tj[dZddlZddlZddlmZddlmZddlmZddlZddlm Z ddl m Z dd l mZdd lmZdd lmZdd lmZmZdd lmZddlmZddlmZmZmZmZmZm Z ddl!m"Z"m#Z#m$Z$m%Z%m&Z&ej'e(Z)dZ*dZ+dZ,e&e$ze%zZ-edeGddeZ.edeGddeZ/edeGddeZ0Gdde j1Z2Gd d!e j1Z3Gd"d#e j1Z4Gd$d%e j1Z5Gd&d'e j1Z6Gd(d)e j1Z7Gd*d+e j1Z8Gd,d-e j1Z9Gd.d/eZ:Gd0d1e j1Z;Gd2d3e j1ZeGd8d9e=Z?eGd:d;e=Z@eGd<d=e=ZAGd>d?e j1ZBGd@dAe j1ZCGdBdCe j1ZDedDGdEdFe=ZEGdGdHe j1ZFGdIdJe j1ZGGdKdLe j1ZHGdMdNe j1ZIedOGdPdQe=ZJgdRZKdS)SzPyTorch FLAVA model.N) OrderedDict) dataclass)Any)nn)initialization)ACT2FN)create_bidirectional_mask)GradientCheckpointingLayer)BaseModelOutputBaseModelOutputWithPooling)PreTrainedModel)Unpack) ModelOutputTransformersKwargsauto_docstringcan_return_tuplelogging torch_int) FlavaConfigFlavaImageCodebookConfigFlavaImageConfigFlavaMultimodalConfigFlavaTextConfigzfacebook/flava-image-codebookg$(~k@a Output from FlavaModel containing embeddings and outputs from individual encoders. Note that `image_embeddings` and `text_embeddigns` returned are similar to pooled output returned from a transformer. If you want embeddings for contrastive loss or retrieval use a FLAVA model's `image_projection` and `text_projection` layers on `image_embeddings` and `text_embeddings` respectively. ) custom_introceZdZUdZdZejdzed<dZe dzed<dZ ejdzed<dZ e dzed<dZ ejdzed<dZ e dzed<d eefd ZdS) FlavaModelOutputa image_embeddings (`torch.FloatTensor` of shape `(batch_size, output_dim)`, *optional*, returned when `pixel_values` are present): The image embeddings which are basically the pooled output of [`FlavaImageModel`]. image_output (`BaseModelOutputWithPooling`, *optional*, returned when `pixel_values` are present): The output of the [`FlavaImageModel`]. text_embeddings (`torch.FloatTensor` of shape `(batch_size, output_dim)`, *optional*, returned when `input_ids` are present): The text embeddings which are basically the pooled output of [`FlavaTextModel`]. text_output (`BaseModelOutputWithPooling`, *optional*, returned when `input_ids` are present): The output of the [`FlavaTextModel`]. multimodal_embeddings (`torch.FloatTensor` of shape `(batch_size, output_dim)`, *optional*, returned when `input_ids` and `pixel_values` are present and `skip_multimodal_encoder` is `None` or `False`): The multimodal embeddings which are basically the pooled output of [`FlavaTextModel`]. multimodal_output (`BaseModelOutputWithPooling`, returned when `input_ids` and `pixel_values` are present and `skip_multimodal_encoder` is `None` or `False`): The output of the [`FlavaMultimodalModel`]. Nimage_embeddings image_outputtext_embeddings text_outputmultimodal_embeddingsmultimodal_outputreturnc^tfdDS)Nc3tK|]2}|dvr|n!t|V3dS))r"r r$Ngetattrto_tuple).0kselfs f/var/www/html/CA-Chatbot/venv/lib/python3.11/site-packages/transformers/models/flava/modeling_flava.py z,FlavaModelOutput.to_tuple..Vsc   TTTDGGZabfhiZjZjZsZsZuZu      tuplekeysr-s`r.r*zFlavaModelOutput.to_tupleUsC    YY[[      r0)__name__ __module__ __qualname____doc__rtorch FloatTensor__annotations__r r r!r"r#r$r2rr*r0r.rr4s  26e'$.5556:L,t3:::04OU&-44459K+d29996:5,t3:::;?1D8??? %*      r0rz@ Class representing pretraining losses from FLAVA model ceZdZUdZdZejdzed<dZejdzed<dZ ejdzed<dZ ejdzed<dZ ejdzed<dZ ejdzed<d e fd ZdS) FlavaLossesa mim (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `mim_labels` and `pixel_values` are present, `input_ids_masked` is absent and `mim_weight` > 0.): Masked Image Modeling loss as used in BeIT calculated only for unimodal image data. mlm (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `mlm_labels` and `input_ids_masked` are present, `pixel_values` is absent and `mlm_weight` > 0.): Masked Language Modeling loss as used in BERT calculated only for unimodal text data. itm (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `itm_labels`, `input_ids_masked`, `pixel_values` are present and `itm_weight` > 0.): Image Text Matching (ITM) loss calculated for paired image-text data. Note that ITM loss is calculated on masked pairs in FLAVA. global_contrastive (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `input_ids` and `pixel_values` are present and `global_contrastive_weight` > 0.): Contrastive loss for image-text similarity similar to CLIP but calculated globally for paired image-text data. This is calculated on unmasked images and texts. mmm_image (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `mim_labels`, `pixel_values` and `input_ids_masked` are present and `mmm_image_weight` > 0.): Masked Multimodal Modeling loss's image component calculated on paired image-text data. mmm_text (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `mlm_labels`, `pixel_values` and `input_ids_masked` are present and `mmm_text_weight` > 0.): Masked Multimodal Modeling loss's text component calculated on paired image-text data. Nmimmlmitmglobal_contrastive mmm_imagemmm_textr%cDd}|D]}|d}n |S)NTF)values)r-all_nonevs r.rGzFlavaLosses.all_none{s9  A} r0)r5r6r7r8r?r9r:r;r@rArBrCrDboolrGr<r0r.r>r>\s"%)C T !