sg+dZddlmZddlZddlZddlmZmZmZm Z ddl Z ddl m Z ddlmZmZmZmZddlmZmZmZmZmZmZmZmZdd lmZdd lmZm Z m!Z!d d l"m#Z#d Z$gdgdgZ%gdgdgdggdgdgdggZ&dZ'GddejPjRZ*GddejPjRZ+GddejPjRZ,GddejPjRZ-Gdd ejPjRZ.Gd!d"ejPjRZ/Gd#d$ejPjRZ0Gd%d&ejPjRZ1Gd'd(ejPjRZ2eGd)d*ejPjRZ3Gd+d,eZ4d-Z5d.Z6ed/e5Gd0d1e4Z7Gd2d3ejPjRZ8ed4e5Gd5d6e4eZ9ed7e5Gd8d9e4eZ:ed:e5Gd;dB xx YbhhG  ^ zz,O0DE ww|j!_= rTctjtj||j|j}tj |d|z}||jz}|S)z} Replace non-padding symbols with their position numbers. Position numbers begin at padding_token_index + 1. rr)rYcast not_equalrrcumsum)rO input_idsmaskrs rS"create_position_ids_from_input_idsz=TFLayoutLMv3TextEmbeddings.create_position_ids_from_input_idssTwwr||It/G/GH)//ZyyA.5 #d&>&>> rTcJ||j|S|j|Srk)rr)rOr inputs_embedss rScreate_position_idsz.TFLayoutLMv3TextEmbeddings.create_position_idss*  >>}M M::9E ErTc||j||}|tj|}ntj|dd}|!tj||j}|1t ||j j|j |}|j|}||z} |j|} | | z } |j|} | | z } |j| } |j| |} | S)NrXrtraining) rrYrzerosrrrv input_dimrxr|rryr) rOrrtoken_type_idsrrrrfrxr]r|rs rSr^zTFLayoutLMv3TextEmbeddings.calls  33I}ML  ((9-K((=1#26K  !XXk9K9KLN  *9d6J6J6T6T U 00;M $ : :> J"%:: "66|D)) &*&P&PQU&V#11 ^^J/ \\*x\@ rTc|jryd|_t|ddMtj|jj 5|jj ddddt|ddMtj|jj 5|jj ddddt|dddtj|jj 5|jj dd|jjgdddt|ddMtj|jj 5|jj ddddt|ddMtj|jj 5|jj ddddt|ddMtj|jj 5|jj ddddt|ddMtj|jj 5|jj ddddt|d dNtj|jj 5|jj ddddyy#1swYxYw#1swYLxYw#1swYxYw#1swYxYw#1swYNxYw#1swYxYw#1swYxYw#1swYyxYw) NTrvrxryr|r}r~rr)r`rarYrbrvr?rcrxryrNrJr|r}r~rrres rSrcz TFLayoutLMv3TextEmbeddings.build s ::  4*D 1 =t33889 1$$**40 1 40$ 7 Ct99>>? 7**006 7 4d + 7t~~223 L$$dD$++2I2I%JK L 4. 5 At77<<= 5((..t4 5 40$ 7 Ct99>>? 7**006 7 40$ 7 Ct99>>? 7**006 7 40$ 7 Ct99>>? 7**006 7 40$ 7 Ct99>>? 7**006 7 7 D) 1 1 7 7 L L 5 5 7 7 7 7 7 7 7 7s` K>%L ?3L0L% L2$L?>M M>L LL"%L/2L<?M  MM!rg)rrirjri)rrirjri)rrirjri)rrirrirjri)NNNNNF)rtf.Tensor | Nonerrirrrrrrrboolrjrirk) rmrnrorprBrrrrr^rcrqrrs@rSrtrtis0d'+R F'++/)-*.####) # ' # ( ## #J7rTrtcheZdZdfd ZddZdd dZ d d dZd dZxZS) TFLayoutLMv3SelfAttentionc t|d i||j|jzdk7r&t d|jd|jd|j|_t |j|jz |_|j|j z|_tj|j |_ tjj|jt|jd|_tjj|jt|jd|_tjj|jt|jd|_tjj'|j(|_|j,|_|j.|_||_y) NrzThe hidden size (z6) is not a multiple of the number of attention heads ()queryr>r?keyvaluer@)rArBrJnum_attention_heads ValueErrorintattention_head_size all_head_sizemathsqrtattention_score_normaliserrrHDenserrKrrrrattention_probs_dropout_probrhas_relative_attention_biashas_spatial_attention_biasrNrs rSrBz"TFLayoutLMv3SelfAttention.__init__+s "6"    : : :a ?#F$6$6#78 445Q8  $*#=#= #&v'9'9F&%)'+3 3)3$ 3  3 " 3%33 >3j HrTrc2eZdZdfd ZdddZddZxZS)TFLayoutLMv3SelfOutputc xt|di|tjj |j t |jd|_tjj|jd|_ tjj|j|_||_yNdenseunitsr>r?ryrz)rater@rArBrrHrrJrrKrrrryrrrrNrs rSrBzTFLayoutLMv3SelfOutput.