T^(hBJJdZddlmZmZmZmZmZddlZddlm Z ddlm Z ddlm Z m Z m Z ddlmZmZddlmZddlmZmZmZmZdd lmZdd lmZmZmZmZdd lm Z ejBe"Z#d Z$d Z%gdZ&Gdde jNZ(Gdde jNZ)Gdde jNZ*Gdde jNZ+dZ,dZ-GddeZ.ede,Gdde.Z/ede,Gdd e.Z0ed!e,Gd"d#e.e Z1gd$Z2y)%zPyTorch TextNet model.)AnyListOptionalTupleUnionN)Tensor)BCEWithLogitsLossCrossEntropyLossMSELoss)PreTrainedModeladd_start_docstrings)ACT2CLS)BackboneOutputBaseModelOutputWithNoAttention(BaseModelOutputWithPoolingAndNoAttention$ImageClassifierOutputWithNoAttention) TextNetConfig)add_code_sample_docstrings%add_start_docstrings_to_model_forwardloggingreplace_return_docstrings) BackboneMixinrzczczup/textnet-base)ic\eZdZdeffd Zdej dej fdZxZS)TextNetConvLayerconfigct||j|_|j|_|j |_t|jtr$|jddz|jddzfn|jdz}tj|j|j|j|j|d|_tj|j|j |_tj$|_|jt)|j|_yy)NrrF) kernel_sizestridepaddingbias)super__init__stem_kernel_sizer! stem_strider" stem_act_funcactivation_function isinstancetuplennConv2dstem_num_channelsstem_out_channelsconv BatchNorm2dbatch_norm_eps batch_normIdentity activationr)selfrr# __class__s j/var/www/skyplay_api_hub/venv/lib/python3.12/site-packages/transformers/models/textnet/modeling_textnet.pyr&zTextNetConvLayer.__init__3s !22(( #)#7#7 &1159   "a '););A)>!)C D((A-  II  $ $  $ $//%%   ..)A)A6CXCXY++-  # # /%d&>&>?ADO 0 hidden_statesreturnch|j|}|j|}|j|SN)r1r4r6)r7r;s r9forwardzTextNetConvLayer.forwardNs- -0  6 }--r:) __name__ __module__ __qualname__rr&torchrr? __classcell__r8s@r9rr2s,B}B6.U\\.ell.r:rc peZdZdZdededededef fd Zdejd ejfd Z xZ S) TextNetRepConvLayera This layer supports re-parameterization by combining multiple convolutional branches (e.g., main convolution, vertical, horizontal, and identity branches) during training. At inference time, these branches can be collapsed into a single convolution for efficiency, as per the re-parameterization paradigm. The "Rep" in the name stands for "re-parameterization" (introduced by RepVGG). r in_channels out_channelsr!r"ctt |||_||_||_||_|ddz dz|ddz dzf}t j|_t j|||||d|_ t j||j|_ |ddz dzdf}d|ddz dzf}|ddk7rLt j|||ddf||d|_t j||j|_nd\|_|_|ddk7rLt j||d|df||d|_t j||j|_nd\|_|_||k(r,|dk(r't j||j|_yd|_y)Nrrr F)rHrIr!r"r#r$) num_featuresepsNN)r%r& num_channelsrIr!r"r-ReLUr*r. main_convr2r3main_batch_norm vertical_convvertical_batch_normhorizontal_convhorizontal_batch_norm rbr_identity) r7rrHrIr!r"r#vertical_paddinghorizontal_paddingr8s r9r&zTextNetRepConvLayer.__init__^s '(& NQ&1,{1~/Aa.GH#%779 #%#   "~~Q !#')(^Q/( "D (*~~Q #%99') A/* $D *,\W]WlWl)mD &?I )rPrQrRrSrTrUrVr*)r7r; main_outputsvertical_outputshorizontal_outputsid_outs r9r?zTextNetRepConvLayer.forwards~~m4 ++L9     )#11-@ #778HI '*::L    +!%!5!5m!D !%!;!;)rk)r7 hidden_stateblocks r9r?zTextNetStage.forwards%ZZ /E .L /r:)r@rArBrr_r&r?rDrEs@r9raras*}*S*"r:rac beZdZdeffd Z ddej deedeede fdZ xZ S) TextNetEncoderrct|g}t|j}t |D]}|j t ||tj||_ yr>) r%r&rfrdrangerirar-rjstages)r7rrw num_stagesstage_ixr8s r9r&zTextNetEncoder.__init__s\ 778 j) :H MM,vx8 9 :mmF+ r:rqoutput_hidden_states return_dictr<c|g}|jD]}||}|j||s |f}|r||fzS|St||S)N)last_hidden_stater;)rwrir)r7rqrzr{r;rkoutputs r9r?zTextNetEncoder.forwardse & [[ /E .L   . /"_F0D6],, P& P- \ijjr:rM) r@rArBrr&rCrrboolrr?rDrEs@r9rtrtsS,},04&* kllk'tnkd^ k ( kr:rtaI 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 ([`TextNetConfig`]): 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. aE Args: pixel_values (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)`): Pixel values. Pixel values can be obtained using [`AutoImageProcessor`]. See [`TextNetImageProcessor.