((($(C T !((($(C T !(((37)D0777*.Iu 4'...)-He$&---$r0r>a Output from FlavaForPreTraining containing embeddings, and outputs from individual encoders. Note that `image_embeddings` and `text_embeddings` returned are similar to pooled output returned from a transformer. If you want embeddings for contrastive loss or retrieval use a FLAVA model's `image_projection` and `text_projection` layers on `image_embeddings` and `text_embeddings` respectively. cVeZdZUdZdZejdzed<dZe ed<dZ ejdzed<dZ e dzed<dZ ejdzed<dZe dzed<dZejdzed <dZe dzed <dZejdzed <dZe dzed <dZejdzed <dZe dzed<dZejdzed<dZe dzed<dZejdzed<dZejdzed<dZejdzed<dZejdzed<dZejdzed<dZejdzed<dZejdzed<deefdZ dS)FlavaForPreTrainingOutputay loss (`torch.FloatTensor`, *optional*, returned when `return_loss` is True): Total loss calculated for this model. loss_info (`FlavaLosses`): Detailed info for FLAVA Pretraining losses. Check `FlavaLosses` class description for the information on the keys. image_embeddings (`torch.FloatTensor` of shape `(batch_size, output_dim)`, *optional*, returned when `pixel_values` are present): The image embeddings which are basically the pooled output of [`FlavaImageModel`]. image_output (`BaseModelOutputWithPooling`, *optional*, returned when `pixel_values` are present): The output of the [`FlavaImageModel`]. text_embeddings (`torch.FloatTensor` of shape `(batch_size, output_dim)`, *optional*, returned when `input_ids` are present): The text embeddings which are basically the pooled output of [`FlavaTextModel`]. text_output (`BaseModelOutputWithPooling`, *optional*, returned when `input_ids` are present): The output of the [`FlavaTextModel`]. multimodal_embeddings (`torch.FloatTensor` of shape `(batch_size, output_dim)`, *optional*, returned when `input_ids` and `pixel_values` are present and `skip_unmasked_multimodal_encoder` is `None` or `False`): The multimodal embeddings which are basically the pooled output of [`FlavaTextModel`]. multimodal_output (`BaseModelOutputWithPooling`, returned when `input_ids` and `pixel_values` are present and `skip_unmasked_multimodal_encoder` is `None` or `False`): The output of the [`FlavaMultimodalModel`]. image_masked_embeddings (`torch.FloatTensor` of shape `(batch_size, output_dim)`, *optional*, returned when `pixel_values` are present): The image embeddings which are basically the pooled output of [`FlavaImageModel`]. Uses `bool_masked_pos` to create masked images. image_masked_output (`BaseModelOutputWithPooling`, *optional*, returned when `pixel_values` are present): The output of the [`FlavaImageModel`]. Uses `bool_masked_pos` to create masked images. text_masked_embeddings (`torch.FloatTensor` of shape `(batch_size, output_dim)`, *optional*, returned when `input_ids_masked` are present): The text embeddings which are basically the pooled output of [`FlavaTextModel`]. text_masked_output (`BaseModelOutputWithPooling`, *optional*, returned when `input_ids_masked` are present): The output of the [`FlavaTextModel`]. multimodal_masked_embeddings (`torch.FloatTensor` of shape `(batch_size, output_dim)`, *optional*, returned when `input_ids` and `pixel_values` are present): The multimodal embeddings which are basically the pooled output of [`FlavaTextModel`]. multimodal_masked_output (`BaseModelOutputWithPooling`, *optional*, returned when `input_ids_masked` and `pixel_values` are present): The output of the [`FlavaMultimodalModel`]. mim_logits (`torch.FloatTensor` of shape `(batch_size, num_image_patches, image_vocab_size)` or of shape `(total_masked_patches, image_vocab_size)` , *optional*, returned when `pixel_values` are present and `input_ids_masked` are not): The logits for MIM unimodal loss. Uses `book_masked_pos` to get masked patches. The flattened output is returned when `bool_masked_pos` has some of the patches masked. mlm_logits (`torch.FloatTensor` of shape `(batch_size, text_seq_length, text_vocab_size)` or of shape `(total_masked_seq_length, text_vocab_size)`, *optional*, returned when `input_ids_masked` are present and `pixel_values` are not): The logits for MLM unimodal loss. The flattened output is returned when `input_ids_masked` has some of the tokens masked. itm_logits (`torch.FloatTensor` of shape `(batch_size, 2)`, *optional*, returned when `input_ids_masked` and `pixel_values` are present): The logits for ITM loss. Note that ITM loss is calculated on masked pairs in FLAVA. contrastive_logits_per_image (`torch.FloatTensor` of shape `(image_batch_size, text_batch_size)`): The scaled dot product scores between `image_embeddings` and `text_embeddings` but passed through FLAVA's `image_projection` and `text_projection` layers respectively. This represents the image-text similarity scores. This is calculated on unmasked images and texts. contrastive_logits_per_text (`torch.FloatTensor` of shape `(text_batch_size, image_batch_size)`): The scaled dot product scores between `text_embeddings` and `image_embeddings` but passed through FLAVA's `text_projection` and `image_projection` layers respectively. This is calculated on unmasked images and texts. mmm_image_logits (`torch.FloatTensor` of shape `(batch_size, num_image_patches, image_vocab_size)` or of shape`(total_masked_patches, image_vocab_size)`, *optional*, returned when `pixel_values` and `input_ids_masked` are present): The logits for MMM image multimodal loss. Uses `book_masked_pos` to get masked patches. The flattened output is returned when `bool_masked_pos` has some of the patches masked. mmm_text_logits (`torch.FloatTensor` of shape `(batch_size, text_seq_length, text_vocab_size)` or of shape `(`(total_masked_seq_length, text_vocab_size)`), *optional*, returned when `pixel_values` and `input_ids_masked` are present): The logits for MMM text multimodal loss. The flattened output is returned when `input_ids_masked` has some of the tokens masked. Nloss loss_inforr r!r"r#r$image_masked_embeddingsimage_masked_outputtext_masked_embeddingstext_masked_outputmultimodal_masked_embeddingsmultimodal_masked_output mim_logits mlm_logits itm_logitscontrastive_logits_per_imagecontrastive_logits_per_textmmm_image_logitsmmm_text_logitsr%cjgdtfdDS)N)r"r r$rQrOrSc3tK|]2}|vr|n!t|V3dSNr()r+r,r-transformer_outputss r.r/z5FlavaForPreTrainingOutput.to_tuple..sNssbc)< < rr r r!r"r#r$rNrOrPrQrRrSrTrUrVrWrXrYrZr2rr*r<r0r.rKrKs955n&*D% d ")))!I{!!!15e'$.5556:L,t3:::04OU&-44459K+d29996:5,t3:::;?1D8???8<U.5<<<=A3d:AAA7;E-4;;;<@2T9@@@=A %"3d":AAABF84?FFF+/J!D(///+/J!D(///+/J!D(///=A %"3d":AAA<@!2T!9@@@15e'$.55504OU&-444 t%* t t t t t tr0rKc eZdZdZddededdffd Zdejd e d e dejfd Z dd ejd ej dzdedejfdZ xZ S)FlavaImageEmbeddingszb Construct the CLS token, position and patch embeddings. Optionally, also the mask token. Fconfiguse_mask_tokenr%Ncft|p|j}tjt jdd|j|_|r-tjt jdd|jnd|_t|j |j |j |j|_ |j j}tjt jd|dz|j|_tj|j|_|j |_ ||_dS)Nr) image_size patch_size num_channels embed_dim)super__init__ mask_tokenr Parameterr9zeros hidden_size cls_tokenPatchEmbeddingsrdrerfpatch_embeddings num_patchesposition_embeddingsDropouthidden_dropout_probdropoutra)r-rarbrq __class__s r.rizFlavaImageEmbeddings.__init__s '<6+<ek!Q8J&K&KLLQ_i",u{1a9K'L'LMMMei /((,( ! ! !  +7 #%< A{QPVPb0c0c#d#d z&"<==  + r0 embeddingsheightwidthc|jddz }|jjddz }tjs||kr ||kr|jS|jddddf}|jddddf}|jd}||jz} ||jz} t |dz} |d| | |}|dddd}tj || | fdd }|dddd dd|}tj ||fd S) a This method allows to interpolate the pre-trained position encodings, to be able to use the model on higher resolution images. This method is also adapted to support torch.jit tracing. Adapted from: - https://github.com/facebookresearch/dino/blob/de9ee3df6cf39fac952ab558447af1fa1365362a/vision_transformer.py#L174-L194, and - https://github.com/facebookresearch/dinov2/blob/e1277af2ba9496fbadf7aec6eba56e8d882d1e35/dinov2/models/vision_transformer.py#L179-L211 rNg?rrbicubicF)sizemode align_cornersdim)shaperrr9jit is_tracingrerreshapepermuter functional interpolateviewcat) r-rwrxryrq num_positionsclass_pos_embedpatch_pos_embedr new_height new_widthsqrt_num_positionss r.interpolate_pos_encodingz-FlavaImageEmbeddings.interpolate_pos_encodingsr!&q)A- 06q9A= y##%% ,+*F*F6UZ??+ +2111bqb592111abb59r"t. T_, &}c'9::)11!5GI[]`aa)11!Q1==-33 i( 4  *11!Q1==BB1b#NNy/?;CCCCr0 pixel_valuesbool_masked_posrc|j\}}}}|||}|\}} } ||j|| d} |dkr)||dd}|d| } |d| z z| | zz}|j |dd} tj | |fd}|r|| |||z}n ||j z}||}|S)N)rr{rrg?rr)rrpr~rjexpandrr unsqueezetype_asrnr9rrrrru)r-rrr batch_sizerfrxryrwseq_len_ mask_tokensmask cls_tokenss r.forwardzFlavaImageEmbeddings.forward*sY 3?2D/ L&%**> Y J7Q???  $ ?#d&C&CJPVX]&^&^^JJ#d&>>J\\*-- r0FNF)r5r6r7r8rrIrir9Tensorintr BoolTensorr __classcell__rvs@r.r`r`s/RV&&D5<&D&DUX&D]b]i&D&D&D&DV48). l)D0#'   r0r`c eZdZdZ ddeeezeeefzdeeeefzded effd Zdd ej d e dej fdZ xZ S)roz# Image to Patch Embedding. rrdrerfrgc~tt|tjjs||f}t|tjjs||f}|d|dz|d|dzz}||_||_||_tj |||||_ dS)Nrr) kernel_sizestride) rhri isinstance collectionsabcIterablerdrerqrConv2d projection)r-rdrerfrgrqrvs r.rizPatchEmbeddings.__init__Ss *ko&>?? 2$j1J*ko&>?? 2$j1J!!} 1 5*Q-:VW=:XY $$&)L)\fgggr0Frrr%c B|j\}}}}|sT||jdks||jdkr2td|d|d|jdd|jdd ||ddd}|S)NrrzInput image size (*z) doesn't match model (z).r|)rrd ValueErrorrflatten transpose)r-rrrrfrxryxs r.rzPatchEmbeddings.forwardfs2>2D/ L&%' +++u8J/J/J EEE%EE+EE.2oa.@EEE OOL ) ) 1 1! 4 4 > >q! D Dr0)rrrrr) r5r6r7r8rlistr2rir9rrIrrrs@r.roroNs 9<,. hh$s)OeCHo5h%S/)h h  hhhhhh&  EL D ]b]i        r0rocneZdZdZfdZ ddejdzdejdzdejdzfdZxZS) FlavaTextEmbeddingszGConstruct the embeddings from word, position and token_type embeddings.cttj|j|j|j|_tj|j|j|_ tj|j |j|_ tj |j|j |_ tj|j|_|dt%j|jdd|dt%j|jt$jddS) N) padding_idxeps position_idsrr{F) persistenttoken_type_ids)dtype)rhrir Embedding vocab_sizerm pad_token_idword_embeddingsmax_position_embeddingsrrtype_vocab_sizetoken_type_embeddings LayerNormlayer_norm_epsrsrtruregister_bufferr9arangerrlrr~longr-rarvs r.rizFlavaTextEmbeddings.__init__us3 !|F,=v?Q_e_rsss#%<0NPVPb#c#c %'\&2H&J\%]%]"f&8f>STTTz&"<==   EL)GHHOOPWXXej      ek$*;*@*@*B*B%*UUUbg      r0N input_idsrrc,|}|d}||jddd|f}|mt|dr2|jddd|f}||d|}|}n+t j|t j|jj}| |}| |} || z} | |} | | z } | | } | | } | S)Nrrr)rdevice)r~rhasattrrrr9rlrrrrrrrru) r-rrr input_shape seq_lengthbuffered_token_type_ids buffered_token_type_ids_expanded inputs_embedsrrwrrs r.rzFlavaTextEmbeddings.forwards.  nn&&  ^  ,QQQ  ^ J J"%:: "66|DD)) ^^J// \\*-- r0NNN) r5r6r7r8rir9rrrrs@r.rrrsQQ     $*..2,0  <$&  t+ lT)        r0rc eZdZdeddffd Z d dejdejdzdedeejejfeejzfd Z xZ S) FlavaSelfAttentionrar%Nct|j|jzdkr0t |ds t d|jd|jd|j|_t |j|jz |_|j|jz|_tj |j|j|j |_ tj |j|j|j |_ tj |j|j|j |_tj|j|_dS)Nrembedding_sizezThe hidden size z4 is not a multiple of the number of attention heads .bias)rhrirmnum_attention_headsrrrattention_head_size all_head_sizerLinearqkv_biasquerykeyvaluersattention_probs_dropout_probrurs r.rizFlavaSelfAttention.__init__s.    : :a ? ?PVXhHiHi ?76#5773777  $*#= #&v'9F>HHANN jj//44\BBLLQPQRR !< Y5H5HR5P5PQQ+di8P.Q.QQ  %/.@ -//0@b/II,,77 _kBB %--aAq99DDFF "/"4"4"6"6ss";t?Q>S"S* *,CD 6G]=/22mM]r0r r5r6r7FlavaPossibleConfigsrir9rrIr2rrrs@r.rrsG3GGGGGGG*/3"' ##|# t+# # u|U\) *U5<-@ @ ########r0rc^eZdZdZdeddffd ZdejdejdejfdZxZ S) FlavaSelfOutputz The residual connection is defined in FlavaLayer (same as ViTLayer) instead of here (as is the case with other models), due to the layernorm applied before each block. rar%Ncttj|j|j|_tj|j|_dSr]) rhrirrrmdensersrtrurs r.rizFlavaSelfOutput.__init__sJ Yv163EFF z&"<== r0r input_tensorcZ||}||}|Sr]rrur-rrs r.rzFlavaSelfOutput.forwards* =11  ]33 r0) r5r6r7r8rrir9rrrrs@r.rrs >3>>>>>>> U\RWR^r0rc eZdZdeddffd Z d dejdejdzdedeejejfeejzfd Z xZ S) FlavaAttentionrar%Nctt||_t ||_dSr])rhrir attentionroutputrs r.rizFlavaAttention.__init__s; +F33%f-- r0Frrrc||||}||d|}|f|ddz}|SN)rrrr)r r)r-rrr self_outputsattention_outputrs r.rzFlavaAttention.forwardsY ~~ .L]&   ;;|A FF#% QRR(88r0rrrs@r.r r s.3......./3"'  |  t+   u|U\) *U5<-@ @        r0r cLeZdZdeddffd ZdejdejfdZxZS)FlavaIntermediaterar%Ncttj|j|j|_t|jtrt|j|_ dS|j|_ dSr]) rhrirrrmintermediate_sizerr hidden_actstrr intermediate_act_fnrs r.rizFlavaIntermediate.__init__ sn Yv163KLL f' - - 9'-f.?'@D $ $ $'-'8D $ $ $r0rcZ||}||}|Sr])rrr-rs r.rzFlavaIntermediate.forwards, =11 00?? r0 r5r6r7rrir9rrrrs@r.rr sr939999999U\elr0rcZeZdZdeddffd ZdejdejdejfdZxZS) FlavaOutputrar%Ncttj|j|j|_tj|j|_ dSr]) rhrirrrrmrrsrtrurs r.rizFlavaOutput.__init__sJ Yv79KLL z&"<== r0rrcd||}||}||z}|Sr]rr s r.rzFlavaOutput.forward!s4 =11  ]33 % 4 r0rrs@r.rrs}>3>>>>>>> U\RWR^r0rc eZdZdZdeddffd Z d dejdejdzd ede ejejfe ejzfd Z xZ S) FlavaLayerz?This corresponds to the Block class in the timm implementation.rar%Nczt|j|_d|_t ||_t ||_t||_ tj |j |j |_tj |j |j |_dS)Nrr)rhrichunk_size_feed_forward seq_len_dimr r r intermediaterrrrrmrlayernorm_beforelayernorm_afterrs r.rizFlavaLayer.__init__-s '-'E$'//-f55!&)) !# V-?VEZ [ [ [!|F,>FDYZZZr0Frrrc|||||}|d}|dd}||z}||}||}|||}|f|z}|Sr)r r'r(r&r)r-rrrself_attention_outputsrr layer_outputs r.rzFlavaLayer.forward9s "&  ! !- 0 0)/"0" "  2!4(,)=8 ++M:: ((66 {{<?? /G+r0r) r5r6r7r8rrir9rrIr2rrrs@r.r"r"*sII [3 [ [ [ [ [ [ [/3"' | t+  u|U\) *U5<-@ @ r0r"creZdZdeddffd Z d dejdejdzd ed ed edee zf d Z xZ S) FlavaEncoderrar%Nct|_tjfdt jD|_d|_dS)Nc.g|]}tSr<)r")r+rras r. z)FlavaEncoder.__init__..Zs!#`#`#`1Jv$6$6#`#`#`r0F) rhrirar ModuleListrangenum_hidden_layerslayergradient_checkpointingrs `r.rizFlavaEncoder.__init__Ws`  ]#`#`#`#`fF^@_@_#`#`#`aa &+###r0FTrrroutput_hidden_states return_dictc|rdnd}|rdnd}t|jD]0\}} |r||fz}| |||} | d}|r || dfz}1|r||fz}|std|||fDSt|||S)Nr<rrc3K|]}||V dSr]r<)r+rHs r.r/z'FlavaEncoder.forward..ws(mmq_`_l_l_l_l_lmmr0)last_hidden_stater attentions) enumerater4r2r ) r-rrrr6r7all_hidden_statesall_self_attentionsi layer_module layer_outputss r.rzFlavaEncoder.forward]s#7@BBD$5?bb4(44 P POA|# I$58H$H!(LHYZZM)!,M  P&9]1=MSTTT->Hk&)))D  r0r%c|jjSr]rwrpr4s r.get_input_embeddingsz$FlavaImageModel.get_input_embeddingss //r0rc||j_dSr]rnr-rs r.set_input_embeddingsz$FlavaImageModel.set_input_embeddingss+0(((r0Nrrrrr6r7c ||n |jj}||n |jj}||n |jj}|t d||||} || ||||} | d} || } |j|| nd} |s| | f| ddzSt| | | j | j S)z bool_masked_pos (`torch.BoolTensor` of shape `(batch_size, image_num_patches)`): Boolean masked positions. Indicates which patches are masked (1) and which aren't (0). Nz You have to specify pixel_values)rrrrr6r7rrr: pooler_outputrr;) rarr6r7rrwrhrirjr rr;) r-rrrrrr6r7kwargsembedding_outputencoder_outputssequence_outputrJs r.rzFlavaImageModel.forwards/ 2C1N--TXT_Tq$8$D $+Jj &1%STTT->Hk&)))D  r0r%c|jjSr]rwrr4s r.roz#FlavaTextModel.get_input_embeddingss ..r0rc||j_dSr]rrqs r.rrz#FlavaTextModel.set_input_embeddingss*/'''r0Nrrrrrr6r7c ||n |jj}||n |jj}||n |jj}|t d||||} t |j| |}|| ||||} | d} || } |j | | nd} |s| | f| ddzSt| | | j | j S) a input_ids (`torch.LongTensor` of shape `(batch_size, text_seq_length)`): Indices of input sequence tokens in the vocabulary. Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and [`PreTrainedTokenizer.__call__`] for details. [What are input IDs?](../glossary#input-ids) token_type_ids (`torch.LongTensor` of shape `(batch_size, text_seq_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. [What are token type IDs?](../glossary#token-type-ids) NzYou have to specify input_ids)rrrrarrrtrrru) rarr6r7rrwr rhrirjr rr;) r-rrrrrr6r7rwrxryrzrJs r.rzFlavaTextModel.forwardsQ02C1N--TXT_Tq$8$D $+Jj &1%STTT->Hk&)))D  r0Nrrr6r7r%c F||n |jj}||n |jj}||n |jj}|\}}} |jr9|j|dd} tj | |fd}|dz }t|j||}| |||||} | d} | | } |j | | nd} |s| | f| ddzSt| | | j| jS) z hidden_states (`torch.FloatTensor` of shape `(batch_size, image_num_patches + text_seq_len, hidden_size)`): The concatenated hidden states of unimodal encoders. Nr{rrrrtrru)rarr6r7r~rXrnrr9rr rhrirjr rr;)r-rrrr6r7rwrrrrryrzrJs r.rzFlavaMultimodalModel.forwardssv2C1N--TXT_Tq$8$D $+Jj &1%> 0+,,000  &-/?@@ 1,--111  &24IJJ _AV%= > >AAAB  ( * "4$3 + 7!-!9/;(55*<88 45F G G " $*@$BU V V!y)>@STT< T[5W(X(XYY&(i0FH[&\&\#%'Yt/DdFY%Z%Z" r0Nrrrrrwr%c n|jd||||dd|}|j}|||_|S)a  input_ids (`torch.LongTensor` of shape `(batch_size, text_seq_length)`): Indices of input sequence tokens in the vocabulary. Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and [`PreTrainedTokenizer.__call__`] for details. [What are input IDs?](../glossary#input-ids) token_type_ids (`torch.LongTensor` of shape `(batch_size, text_seq_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. [What are token type IDs?](../glossary#token-type-ids) Examples: ```python >>> import torch >>> from transformers import AutoProcessor, FlavaModel >>> model = FlavaModel.from_pretrained("{0}") >>> processor = AutoProcessor.from_pretrained("{0}") >>> inputs = processor( ... text=["a photo of a cat", "a photo of a dog"], max_length=77, padding="max_length", return_tensors="pt" ... ) >>> with torch.inference_mode(): ... text_features = model.get_text_features(**inputs) ``` T)rrrrr7r<)rr:rrv)r-rrrrrw text_outputsr:s r.get_text_featureszFlavaModel.get_text_featuressdL4C4?4 ))% 4 4  4 4 ):%)%9%9:K%L%L "r0rrrc n|jd||||dd|}|j}|||_|S)a  bool_masked_pos (`torch.BoolTensor` of shape `(batch_size, image_num_patches)`): Boolean masked positions. Indicates which patches are masked (1) and which aren't (0). Examples: ```python >>> import torch >>> from transformers import AutoProcessor, FlavaModel >>> from transformers.image_utils import load_image >>> model = FlavaModel.from_pretrained("{0}") >>> processor = AutoProcessor.from_pretrained("{0}") >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg" >>> image = load_image(url) >>> inputs = processor(images=image, return_tensors="pt") >>> with torch.inference_mode(): ... image_features = model.get_image_features(**inputs) ``` T)rrrrr7r<)rr:rrv)r-rrrrrw image_outputsr:s r.get_image_featureszFlavaModel.get_image_features seB5ED4D5 %+)%= 5 5  5 5 *;&*&;&;>> from PIL import Image >>> import httpx >>> from io import BytesIO >>> from transformers import AutoProcessor, FlavaModel >>> model = FlavaModel.from_pretrained("facebook/flava-full") >>> processor = AutoProcessor.from_pretrained("facebook/flava-full") >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg" >>> with httpx.stream("GET", url) as response: ... image = Image.open(BytesIO(response.read())) >>> inputs = processor(text=["a photo of a cat"], images=image, return_tensors="pt", padding=True) >>> outputs = model(**inputs) >>> image_embeddings = outputs.image_embeddings >>> text_embeddings = outputs.text_embeddings >>> multimodal_embeddings = outputs.multimodal_embeddings >>> outputs.image_embeddings.shape torch.Size([1, 197, 768]) >>> text_embeddings.shape torch.Size([1, 7, 768]) >>> multimodal_embeddings.shape torch.Size([1, 205, 768]) ``` NzRFLAVA model requires hidden states to work. Please set `output_hidden_states=True`)rrrrr6r7rr|r{)rrrrrr6r7rrr)rr7)rr r!r"r#r$)rar7rrrrrrrrXr9onesrrr)r-rrrrrrrrrr6r7rwr image_statesimage_mm_projectionr r! text_statestext_mm_projectionr"r#r$rrrattention_mask_imageattention_multimodalmultimodal_inputs r.rzFlavaModel.forward:s?L&1%;q>[O!%!;!;KO!L!L  $   */A/MVm/M))<)B& GQ(6!qLG',z*gNaNh'i'i'i$',y2F1W]^'_'_'_$$'+$$y*=?Q)RXYZZZ  $ 5 5 1ES^!6!! %6a$8 !  %!   -%+#"7/     r0r) NNNNNNNNNTN)r5r6r7rr;rirrr9rrrr2r rrrIr LongTensorr:rrrrs@r.rYrYs ){))))))V/3.2,0 //</ t+/ t+ / lT) / +, / + +///^/b4804.2 **l*)D0*#'+ *  t+ * +, * + +***^*X.215.2.2/30448/3)-%)#'N N #d*N '$.N  t+ N  t+ N , N &-N $lT1N "&N  $;N #N D[N  ! !N N N ^N N N N N r0rYcLeZdZdedeffd ZdejdejfdZxZS)FlavaImageCodebookResPathin_sizeout_sizec (t|dz}t}tj|d<tj||dd|d<tj|d<tj||dd|d<tj|d <tj||dd|d <tj|d <tj||dd |d <tj||_dS)Nrelu_1rrrpaddingconv_1relu_2conv_2relu_3conv_3relu_4rconv_4)rhrirrReLUr Sequentialpath)r-rrrwhid_sizerrvs r.riz"FlavaImageCodebookResPath.__init__s q=}}X7H!QOOOXX8X1aPPPXX8X1aPPPXX8X1aPPPXM$'' r0rr%c,||Sr])rr-rs r.rz!FlavaImageCodebookResPath.forwardsyy||r0 r5r6r7rrir9rrrrs@r.rrsq((s(((((( %,r0rcPeZdZdededeffd ZdejdejfdZxZS)FlavaImageCodebookBlockrr num_layersc td|dzz |_||krtj||dd|_ntj|_t|||_dS)Nrr|rr) rhri post_gainrrid_pathIdentityrres_path)r-rrrrwrvs r.riz FlavaImageCodebookBlock.__init__sr j!m, h  9WhAqQQQDLL;==DL1'8DD r0rr%ch|||j||zzSr])rrrrs r.rzFlavaImageCodebookBlock.forwards*||A$--2B2B!BBBr0rrs@r.rrs E Es E E E E E E ECC%,CCCCCCCCr0rc ZeZdZd dededededef fd Zdejd ejfd ZxZ S) FlavaImageCodebookLayerGroupT num_blocksrrruse_poolcdtt}t|D]=}|dkrt ||||d|dz<#t ||||d|dz<>|rt jd|d<t j||_dS)Nrblock_rr|)rpool) rhrirr2rr MaxPool2drgroup) r-rrrrrblocksr?rvs r.riz%FlavaImageCodebookLayerGroup.__init__s z"" c cAAvv+B7HV`+a+a'A''((+B8XWa+b+b'A''((  9\a888F6N]6** r0rr%c,||Sr])rrs r.rz$FlavaImageCodebookLayerGroup.forwardszz!}}r0r{) r5r6r7rrIrir9rrrrs@r.rrs + +3 +C +# +QT +`d + + + + + +%,r0ra" The FLAVA's image codebook model inspired from DALL-E's original encoder. Outputs raw hidden states and can be used to generate image tokens for an image based on DALL-E's vocab. Used to generate labels for MIM. Use `get_codebook_indices` to get image tokens for an image. ceZdZUdZeed<dZdZdZdede ffd Z de j de j fd Z de j de j fd Zde jde j fd ZxZS) FlavaImageCodebookmodelrarrcFrwc &t|||_|j|_|j|_|j|_|j|_|j|_|j|jz}t}tj |d<tj d|jz|jdd|d<t}tj |jd|jzdd|d <t|j|d|jzd|jz|d <t|j|d|jzd |jz|d <t|j|d |jzd |jz|d<t|j|d |jzd|jzd|d<tj ||d<tj ||_||jjr|D] }d|_ dSdS)Nrelurrrconvrinputgroup_1r|group_2rgroup_3F)rgroup_4r)rhrira num_groupsinput_channelsnum_blocks_per_grouprmrrrrrrrrrkfreeze parameters requires_grad)r-rarwr output_blocksrparamrvs r.rizFlavaImageCodebook.__init__s&      +$3$*$?!!- +_t'@@ # " f " !d.>*>]^hi j j j f)D$7T=M9M[\fghhhw8  %z1t7G3GTM]I]  y9  %z1t7G3GTM]I]  y9  %z1t7G3GTM]I]  y9  %z1t7G3GTM]I]hm   y=77xmF++   ;  ,** , ,&+## , , , ,r0r%c~dtdtd||}tj|dS)NaI Args: pixel_values (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)`): Pixel values. Codebook pixel values can be obtained using [`AutoImageProcessor`] by passing `return_codebook_pixels=True`. See [`FlavaImageProcessor.__call__`] for details. Examples: ```python >>> from PIL import Image >>> import httpx >>> from io import BytesIO >>> from transformers import AutoImageProcessor, FlavaImageCodebook >>> model = FlavaImageCodebook.from_pretrained("E") >>> image_processor = AutoImageProcessor.from_pretrained("a") >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg" >>> with httpx.stream("GET", url) as response: ... image = Image.open(BytesIO(response.read())) >>> inputs = image_processor([image], return_codebook_pixels=True, return_tensors="pt") >>> inputs = dict(pixel_values=inputs.codebook_pixel_values) >>> outputs = model.get_codebook_indices(**inputs) ``` r)axis)_CHECKPOINT_FOR_CODEBOOK_DOCrr9argmaxr-rz_logitss r.get_codebook_indicesz'FlavaImageCodebook.get_codebook_indices@sZ :V  D`    4;;|,,|H1----r0ch||}tjd|S)Nrr)rrSoftmaxrs r.get_codebook_probsz%FlavaImageCodebook.get_codebook_probs^s0;;|,, rza   ***r0c (dtdtdt|jdkrtd|jd|jd|jkr%td|jdd |j||S) NaJ Args: pixel_values (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)`): Pixel values. Codebook pixel values can be obtained using [`AutoImageProcessor`] by passing `return_codebook_pixels=True`. See [`FlavaImageProcessor.__call__`] for details. Examples: ```python >>> from PIL import Image >>> import httpx >>> from io import BytesIO >>> from transformers import AutoImageProcessor, FlavaImageCodebook >>> model = FlavaImageCodebook.from_pretrained("ra") >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg" >>> with httpx.stream("GET", url) as response: ... image = Image.open(BytesIO(response.read())) >>> inputs = image_processor([image], return_codebook_pixels=True, return_tensors="pt") >>> inputs = dict(pixel_values=inputs.codebook_pixel_values) >>> outputs = model(**inputs) >>> print(outputs.shape) (1, 196) ``` rz input shape z is not 4drz input has z channels but model built for )rlenrrrr)r-rrws r.rzFlavaImageCodebook.forwardbs :V  D`    : |! " "a ' 'JL,>JJJKK K  a D$7 7 7t,*STTTr0c||}||}||}|Sr])rrrrs r.rz$FlavaPredictionHeadTransform.forwards= =11 --m<< }55 r0r5r6r7rirrrs@r.rrsLUUUUUr0rc&eZdZdfd ZdZxZS)rSNcJt||_t||_t j|j|jd|_ t j tj |j|_ |||j _dSdS)NTr)rhrirar transformrrrmrdecoderrkr9rlrweight)r-rarrvs r.riz"FlavaMaskedPredictionHead.__init__s  5f==y!3V5FTRRR LV->!?!?@@  "(DL     r0cZ||}||}|Sr])rrrs r.rz!FlavaMaskedPredictionHead.forwards' NN1   LLOOr0r]rrs@r.rSrSsL))))))r0rSc$eZdZfdZdZxZS) FlavaITMHeadct||_t||_t j|jd|_dS)Nr|) rhrirarCrjrrrmseq_relationshiprs r.rizFlavaITMHead.