__init__ "6"\\''$$IaIa9bip(  88AVAV]h8i||++1K1K+L  rTcz|j|}|j||}|j||z}|SNinputs)r rrrryrOr input_tensorrs rSr^zTFLayoutLMv3SelfOutput.call? - 8  MH M ml.JK rTc"|jryd|_t|dddtj|jj 5|jj dd|jjgdddt|ddetj|jj 5|jj dd|jjgdddyy#1swY|xYw#1swYyxYwNTrry) r`rarYrbrr?rcrNrJryres rSrczTFLayoutLMv3SelfOutput.builds ::  4$ ' 3tzz/ H   $dkk.E.E!FG H 4d + 7t~~223 L$$dD$++2I2I%JK L L 8 H H L L 3C9<3D9DDrgFrrir rirrrjrirkrmrnrorBr^rcrqrrs@rSrr LrTrcVeZdZdfd Z d ddZddZxZS)TFLayoutLMv3Attentionc lt|di|t|d|_t |d|_y)NrOr?outputr@)rArBrself_attentionr self_outputrs rSrBzTFLayoutLMv3Attention.__init__s1 "6"7VL1&xHrTc v|j|||||||}|j|d||} | f|ddz} | S)Nrrr)rr) rOrrrrrrr self_outputsattention_outputrs rSr^zTFLayoutLMv3Attention.callsf**      +  ++LO]U]+^#% QR(88rTc|jryd|_t|ddMtj|jj 5|jj ddddt|ddNtj|jj 5|jj ddddyy#1swYexYw#1swYyxYw)NTrr)r`rarYrbrr?rcrres rSrczTFLayoutLMv3Attention.builds ::  4)4 0 <t22778 0##))$/ 0 4 - 9t//445 -  &&t, - - : 0 0 - - C %C CC rgrrrkrrrs@rSrrslI%)'+ )$    " % >. -rTrc0eZdZdfd ZddZddZxZS)TFLayoutLMv3Intermediatec Tt|di|tjj |j t |jd|_t|jtr"t|j|_ ||_y|j|_ ||_y)Nrrr@)rArBrrHrintermediate_sizerrKrrC hidden_actstrr intermediate_act_fnrNrs rSrBz!TFLayoutLMv3Intermediate.__init__s "6"\\''**vOgOg?hov(   f'' -'89J9J'KD $ (.'8'8D $ rTcL|j|}|j|}|S)Nr)rr')rOrs rSr^zTFLayoutLMv3Intermediate.calls( - 8 00? rTc(|jryd|_t|ddetj|jj 5|jj dd|jjgdddyy#1swYyxYw)NTr) r`rarYrbrr?rcrNrJres rSrczTFLayoutLMv3Intermediate.builds} ::  4$ ' 3tzz/ H   $dkk.E.E!FG H H 4 H Hs 3BBrg)rrirjrirkrrrs@rSr"r"s  HrTr"c2eZdZdfd ZdddZddZxZS)TFLayoutLMv3Outputc xt|di|tjj |j t |jd|_tjj|jd|_ tjj|j|_||_yrrrs rSrBzTFLayoutLMv3Output.__init__rrTcz|j|}|j||}|j||z}|Srr r s rSr^zTFLayoutLMv3Output.callr rTc"|jryd|_t|dddtj|jj 5|jj dd|jjgdddt|ddetj|jj 5|jj dd|jjgdddyy#1swY|xYw#1swYyxYwr) r`rarYrbrr?rcrNr$ryrJres rSrczTFLayoutLMv3Output.build"s ::  4$ ' 3tzz/ N   $dkk.K.K!LM N 4d + 7t~~223 L$$dD$++2I2I%JK L L 8 N N L Lrrgrrrkrrrs@rSr+r+rrTr+cVeZdZdfd Z d ddZddZxZS)TFLayoutLMv3Layerc t|di|t|d|_t |d|_t |d|_y)N attentionr intermediaterr@)rArBrr2r"r3r+ bert_outputrs rSrBzTFLayoutLMv3Layer.__init__/s? "6".vKH4V.Q-f8DrTc |j|||||||}|d} |dd} |j| } |j| | |} | f| z} | S)N)rrrrrrr)r2r3r4) rOrrrrrrrself_attention_outputsrrintermediate_output layer_outputs rSr^zTFLayoutLMv3Layer.call5s"&   /!"0" 2!4(,"//0@A''(;=MX`'a /G+rTc|jryd|_t|ddMtj|jj 5|jj ddddt|ddMtj|jj 5|jj ddddt|ddNtj|jj 5|jj ddddyy#1swYxYw#1swYqxYw#1swYyxYw)NTr2r3r4) r`rarYrbr2r?rcr3r4res rSrczTFLayoutLMv3Layer.buildOs  ::  4d + 7t~~223 +$$T* + 4 . :t00556 .!!''- . 4 - 9t//445 -  &&t, - - :  + + . . - -s$ D%%D1?D=%D.1D:=Ergrrrkrrrs@rSr0r0.slE%)'+ )$    " % >4 -rTr0ceZdZdfd Zd dZ d dZd dZd dZ d ddZddZ xZ S)TFLayoutLMv3Encoderc Dt|di|||_t|jDcgc]}t |d|c}|_|j|_|j|_|jrg|j|_ |j|_ tjj|jt|j dd|_|jr|j$|_|j&|_tjj|jt|j dd|_tjj|jt|j dd|_yycc}w) Nzlayer.rF rel_pos_bias)rr>r=r?rel_pos_x_biasrel_pos_y_biasr@)rArBrNrnum_hidden_layersr0layerrr rel_pos_bins max_rel_posrrHrrrrKr=max_rel_2d_posrel_2d_pos_binsr>r?)rOrNrPirRs rSrBzTFLayoutLMv3Encoder.__init___sf "6" LQRXRjRjLklq'vaS\Bl +1+M+M(*0*K*K'  + + & 3 3D %11D  % 2 200#263K3K#L# !3!D   * *"("7"7D #)#9#9D "',,"4"400#263K3K#L% #5#D  #(,,"4"400#263K3K#L% #5#D  +msFc|dz}tj|}|dz}||k}tjjtj|tj |z }tj||z }||z ||z z} || z} tj| |j } tj| |dz } tj|dkD|j |ztj||| zS)Nrrr) rYabsrlogrfloat32rminimumwhere) rOrelative_positions num_buckets max_distancebucketsmax_exact_bucketsis_smallbuckets_log_ratiodistance_log_ratiobuckets_big_offset buckets_bigs rSrelative_position_bucketz,TFLayoutLMv3Encoder.relative_position_buckets"Q& &&+,(1,..GGKK(DGX(XY!XXl5F&FG  2 2kDU6U V (*<< ggk7==9 jjkAo> *Q. >LPRPXPX g{Q   rTcPtj|dtj|dz }|j|||}tj|||j}||}tj |gd}tj ||j}|S)NrrX)depthr)rrrrr)rYrrWone_hot compute_dtyperZr) rO dense_layerrrNrOrel_pos_matrixrrel_pos_one_hot embeddings rS _cal_pos_embz TFLayoutLMv3Encoder._cal_pos_embs 2>P\ceAff// \Z**WKtGYGYZ0 LLL9 GGIT-?-?@ rTcf|j|j||j|jSrk)rar=rBrC)rOrs rS_cal_1d_pos_embz#TFLayoutLMv3Encoder._cal_1d_pos_embs,  !2!2L$BSBSUYUeUeffrTc|dddddf}|dddddf}|j|j||j|j}|j|j||j|j}||z}|S)Nrr)rar>rErDr?)rOrposition_coord_xposition_coord_y rel_pos_x rel_pos_yrs rS_cal_2d_pos_embz#TFLayoutLMv3Encoder._cal_2d_pos_embs1a=1a=%%          %%          * rTc |rdnd} |rdnd} |jr|j|nd} |jr|j|nd} t |j D]6\}}|r| |fz} |||nd}|||||| | | }|d}|s.| |dfz} 8|r| |fz} |rt || | Std|| | fDS)Nr@)rrrrrlast_hidden_stater attentionsc3&K|] }|| ywrkr@).0rs rS z+TFLayoutLMv3Encoder.call..s^c^os)rrcrri enumeraterAr tuple)rOrrrrroutput_hidden_states return_dictrrall_hidden_statesall_self_attentionsrrrF layer_modulelayer_head_mask layer_outputss rSr^zTFLayoutLMv3Encoder.calls"#7BD$5b48<8X8X$&&|4^b373R3RT))$/X\ (4 POA|#$58H$H!.7.CilO(!%!M*!,M &9]1=Mr?rA) r`rarYrbr=r?rcrBr>rEr?rA)rOrfrAs rSrczTFLayoutLMv3Encoder.builds ::  4 . :t00556 I!!''tT5F5F(GH I 4)4 0 <t22778 N##))4t7K7K*LM N 4)4 0 <t22778 N##))4t7K7K*LM N 4$ ' 3 &]]5::.&KK%&& & 4 I I N N N N&&s0 )F#2)F0)F<G#F-0F9<GG rg)rMrirNrrOr)r]zkeras.layers.DenserrirNrrOr)rri)rri)NNNFFTNF)rrirrrrrrrrrsrrtrrrrrrjoUnion[TFBaseModelOutput, Tuple[tf.Tensor], Tuple[tf.Tensor, tf.Tensor], Tuple[tf.Tensor, tf.Tensor, tf.Tensor]]rk) rmrnrorBrWrarcrir^rcrqrrs@rSr;r;^s D 2'     g*"&+/&*"'%* )-7 77) 7 $ 7  7#77'77 7r&rTr;ceZdZeZd fd Zd dZddZddZdZ dddZ ddZ ddZ dd Z dd Ze d dd ZxZS)TFLayoutLMv3MainLayerc dt|d i|||_|jrt |d|_|j rt|d|_tjj|jd|_ tjj|jd|_|j"s |j$r-|j&|j(z}|j+||ftjjdd |_t/|d |_y) Nr]r patch_embedryrzr) image_sizegư>normencoderr@)rArBrN text_embedrtr] visual_embedr3rrrHrrryrrrrrrLrDinit_visual_bboxrr;r)rOrNrPrrRs rSrBzTFLayoutLMv3MainLayer.__init__s "6"   8lSDO   :6 VD "\\< ggkJ1+=> ggk7 rTc*|j|}tj|d}tj|j|ddg}tj ||gd}t |dd||jz }|j|}|S)Nrrrr) rrYrrrrrarr)rOr\r]r cls_tokenss rS embed_imagez!TFLayoutLMv3MainLayer.embed_imagews%%l3 XXj)!, WWT^^j!Q-?