__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. c"eZdZdZeZdZdZdZy)TextNetPreTrainedModelz An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained models. textnet pixel_valuesct|tjtjfrm|jj j d|jj|j%|jj jyyt|tjrW|jj jd|j%|jj jyyy)Ng)meanstdg?) r+r-Linearr.weightdatanormal_rinitializer_ranger$zero_r2fill_)r7modules r9 _init_weightsz$TextNetPreTrainedModel._init_weightss fryy"))4 5 MM   & &CT[[5R5R & S{{&   &&('  / MM   $ $S ){{&   &&('0r:N) r@rArBr^r config_classbase_model_prefixmain_input_namerr:r9rrs !L!$O)r:rzPThe bare Textnet model outputting raw features without any specific head on top.ceZdZfdZeeeeee de  d de de e de e deeeeefeeeffdZxZS) TextNetModelct||t||_t ||_t jd|_|jy)N)r r ) r%r&rstemrtencoderr-AdaptiveAvgPool2dpooler post_initr7rr8s r9r&zTextNetModel.__init__sD  $V, %f- **62  r:vision) checkpoint output_typermodalityexpected_outputrrzr{r<c:||n|jj}||n|jj}|j|}|j |||}|d}|j |}|s||f}|r ||dfzS|St |||r |dSdS)Nrzr{rr)r} pooler_outputr;)ruse_return_dictrzrrrr) r7rrzr{rqencoder_outputsr} pooled_outputr~s r9r?zTextNetModel.forwards&1% 1` a classification loss is computed (Cross-Entropy). Returns: Examples: ```python >>> import torch >>> import requests >>> from transformers import TextNetForImageClassification, TextNetImageProcessor >>> from PIL import Image >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg" >>> image = Image.open(requests.get(url, stream=True).raw) >>> processor = TextNetImageProcessor.from_pretrained("czczup/textnet-base") >>> model = TextNetForImageClassification.from_pretrained("czczup/textnet-base") >>> inputs = processor(images=image, return_tensors="pt") >>> with torch.no_grad(): ... outputs = model(**inputs) >>> outputs.logits.shape torch.Size([1, 2]) ```Nrrr regressionsingle_label_classificationmulti_label_classificationrr )losslogitsr;)rrrrr problem_typerdtyperClongr_r squeezer viewr rr;) r7rrrzr{outputsr}layerrrloss_fctr~s r9r?z%TextNetForImageClassification.forwardWsH&1%)r%r&_init_backbonerrrgrKrrs r9r&zTextNetBackbone.__init__sC   v&#F+ "// r:rrrzr{r<c||n|jj}||n|jj}|j|d|}|r |jn|d}d}t |j D]\}}||jvs|||fz }|s |f} |r|r |jn|d}| |fz } | St||r|jdSddS)a Returns: Examples: ```python >>> import torch >>> import requests >>> from PIL import Image >>> from transformers import AutoImageProcessor, AutoBackbone >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg" >>> image = Image.open(requests.get(url, stream=True).raw) >>> processor = AutoImageProcessor.from_pretrained("czczup/textnet-base") >>> model = AutoBackbone.from_pretrained("czczup/textnet-base") >>> inputs = processor(image, return_tensors="pt") >>> with torch.no_grad(): >>> outputs = model(**inputs) ```NTrr r) feature_mapsr; attentions) rrrzrr; enumerate stage_names out_featuresr) r7rrzr{rr;ridxrkr~s r9r?zTextNetBackbone.forwards4&1%Xos1 "1 :B4.1 ^fgk^l1 uU|^+ ,1 YE1 r:r)rrrr)3r^typingrrrrrrCtorch.nnr-rr r r transformersr r transformers.activationsrtransformers.modeling_outputsrrrr1transformers.models.textnet.configuration_textnetrtransformers.utilsrrrr!transformers.utils.backbone_utilsr get_loggerr@loggerrrrModulerrGrartTEXTNET_START_DOCSTRINGrrrrr__all__rr:r9rs]44 AA>, L <   H %"+).ryy.DW6"))W6t2990kRYYk:  )_)*V) )) ) X Us$:UsUsp  > ,m>  > B ir: