9inSSKJrJr SSKrSSKJr SSKJr SSKJ r SSK J r SSK J r "S S \ 5rg) )AnyDictN)Image) transforms) load_image)ModeKeys)OfaBasePreprocessorc^\rSrSrSr\R 4U4SjjrS\\ \ 4S\\ \ 44Sjr S\\ \ 4S\\ \ 44Sjr S\\ \ 4S\\ \ 44Sjr S rU=r$) OfaVisualEntailmentPreprocessor z/ OFA preprocessor for visual entailment tasks. c p>[[U] "XU/UQ70UD6 [R"S[R "UR UR 4[RRS9[R"5[R"URURS9/5Ul g)zpreprocess the data Args: cfg(modelscope.utils.config.ConfigDict) : model config model_dir (str): model path, mode: preprocessor mode (model mode) c$URS5$)NRGB)convert)images n/var/www/html/land-doc-ocr/venv/lib/python3.13/site-packages/modelscope/preprocessors/ofa/visual_entailment.py:OfaVisualEntailmentPreprocessor.__init__..#s %--.) interpolation)meanstdN)superr __init__rComposeResizepatch_image_sizeInterpolationModeBICUBICToTensor Normalizerrpatch_resize_transform)selfcfg model_dirmodeargskwargs __class__s rr(OfaVisualEntailmentPreprocessor.__init__s -   T D48 D(::BB D    !  diiTXX > : ' #rdatareturncUR[R:XaURU5$UR U5$)N)r'rTRAIN_build_train_sample_build_infer_sample)r$r,s r__call__(OfaVisualEntailmentPreprocessor.__call__+s4 99 &++D1 1++D1 1rcURU5nSRUS5nUSS0US'URUSSS9nURS:XaH[R "UR U/5n[R "USS UR/5nOURS :XaA[R "US U/5n[R "USS UR/5nOZURS :XaD[R "US S S U/5n[R "USS UR/5nO[eURRUS [U5*S- &XbS'XRS'URb[R"[U5[UR545R5n[U5[U5- S- n[![U5[U5- S- [U55HSn URR#5/XhU R%5-n URR'U 5n SXyU 'MU XrS'U$)a Building training samples. step 1. Preprocess the data using the logic of `_build_infer_sample` and make sure the label data in the result. step 2. Preprocess the label data to generate the `target` and `prev_output_tokens`. - tokenize the label data. - calculate the target item. 1) if `promp_type` is `None`, using tokenized label data. 2) if `promp_type` is `src`, concatenating the `source` data and tokenized label data. 3) if `promp_type` is `prev_output`, concatenating the `source` data without eos token and tokenized label data step 3. Add constraint mask Args: data (`Dict[str, Any]`): Input data, should contains the key of `text` `text2` and `label` are optional. Return: A dict object, contains source text input, patch images, patch masks with `Tensor([True])` value, decoder prompt, label, target, previous output tokens and constraint mask. z {}labelg?ref_dictF)add_bosadd_eosnoner Nsrcsource prev_outputtargetprev_output_tokensTconstraint_mask)r1format tokenize_text prompt_typetorchcatbos_itemeos_itemNotImplementedError tokenizer pad_token_idlenconstraint_triezerostgt_dictboolrangebostolistget_next_layer) r$r,sampler>tgt_itemprev_output_item target_itemr@ start_idxiconstraint_prefix_tokenconstraint_nodess rr03OfaVisualEntailmentPreprocessor._build_train_sample1sK2))$/fWo.$Wos3z%%feU%K   v %$yy$--)BC ))%5ab%94==$IJK    &$yy&*:H)EF ))%5ab%94==$IJK    .$yy&*:3B*?)JK ))%5ab%94==$IJK% %+/>>+F+F 'c(m^a'(&x'7#$    +#kk[!3t}}#5688< K(3x=81!?@11%8 $ ))$v/F*G*.*=*=?JXX^^''(JLF==,DT> M%? }#M M>&&tOOG,D'E'+':':?F7O r)r#)__name__ __module__ __qualname____firstlineno____doc__r INFERENCErrstrrr2r0r1__static_attributes__ __classcell__)r*s@rr r s(( 22T#s(^2S#X2 <S#X<4S><|?S#X?4S>??rr )typingrrrDPILr torchvisionrmodelscope.preprocessors.imagermodelscope.utils.constantrbaser r rrrs) "5.%a&9ar