@ YY J7a@  4d + 7 $.. (JYYz* rTc\t|j}|dk(rtj|d}nM|dk(r/tj|d}tj|d}nt d|jdtj ||j }d|z tz}|S)Nrrrrz&Wrong shape for attention_mask (shape ).g?)lenrrYrrrr\LARGE_NEGATIVE)rOrn_dimsextended_attention_masks rSget_extended_attention_maskz1TFLayoutLMv3MainLayer.get_extended_attention_masks^))* Q;&(nn^!&L # q[')nn^!&L #&(nn5LST&U #EnFZFZE[[]^_ _#%''*A4CUCU"V#&)@#@N"R&&rTc|dg|jjzStj|}|dk(rtj|d}tj|d}tj|d}tj|d}tj ||jjddddg}ni|dk(rFtj|d}tj|d}tj|d}n|dk7rt d|jdtj|dk(sJd tj|d tj||j}|S) NrrrrXrr z!Wrong shape for head_mask (shape rzGot head_mask rank of z, but require 5.) rNr@rYrankrrrrrr\)rOrrs rS get_head_maskz#TFLayoutLMv3MainLayer.get_head_masksD  6DKK999 9# Q;yq9Iyq9Iyr:Iyr:IDKK991aAFIq[yq9Iyr:Iyr:I q[@@QQSTU Uwwy!Q&e*@AS@TTd(ee&GGIt'9'9: rTc | | n|jj} | | n|jj} | | n|jj} | t j |} | d}| d}nH| t j |} | d}| d}n&|t j |d}n t d| |j}n=| |j}n.| |j}n| |j}ntj}||i|t j|f|}|t j|f|}|t j|df|}|j|||||| }d}d}||j|}t j|t j |df|}||}nt j||gd}|jjr0|j||}||}nt j||gd}|jj s|jjrt j"dt j |d|}t j$|d}t j&||dg}||_t j$t j"d|d}t j&||dg}t j||gd}n|}|||}nt j|gd}|j)|}|j+|| }n|jj s|jjru|jj rGt j$t j"d|d}t j&||dg}|}|jjr|}|j-|}|j/|}|j1||||| | | }|d}| s |f|ddzSt3||j4|j6 S) NrrzEYou have to specify either input_ids or inputs_embeds or pixel_valuesrr)rrrrrrrr)rrrrrrsrtrk)rNrrsrtrYrrrronesrr]rrrrrrrrryrrrrr rrm)rOrrrrrrrr\rrsrtrrfr seq_length int_dtypeembedding_output final_bboxfinal_position_idsvisual_embeddingsvisual_attention_maskrvisual_position_idsrencoder_outputssequence_outputs rSr^zTFLayoutLMv3MainLayer.callsW62C1N-TXT_T_TqTq$8$D $++JjJj &1% %'GGZBS9TUV9W,X`i$j !%!6!#NQ!GJ{{66$++:`:`&(hhq"((;L2Ma2PXa&b#&(nn5Hq&Q#&(gg.AJPQ?&S#(M,E#%>>"((1jPY2Zab#cL#%77<*a#IL)+LBU3V]^)_&)<& ]%:#4 #%99.>@Q-RYZ#[ #~~.>? #||,"R&&y1 ,, +2/!5#'  *!,#%(;; ; -)77&11  rTrgrk)rjzkeras.layers.Layer)rz tf.Variable)i)rzTuple[int, int]rr)rrrztf.DTyperh)rrirjri)rrrjz(Union[tf.Tensor, List[tf.Tensor | None]] NNNNNNNNNNNFrrrrrrrrrrrrrrr\rrOptional[bool]rsrrtrrrrjr{)rmrnror config_classrBrcrrrrrrrrrr^rqrrs@rSr}r} s#LC*&GP/7"K. '62'+!%+/+/)-&**.)-,0/3&*` #` ` ) ` ) ` ' ` $` (` '` *` -` $` `  ` ` rTr}c4eZdZdZeZdZefdZxZ S)TFLayoutLMv3PreTrainedModelz An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained models. layoutlmv3cnt|}tjdtjd|d<|S)N)NNrrr)rAinput_signaturerY TensorSpecr)rOsigrRs rSrz+TFLayoutLMv3PreTrainedModel.input_signaturegs,g%mmORXXFKF  rT) rmrnrorprrbase_model_prefixpropertyrrqrrs@rSrr^s' $L$ rTra This model inherits from [`TFPreTrainedModel`]. Check the superclass documentation for the generic methods the library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads etc.) This model is also a [keras.Model](https://www.tensorflow.org/api_docs/python/tf/keras/Model) subclass. Use it as a regular TF 2.0 Keras Model and refer to the TF 2.0 documentation for all matter related to general usage and behavior. TensorFlow models and layers in `transformers` accept two formats as input: - having all inputs as keyword arguments (like PyTorch models), or - having all inputs as a list, tuple or dict in the first positional argument. The reason the second format is supported is that Keras methods prefer this format when passing inputs to models and layers. Because of this support, when using methods like `model.fit()` things should "just work" for you - just pass your inputs and labels in any format that `model.fit()` supports! If, however, you want to use the second format outside of Keras methods like `fit()` and `predict()`, such as when creating your own layers or models with the Keras `Functional` API, there are three possibilities you can use to gather all the input Tensors in the first positional argument: - a single Tensor with `input_ids` only and nothing else: `model(input_ids)` - a list of varying length with one or several input Tensors IN THE ORDER given in the docstring: `model([input_ids, attention_mask])` or `model([input_ids, attention_mask, token_type_ids])` - a dictionary with one or several input Tensors associated to the input names given in the docstring: `model({"input_ids": input_ids, "token_type_ids": token_type_ids})` Note that when creating models and layers with [subclassing](https://keras.io/guides/making_new_layers_and_models_via_subclassing/) then you don't need to worry about any of this, as you can just pass inputs like you would to any other Python function! Parameters: config ([`LayoutLMv3Config`]): 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 [`~TFPreTrainedModel.from_pretrained`] method to load the model weights. a Args: input_ids (`Numpy array` or `tf.Tensor` of shape `(batch_size, sequence_length)`): Indices of input sequence tokens in the vocabulary. Note that `sequence_length = token_sequence_length + patch_sequence_length + 1` where `1` is for [CLS] token. See `pixel_values` for `patch_sequence_length`. Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and [`PreTrainedTokenizer.__call__`] for details. [What are input IDs?](../glossary#input-ids) bbox (`Numpy array` or `tf.Tensor` of shape `(batch_size, sequence_length, 4)`, *optional*): Bounding boxes of each input sequence tokens. Selected in the range `[0, config.max_2d_position_embeddings-1]`. Each bounding box should be a normalized version in (x0, y0, x1, y1) format, where (x0, y0) corresponds to the position of the upper left corner in the bounding box, and (x1, y1) represents the position of the lower right corner. Note that `sequence_length = token_sequence_length + patch_sequence_length + 1` where `1` is for [CLS] token. See `pixel_values` for `patch_sequence_length`. pixel_values (`tf.Tensor` of shape `(batch_size, num_channels, height, width)`): Batch of document images. Each image is divided into patches of shape `(num_channels, config.patch_size, config.patch_size)` and the total number of patches (=`patch_sequence_length`) equals to `((height / config.patch_size) * (width / config.patch_size))`. attention_mask (`tf.Tensor` of shape `(batch_size, sequence_length)`, *optional*): Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`: - 1 for tokens that are **not masked**, - 0 for tokens that are **masked**. Note that `sequence_length = token_sequence_length + patch_sequence_length + 1` where `1` is for [CLS] token. See `pixel_values` for `patch_sequence_length`. [What are attention masks?](../glossary#attention-mask) token_type_ids (`Numpy array` or `tf.Tensor` of shape `(batch_size, sequence_length)`, *optional*): Segment token indices to indicate first and second portions of the inputs. Indices are selected in `[0, 1]`: - 0 corresponds to a *sentence A* token, - 1 corresponds to a *sentence B* token. Note that `sequence_length = token_sequence_length + patch_sequence_length + 1` where `1` is for [CLS] token. See `pixel_values` for `patch_sequence_length`. [What are token type IDs?](../glossary#token-type-ids) position_ids (`Numpy array` or `tf.Tensor` of shape `(batch_size, sequence_length)`, *optional*): Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0, config.max_position_embeddings - 1]`. Note that `sequence_length = token_sequence_length + patch_sequence_length + 1` where `1` is for [CLS] token. See `pixel_values` for `patch_sequence_length`. [What are position IDs?](../glossary#position-ids) head_mask (`tf.Tensor` of shape `(num_heads,)` or `(num_layers, num_heads)`, *optional*): Mask to nullify selected heads of the self-attention modules. Mask values selected in `[0, 1]`: - 1 indicates the head is **not masked**, - 0 indicates the head is **masked**. inputs_embeds (`tf.Tensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*): Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert *input_ids* indices into associated vectors than the model's internal embedding lookup matrix. output_attentions (`bool`, *optional*): Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned tensors for more detail. 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. zdThe bare LayoutLMv3 Model transformer outputting raw hidden-states without any specific head on top.ceZdZdgZfdZeeeee e  d ddZ ddZ xZ S) TFLayoutLMv3ModelrcPt||g|i|t|d|_y)Nrr)rArBr}r)rOrNr rPrRs rSrBzTFLayoutLMv3Model.__init__s( 3&3F3/\JrT output_typerc @|j||||||||| | | |  } | S)a Returns: Examples: ```python >>> from transformers import AutoProcessor, TFAutoModel >>> from datasets import load_dataset >>> processor = AutoProcessor.from_pretrained("microsoft/layoutlmv3-base", apply_ocr=False) >>> model = TFAutoModel.from_pretrained("microsoft/layoutlmv3-base") >>> dataset = load_dataset("nielsr/funsd-layoutlmv3", split="train", trust_remote_code=True) >>> example = dataset[0] >>> image = example["image"] >>> words = example["tokens"] >>> boxes = example["bboxes"] >>> encoding = processor(image, words, boxes=boxes, return_tensors="tf") >>> outputs = model(**encoding) >>> last_hidden_states = outputs.last_hidden_state ```) rrrrrrrr\rrsrtr)r)rOrrrrrrrr\rrsrtrrs rSr^zTFLayoutLMv3Model.callsC^//))%'%/!5#"  rTc|jryd|_t|ddNtj|jj 5|jj ddddyy#1swYyxYw)NTr)r`rarYrbrr?rcres rSrczTFLayoutLMv3Model.build0si ::  4t , 8t334 ,%%d+ , , 9 , ,s A11A:rrrk)rmrnro"_keys_to_ignore_on_load_unexpectedrBrrLAYOUTLMV3_INPUTS_DOCSTRINGrr _CONFIG_FOR_DOCr^rcrqrrs@rSrrs +:):&K*+FG+r?rrout_projrr@)rArBrrHrrJrrKrclassifier_dropoutrrr num_labelsrrN)rOrNrPrrRs rSrBz'TFLayoutLMv3ClassificationHead.__init__>s "6"\\''   .v/G/GH (  *0)B)B)NF % %TZTnTn ||++ ,   **   .v/G/GH+   rTc|j||}|j|}|j||}|j|}|S)Nr)rrr)rOr rrs rSr^z#TFLayoutLMv3ClassificationHead.callTsG,,v,9**W%,,w,:--(rTc|jryd|_t|dddtj|jj 5|jj dd|jjgdddt|ddMtj|jj 5|jj ddddt|ddetj|jj 5|jj dd|jjgdddyy#1swYxYw#1swYxYw#1swYyxYw)NTrrr) r`rarYrbrr?rcrNrJrrres rSrcz$TFLayoutLMv3ClassificationHead.build[s* ::  4$ ' 3tzz/ H   $dkk.E.E!FG H 4D ) 5t||001 ) ""4( ) 4T * 6t}}112 K ##T41H1H$IJ K K 7  H H ) ) K Ks$ 3E<E3E+EE(+E4rgr)r rirrrjrirkrlrrs@rSrr9s, KrTra  LayoutLMv3 Model with a sequence classification head on top (a linear layer on top of the final hidden state of the [CLS] token) e.g. for document image classification tasks such as the [RVL-CDIP](https://www.cs.cmu.edu/~aharley/rvl-cdip/) dataset. ceZdZdgZdfd Zeeeee e  d ddZ d dZ xZ S) %TFLayoutLMv3ForSequenceClassificationrc |t||fi|||_t|d|_t |d|_y)Nrr classifier)rArBrNr}rrrrs rSrBz.TFLayoutLMv3ForSequenceClassification.__init__vs8 *6* /\J8lSrTrcP| | n|jj} |j|||||||| | | | |  }|ddddddf}|j|| }|dn|j ||}| s|f|ddz}||f|zS|St |||j |jS)a Returns: Examples: ```python >>> from transformers import AutoProcessor, TFAutoModelForSequenceClassification >>> from datasets import load_dataset >>> import tensorflow as tf >>> processor = AutoProcessor.from_pretrained("microsoft/layoutlmv3-base", apply_ocr=False) >>> model = TFAutoModelForSequenceClassification.from_pretrained("microsoft/layoutlmv3-base") >>> dataset = load_dataset("nielsr/funsd-layoutlmv3", split="train", trust_remote_code=True) >>> example = dataset[0] >>> image = example["image"] >>> words = example["tokens"] >>> boxes = example["bboxes"] >>> encoding = processor(image, words, boxes=boxes, return_tensors="tf") >>> sequence_label = tf.convert_to_tensor([1]) >>> outputs = model(**encoding, labels=sequence_label) >>> loss = outputs.loss >>> logits = outputs.logits ```N rrrrrrrsrtrr\rrrrlosslogitsrrm)rNuse_return_dictrrhf_compute_lossr rrm)rOrrrrrrlabelsrrsrtrr\rrrrrrs rSr^z*TFLayoutLMv3ForSequenceClassification.call|sh&1%