iz ddZddlZddlmZddlZddlmZddlmZddlm Z ddl m Z dd l m Z dd lmZmZmZmZmZmZd d lmZer dd lmZmZndZdZej4eZeedGddeZeedGddeZeedGddeZGddej@Z!Gddej@Z"Gddej@Z#dBd ejHd!e%d"e&d#ejHfd$Z'Gd%d&ej@Z(Gd'd(ej@Z)Gd)d*ej@Z*Gd+d,ej@Z+Gd-d.ej@Z,Gd/d0ej@Z-Gd1d2ej@Z.Gd3d4ej@Z/Gd5d6ej@Z0eGd7d8e Z1eGd9d:e1Z2ed;Gd<d=e1Z3ed>Gd?d@e e1Z4gdAZ5y)Cz9PyTorch Dilated Neighborhood Attention Transformer model.N) dataclass)nn)ACT2FN) BackboneMixin)BackboneOutput)PreTrainedModel) ModelOutputOptionalDependencyNotAvailableauto_docstringis_natten_availableloggingrequires_backends) DinatConfig) natten2dav natten2dqkrpbctNr argskwargss d/mnt/e/genesis-system/.venv/lib/python3.12/site-packages/transformers/models/dinat/modeling_dinat.pyrr) ,..ctrrrs rrr,rrzO Dinat encoder's outputs, with potential hidden states and attentions. ) custom_introceZdZUdZdZej dzed<dZe ej dfdzed<dZ e ej dfdzed<dZ e ej dfdzed<y)DinatEncoderOutputa reshaped_hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`): Tuple of `torch.FloatTensor` (one for the output of the embeddings + one for the output of each stage) of shape `(batch_size, hidden_size, height, width)`. Hidden-states of the model at the output of each layer plus the initial embedding outputs reshaped to include the spatial dimensions. Nlast_hidden_state. hidden_states attentionsreshaped_hidden_states) __name__ __module__ __qualname____doc__r!torch FloatTensor__annotations__r"tupler#r$rrr r 6s}37u((4/6:>M5**C/047>7;Je'',-4;CGE%"3"3S"89D@Grr zW Dinat model's outputs that also contains a pooling of the last hidden states. ceZdZUdZdZej dzed<dZej dzed<dZ e ej dfdzed<dZ e ej dfdzed<dZ e ej dfdzed<y) DinatModelOutputa pooler_output (`torch.FloatTensor` of shape `(batch_size, hidden_size)`, *optional*, returned when `add_pooling_layer=True` is passed): Average pooling of the last layer hidden-state. reshaped_hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`): Tuple of `torch.FloatTensor` (one for the output of the embeddings + one for the output of each stage) of shape `(batch_size, hidden_size, height, width)`. Hidden-states of the model at the output of each layer plus the initial embedding outputs reshaped to include the spatial dimensions. Nr! pooler_output.r"r#r$) r%r&r'r(r!r)r*r+r0r"r,r#r$r-rrr/r/Ls 37u((4/6.2M5$$t+2:>M5**C/047>7;Je'',-4;CGE%"3"3S"89D@Grr/z1 Dinat outputs for image classification. ceZdZUdZdZej dzed<dZej dzed<dZ e ej dfdzed<dZ e ej dfdzed<dZ e ej dfdzed<y) DinatImageClassifierOutputa7 loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided): Classification (or regression if config.num_labels==1) loss. logits (`torch.FloatTensor` of shape `(batch_size, config.num_labels)`): Classification (or regression if config.num_labels==1) scores (before SoftMax). reshaped_hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`): Tuple of `torch.FloatTensor` (one for the output of the embeddings + one for the output of each stage) of shape `(batch_size, hidden_size, height, width)`. Hidden-states of the model at the output of each layer plus the initial embedding outputs reshaped to include the spatial dimensions. Nlosslogits.r"r#r$) r%r&r'r(r3r)r*r+r4r"r,r#r$r-rrr2r2es &*D%  d ")'+FE   $+:>M5**C/047>7;Je'',-4;CGE%"3"3S"89D@Grr2cfeZdZdZfdZdej dzdeejfdZ xZ S)DinatEmbeddingsz6 Construct the patch and position embeddings. ct|t||_t j |j |_t j|j|_ yr) super__init__DinatPatchEmbeddingspatch_embeddingsr LayerNorm embed_dimnormDropouthidden_dropout_probdropoutselfconfig __class__s rr9zDinatEmbeddings.__init__sG  4V <LL!1!12 zz&"<"<= r pixel_valuesNreturncl|j|}|j|}|j|}|Sr)r;r>rA)rCrF embeddingss rforwardzDinatEmbeddings.forwards4**<8 YYz* \\*- r) r%r&r'r(r9r)r*r,TensorrJ __classcell__rEs@rr6r6s4>E$5$5$<u||ATrr6c`eZdZdZfdZdej dzdejfdZxZ S)r:z This class turns `pixel_values` of shape `(batch_size, num_channels, height, width)` into the initial `hidden_states` (patch embeddings) of shape `(batch_size, height, width, hidden_size)` to be consumed by a Transformer. c Pt||j}|j|j}}||_|dk(rn t dt jt j|j|dzdddt j|dz|ddd|_ y)Nz2Dinat only supports patch size of 4 at the moment.rrrQrQrr) kernel_sizestridepadding) r8r9 patch_size num_channelsr= ValueErrorr SequentialConv2d projection)rCrDrXrY hidden_sizerEs rr9zDinatPatchEmbeddings.__init__s && $*$7$79I9Ik ( ? QR R-- IId'')9vV\fl m IIkQ& PV`f g rrFNrGc|j\}}}}||jk7r td|j|}|j dddd}|S)NzeMake sure that the channel dimension of the pixel values match with the one set in the configuration.rrQrr)shaperYrZr]permute)rCrF_rYheightwidthrIs rrJzDinatPatchEmbeddings.forwards`)5););&< 4,, ,w __\2 ''1a3 r) r%r&r'r(r9r)r*rKrJrLrMs@rr:r:s/  " E$5$5$<  rr:ceZdZdZej fdedejddffd Zde jde jfdZ xZ S) DinatDownsamplerz Convolutional Downsampling Layer. Args: dim (`int`): Number of input channels. norm_layer (`nn.Module`, *optional*, defaults to `nn.LayerNorm`): Normalization layer class. dim norm_layerrGNct|||_tj|d|zdddd|_|d|z|_y)NrQrRrSrTF)rUrVrWbias)r8r9rgrr\ reductionr>)rCrgrhrEs rr9zDinatDownsampler.__init__sE 3CVF\binoq3w' r input_featurec|j|jddddjdddd}|j|}|S)NrrrrQ)rkrar>)rCrls rrJzDinatDownsampler.forwardsJ}'<'