_ dS) NrrGrMr\rYrcsg|]}t|dqS))r>r')rh).0jrcdrop_path_rates out_channelsr%r& sz,ConvNextV2Stage.__init__..) r*r+rZ SequentialrKr]downsampling_layerrsrangelayers)r,rc in_channelsryrZr[depthrxr-rwr&r+s   zConvNextV2Stage.__init__r/cCs||}||}|Sr))r{r}r1r%r%r&r2s  zConvNextV2Stage.forward)rCrCrCNrgr%r%r-r&rts rtcsDeZdZfddZdejeeeeee e fdddZ Z S) ConvNextV2Encoderc stt|_ddtd|jt|j  |j D}|j d}t |j D]H}|j |}t||||dkrrdnd|j |||d}|j||}qPdS)NcSsg|] }|qSr%)tolist)rurRr%r%r&rzsz.ConvNextV2Encoder.__init__..rrCr)r~ryr[rrx)r*r+rZ ModuleListstagesr"ZlinspaceZdrop_path_ratesumZdepthssplitr_r|Z num_stagesrtappend)r,rcrxZprev_chsiZout_chsstager-r%r&r+s$     zConvNextV2Encoder.__init__FT)r0output_hidden_states return_dictrcCsj|rdnd}t|jD]\}}|r,||f}||}q|rD||f}|s^tdd||fDSt||dS)Nr%css|]}|dk r|VqdSr)r%)ruvr%r%r& sz,ConvNextV2Encoder.forward..)last_hidden_stater0) enumeratertupler )r,r0rrZall_hidden_statesrZ layer_moduler%r%r&r2s    zConvNextV2Encoder.forward)FT) r5r6r7r+r"rJrboolrrr r2r<r%r%r-r&rs  rc@s2eZdZdZeZdZdZdZddZ d dd Z d S) ConvNextV2PreTrainedModelz An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained models. convnextv2rdTcCsjt|tjtjfr@|jjjd|jjd|j dk rf|j j n&t|tj rf|j j |jj ddS)zInitialize the weightsr)rHZstdNg?) isinstancerrnr]rAdataZnormal_rcZinitializer_rangerBZzero_ LayerNormZfill_)r,moduler%r%r& _init_weights4s   z'ConvNextV2PreTrainedModel._init_weightsFcCst|tr||_dSr))rrZgradient_checkpointing)r,rvaluer%r%r&_set_gradient_checkpointing@s z5ConvNextV2PreTrainedModel._set_gradient_checkpointingN)F) r5r6r7r8r config_classZbase_model_prefixZmain_input_nameZsupports_gradient_checkpointingrrr%r%r%r&r)s raL This model is a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) subclass. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior. Parameters: config ([`ConvNextV2Config`]): Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the [`~PreTrainedModel.from_pretrained`] method to load the model weights. aI Args: pixel_values (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)`): Pixel values. Pixel values can be obtained using [`ConvNextImageProcessor`]. See [`ConvNextImageProcessor.__call__`] for details. output_hidden_states (`bool`, *optional*): Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for more detail. return_dict (`bool`, *optional*): Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple. zSThe bare ConvNextV2 model outputting raw features without any specific head on top.c s^eZdZfddZeeeeee de dd e j e ee eeeefdddZZS) ConvNextV2ModelcsJt|||_t||_t||_tj|j d|j d|_ | dS)NrFrk) r*r+rcrXrfrencoderrrr_Zlayer_norm_epsra post_initrbr-r%r&r+cs    zConvNextV2Model.__init__Zvision) checkpoint output_typerZmodalityexpected_outputNrdrrrcCs|dk r |n|jj}|dk r |n|jj}|dkr8td||}|j|||d}|d}||ddg}|s||f|ddSt|||j dS)Nz You have to specify pixel_valuesrrrrFr)r pooler_outputr0) rcruse_return_dictrerfrrarHr r0)r,rdrrembedding_outputZencoder_outputsr pooled_outputr%r%r&r2ps( zConvNextV2Model.forward)NNN)r5r6r7r+rCONVNEXTV2_INPUTS_DOCSTRINGr_CHECKPOINT_FOR_DOCr _CONFIG_FOR_DOC_EXPECTED_OUTPUT_SHAPEr"rJrrrrr2r<r%r%r-r&r]s$   rz ConvNextV2 Model with an image classification head on top (a linear layer on top of the pooled features), e.g. for ImageNet. c sdeZdZfddZeeeeee e dde j e e je ee eeeefdddZZS) ConvNextV2ForImageClassificationcsRt||j|_t||_|jdkr 1` a classification loss is computed (Cross-Entropy). NrrZ regressionZsingle_label_classificationZmulti_label_classificationrFrC)losslogitsr0)rcrrrrZ problem_typerrr"longrIrZsqueezerviewrrr0) r,rdrrroutputsrrrZloss_fctr$r%r%r&r2s>     "     z(ConvNextV2ForImageClassification.forward)NNNN)r5r6r7r+rrr_IMAGE_CLASS_CHECKPOINTrr_IMAGE_CLASS_EXPECTED_OUTPUTr"rJrZ LongTensorrrrr2r<r%r%r-r&rs&  rzT ConvNeXT V2 backbone, to be used with frameworks like DETR and MaskFormer. csPeZdZfddZeeeeedde j e e e e edddZ ZS) ConvNextV2Backbonecst|t|t||_t||_|jdg|j|_i}t |j |j D]\}}t |dd||<qRt ||_|dS)NrrM)rO)r*r+Z_init_backbonerXrfrrr_Z num_featureszipZ _out_featuresZchannelsrKrZ ModuleDicthidden_states_normsr)r,rcrrr^r-r%r&r+s     zConvNextV2Backbone.__init__)rrNrc Cs|dk r |n|jj}|dk r |n|jj}||}|j|ddd}|j}d}tt|jdd|ddD].\}\} } | |j krl|j | | } || f7}ql|s|f} |r| |jf7} | St ||r|jndddS)a Returns: Examples: ```python >>> from transformers import AutoImageProcessor, AutoBackbone >>> import torch >>> from PIL import Image >>> import requests >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg" >>> image = Image.open(requests.get(url, stream=True).raw) >>> processor = AutoImageProcessor.from_pretrained("facebook/convnextv2-tiny-1k-224") >>> model = AutoBackbone.from_pretrained("facebook/convnextv2-tiny-1k-224") >>> inputs = processor(image, return_tensors="pt") >>> outputs = model(**inputs) ```NTrr%r) feature_mapsr0Z attentions) rcrrrfrr0rrZ stage_namesZ out_featuresrr ) r,rdrrrrr0ridxrZ hidden_stater$r%r%r&r2 s2 ,    zConvNextV2Backbone.forward)NN)r5r6r7r+rrrr rr"r:rrr2r<r%r%r-r&rs  r)rF):r8typingrrrr"Ztorch.utils.checkpointrZtorch.nnrrrZ activationsr Zmodeling_outputsr r r rZmodeling_utilsrutilsrrrrrZutils.backbone_utilsrZconfiguration_convnextv2rZ get_loggerr5loggerrrrrrZ(CONVNEXTV2_PRETRAINED_MODEL_ARCHIVE_LISTr:r9rr'Moduler(r=rKrXrhrtrrZCONVNEXTV2_START_DOCSTRINGrrrrr%r%r%r&sZ       ,!0  ;L