__init__sL  !&)) " &*.s$'e'e'ea(8(I(I'e'e'er0c8g|]}tjSr<r)r+rrs r.r0z6FlavaGlobalContrastiveHead.forward..s%&e&e&eau'78H'I'I&e&e&er0r)r9exp distributed is_availableis_initializedrr~rget_world_sizerrr all_gatherr2get_rankrrr) r-rr!r[ temperaturelabelsimage_embeddings_alltext_embeddings_alllocal_batch_size world_sizelogits_per_imagelogits_per_texts `` r.rz"FlavaGlobalContrastiveHead.forwards2i ,,  --// u7H7W7W7Y7Y \"2"7"7":":CSCZ[[[F$4#5 #2"3  /44Q77 *99;;J/ S(-'8';'F'Q'QRb'c'c$&+&7&:&E&P&PQ`&a&a##'e'e'e'eSXYcSdSd'e'e'e$&e&e&e&eSXYcSdSd&e&e&e#!,,-ACSTTT!,,-@/RRR%(9(B(B(D(DDu| )9)@HHHF %y)=>>#i(;<< <(8:M:W:WXY[\:]:]^^all,8L8V8VWXZ[8\8\]]`kk&88r0rrs@r.rrsLNNNNN 9999999r0rzk The FLAVA model for pretraining which outputs losses, embeddings, logits and transformer outputs. c'eZdZdddddZd dedejdzffd Zd ej fd Z e d!d ej dzdej dzdej dzdej dzdej dzdej dzdej dzdej dzdej dzdedzdej dzdej dzdej dzdedzdededzdedzdeej ezf$dZxZS)"FlavaForPreTrainingzmmm_text_head.decoder.biaszmim_head.decoder.biaszmlm_head.decoder.biaszmmm_image_head.decoder.bias)zmmm_text_head.biasz mim_head.biasz mlm_head.biaszmmm_image_head.biasNraimage_codebookct|t||_||_|j |jrt |j|_t|j |_ t|j |_ t||_t|j |_t|j |_t#||_|j j|_|j j|_|j|_|j|_|j|_|j|_|j|_|j|_|j|_|j|_|dS)z image_codebook ([`nn.Module`]): If passed, the image codebook will be set to this. Otherwise, it will be initialized using the image_codebook_config defined in the config first as the first parameter. N)rhrirYrMr( init_codebookrimage_codebook_configrSrmim_headrmlm_headr itm_headmmm_image_head mmm_text_headrglobal_contrastive_headrimage_vocab_sizetext_vocab_size mlm_weight mim_weightglobal_contrastive_weightce_ignore_index itm_weightmmm_image_weightmmm_text_weight skip_unmasked_multimodal_encoderrk)r-rar(rvs r.rizFlavaForPreTraining.__init__sJ    '' ,   &6+? &"4V5Q"R"RD 2&2EFF 1&2DEE $V,, 78KLL6v7IJJ'A&'I'I$ & 3 >%1< + +)/)I&%5 + & 7%5060W- r0rc|dkr)||dd}|S)Nr|rr{)rrr~rs r. _resize_to_2dz!FlavaForPreTraining._resize_to_2ds5 5577Q;;qvvayy"%%Ar0Trinput_ids_maskedrcodebook_pixel_valuesrrrrrr; mlm_labels mim_labels itm_labelsrr6r7 return_lossr%c ||n |jj}||n |jj}| | n|j} ||td|}||||||| | ||d }|||||| |||d }d}|j}|j}|j}|j}|j }dx}x}x}x}x}x} }!dx}"x}#x}$}%dx}&x}'}(||E| C|rA|j td|td|j |} |jdkr(|%|"|})| || } ||}|j| |d<|)dd| d  dddf})| |j}*| |*}+|)|*ddf})||)}"|rVt(j|"d |j|+d }||jz}n||)}"|jdkr|||},| || } |,dd| d  dddf},| |j}*| |*}-|,|*ddf},||,}#|rVt(j|#d |j|-d }||jz}n||,}#|jdkr|||}&| | d}.|.|.z}|r*t(j|&| }!|!|jz}!|||}| | |} | | |} ||}|4|jdkr(|})|d d z }/|)ddd d |/zddf})| || } ||}|j| |d<| |j}*| |*}+|)|*ddf})| |)}%|rVt(j|%d |j|+d }||jz}n| |)}%||j!dkr|},|,dd|d  dddf},| || } | |j}*| |*}-|,|*ddf},|"|,}$|rVt(j|$d |j|-d }||j!z}n|"|,}$|h|e|j#dkrY|j$|dddddf}0t(j%|0d }0|j&|dddddf}1t(j%|1d }1|j'r/|jj(j)*tVtX|-|1|0|jj(\}'}(}2||'|}'|(|}(|2|}2|rRt(j|'|2}3t(j|(|2}4|3|4zd z } | |j#z} t]|||!| || }5|r?|5/s+tad|51D}|s||j2|j23nd||j4|j43nd|j |j5|j53nd||j2|j23nd||j4|j43nd||j5|j53nd|"|#|&|'|(|%|$f}6|r|5/s||5f|6z}6tmd|6DStod%id|d|5d|d|j2d|d|j4d|j d|j5d|d|j2d|d|j4d|d|j5d|"d|#d |&d!|'d"|(d#|%d$|$S)&a input_ids (`torch.LongTensor` of shape `(batch_size, text_seq_len)`): Indices of input sequence tokens in the vocabulary. Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and [`PreTrainedTokenizer.__call__`] for details. [What are input IDs?](../glossary#input-ids) input_ids_masked (`torch.LongTensor` of shape `(batch_size, text_seq_len)`): Indices of input sequence tokens in the vocabulary. These ones are the masked version of the original task to be used with MLM. Indices can be obtained using [`AutoTokenizer`] along with [`DataCollatorForMaskedLanguageModeling`]. See [`PreTrainedTokenizer.encode`] and [`PreTrainedTokenizer.__call__`] for details. [What are input IDs?](../glossary#input-ids) codebook_pixel_values (`torch.FloatTensor` of shape `(batch_size, num_image_patches, patch_size, patch_size, 3)`, *optional*): Pixel values for image patches that are used to compute the image codebook labels for masked image modeling. token_type_ids (`torch.LongTensor` of shape `(batch_size, text_seq_len)`, *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. [What are token type IDs?](../glossary#token-type-ids) bool_masked_pos (`torch.BoolTensor` of shape `(batch_size, image_num_patches)`): Boolean masked positions. Indicates which patches are masked (1) and which aren't (0). image_attention_mask (`torch.FloatTensor` of shape `(batch_size, image_num_patches)`, *optional*): Mask to avoid performing attention on padding token indices specifically for images. Mask values selected in `[0, 1]`: - 1 for tokens that are **not masked**, - 0 for tokens that are **masked**. [What are attention masks?](../glossary#attention-mask) skip_unmasked_multimodal_encoder (*bool*, *optional*): Skip any calculations for multimodal encoder for unmasked inputs. FLAVA pretraining doesn't need unmasked multimodal embeddings or outputs as of now. mlm_labels (`torch.LongTensor` of shape `(batch_size, text_seq_len)`, *optional*): Labels for computing the left-to-right language and multimodal masked modeling loss (next word prediction). Indices should be in `[-100, 0, ..., text_config.vocab_size - 1]` (see `input_ids` docstring). Tokens with indices set to `-100` are ignored (masked), the loss is only computed for the tokens with labels in `[0, ..., text_config.vocab_size - 1]`. mim_labels (`torch.LongTensor` of shape `(batch_size, image_num_patches)`, *optional*): Labels for computing the image and multimodal masked modeling loss. Indices should be in `[-100, 0, ..., image_config.vocab_size - 1]`. Tokens with indices set to `-100` are ignored (masked), the loss is only computed for the tokens with labels in `[0, ..., image_config.vocab_size - 1]`. If not passed, they are generated automatically using the image codebook assigned to the model. By default, it uses [`FlavaImageCodebook`]. See [`FlavaImageCodebook`] to understand how to generate mim_labels. itm_labels (`torch.LongTensor` of shape `(batch_size, 1)`, *optional*): Labels for computing the image-text matching loss. 0 means the pairs don't match and 1 means they match. The pairs with 0 will be skipped for calculation of MMM and global contrastive losses as well. return_loss (`bool`, *optional*, default to None): Whether to return calculated loss or not. Examples: ```python >>> from PIL import Image >>> import httpx >>> from io import BytesIO >>> from transformers import FlavaForPreTraining, AutoProcessor >>> url = "http://images.cocodataset.org/val2017/000000039769.jpg" >>> with httpx.stream("GET", url) as response: ... image = Image.open(BytesIO(response.read())) >>> model = FlavaForPreTraining.from_pretrained("facebook/flava-full") >>> processor = AutoProcessor.from_pretrained("facebook/flava-full") >>> text = ["a photo of a cat"] >>> inputs = processor( ... images=[image], ... text=text, ... return_masks=True, ... return_codebook_pixels=True, ... padding=True, ... max_length=77, ... return_tensors="pt", ... ) >>> output = model(**inputs) ``` Nz`input_ids_masked` isn't passed which means MLM loss won't be calculated correctlySetting it to `input_ids` so that model can work. Please pass it if this is unintentional. This is usually OKAY if you are doing inference on unmasked text...T) rrrrrrrrr6r7) rrrrrrrr6r7z`return_loss` is set to True but the image codebook is not initialized and no `mim_labels` have been passed. Reinstantiate the model with `init_codebook` set to True or pass in your custom `mim_labels`z`codebook_pixel_value` are required to generate `mim_labels` if loss is expected. Call `AutoProcessor` with `return_codebook_pixels` set to Truerrr{r|r)r?r@rArBrCrDc3"K|] }||ndV dSrHr<)r+rLs r.r/z.FlavaForPreTraining.forward..Ks+__T%5TT1______r0c3K|]}||V dSr]r<)r+rs r.r/z.FlavaForPreTraining.forward..ls"<rGsumrFr r*r"r$r2rK)7r-rr>rr?rrrrrr;r@rArBrr6r7rCrw flava_outputflava_masked_outputpos_maskrr!rNrPrR total_lossmim_lossmlm_loss mmm_text_lossmmm_image_lossgc_lossitm_lossrTrUrZrYrVr$r%sequence_for_image masked_tokensmim_labels_filteredsequence_for_textmlm_labels_filtered pos_pairs end_indextext_embeddingimage_embedding gc_labels gc_loss_image gc_loss_text flava_lossesrs7 r.rzFlavaForPreTraining.forwards D&1%eGV^GKK KZK/4D:>> >% # .2N2Z!k!&.&; )0$Y"0EEF[\\  ?Q  #:#FKgKo!8 %!// ;; "&"4"4_"E"E7;7K ?--d334%7JOOA> 0!}::"D,ABBDWD\D\]_D`D`  H/H!]]+=>>  ?Q  #9#EJfJn 6 %!// ;; $5aaa*//!:L:L9L9N9NPQPQPQ6Q$R! * d.B C C &0&?#$5mQQQ6F$G!!]]+<== 0!}::"D,@AACVC[C[\^C_C_  H/H!]]+<==  ?Q  #?#K'CDDJ%&MM!,, $ '770!}:::zRRH/H/;3OPX3Y0)!+H!5J)!+H!5J&5h&?O ( 38MPQ8Q8Q!= /44Q77!;I!3AAAq1y=7H!!!4K!L %!// ;; "&"4"4_"E"E7;7K ?--d334 * d.B C C &0&?#%7 qqq8H%I"#'#6#67I#J#J <%']%@%@(--b$2GHHJ]JbJbceJfJf&&N#d&;;N#'#6#67I#J#J  ( 38Lq8P8P <  1!!!6L6Q6QRS6T6T5T5V5VXYXYXY2Y Z %!// ;; * d.B C C &0&?#$5mQQQ6F$G!"&"4"45F"G"G:$&M$?$?',,R1EFFH[H`H`acHdHd%%M"T%99M"&"4"45F"G"G  'O,GDLjmnLnLn!Z771aaa8PQQN]44^4LLN"j99:J111aQRQRQR7:STTO m55o25NNO} a &+223HJ_```;?;W;W1G<< 8 oy ##3H#= "1(";%h/  : " ; ;
  • Q>]>i#/88:::os,&8D$5>>@@@  +F. <#8#8#:#:   <> *F)E &9%J%J "z "z! ""z# $*:)9% &)8' (.-) *,O+  r0r])NNNNNNNNNNNNNNTNN)r5r6r7_tied_weights_keysrrr~rir9rr=rrr:rIr2rKrrrs@r.r'r's-;00< !!{!BI.2.2/304488<*.*.*.)-%)#'#'%p p #d*p  *T1p '$. p %047 p  t+ p  t+p ,p &-p $lT1p +/+p L4'p L4'p L4'p  $;p #!p "D[#p $D[%p ( u| 8 8)p p p ^p p p p p r0r')r'rrbrYrWrLr)Lr8rrr dataclassesrtypingrr9rrrT activationsr masking_utilsr modeling_layersr modeling_outputsr r modeling_utilsrprocessing_utilsrutilsrrrrrrconfiguration_flavarrrrr get_loggerr5rGrrQrRrrr>rKr~r`rorrrr rrr"r-rCrLrbrrWrYrrrrrrSr rr'__all__r<r0r.rys ######!!!!!! &&&&&&!!!!!!666666999999KKKKKKKK------&&&&&&jjjjjjjjjjjjjjjj  H % %>&)99