+ QV-j8+^RIHtHt^RIt^RIHtHtHt^RI H t H t H t H t ]!4'd^RIt]P!]4t]!] !RR7R4!RR ] 44tR#) )AnyoverloadN)add_end_docstringsis_torch_availablelogging) GenericTensorPipelinePipelineExceptionbuild_pipeline_init_argsT) has_tokenizeraO top_k (`int`, *optional*, defaults to 5): The number of predictions to return. targets (`str` or `list[str]`, *optional*): When passed, the model will limit the scores to the passed targets instead of looking up in the whole vocab. If the provided targets are not in the model vocab, they will be tokenized and the first resulting token will be used (with a warning, and that might be slower). tokenizer_kwargs (`dict`, *optional*): Additional dictionary of keyword arguments passed along to the tokenizer.caa]tRt^toRtRtRtRtV3RlRltV3RlRlt V3RlRlt RV3R lR llt R t RR lt R tRRlt]V3RlRl4t]V3RlRl4tV3RlV3RlltRtVtV;t#)FillMaskPipelineFTc:<V^8dQhRS[RS[P/# input_idsreturnrnpndarray)format __classdict__s"q/Volumes/fast/ai/experiments/ui-tars-smoke/.venv/lib/python3.14/site-packages/transformers/pipelines/fill_mask.py __annotate__FillMaskPipeline.__annotate__Vs-BJJcb\P!WPP8HRR7pV#)Fas_tuple)torchnonzero tokenizer mask_token_id)selfr masked_indexs&& rget_masked_index!FillMaskPipeline.get_masked_indexVs&}}Y..2N2N%NY^_ rc:<V^8dQhRS[RS[P/#rr)rrs"rrrZs "**rcVPV4p\P!VP4pV^8d:\ RVP P RVPP R24hR#) fill-maskzNo mask_token (z) found on the inputN) r&rprodshaper modelbase_model_prefixr" mask_token)r$rr%numels&& r_ensure_exactly_one_mask_token/FillMaskPipeline._ensure_exactly_one_mask_tokenZsg,,Y7  **+ 19# ,,!$..";";!<7rc6<V^8dQhRS[S[S[3,/#)rr)dictstrr)rrs"rrrls   c= ! rc jVfRpVf/pVP!V3RV/VBpVPV4V#)Nptreturn_tensors)r"r:)r$inputsrAtokenizer_kwargspreprocess_parametersr5s&&&&, r preprocessFillMaskPipeline.preprocesslsH  !!N  #! ~~f`^`O_`  **<8rcBVP!R/VBpVR,VR&V#)r)r.)r$r5 model_outputss&& r_forwardFillMaskPipeline._forwardxs( 2\2 %1+%> k"rc @Ve,VP^,V8dVP^,pVR,^,pVR,p\P!W@PP8HRR7P R 4pV^VR3,pVP R R7pVe VRV3,pVPV4wr.p V P^,^8Hp \\V P4V P444Fwp wr.p\W4FwppVP4P4pVeVV,P4pVVWm,&V\P!VVPP8g4,pVPP!VV R7pR VR VR VPP!V.4R V/pVP#V4K V P#V4K V 'd V ^,#V #)NrlogitsFr:NNN)dim.)skip_special_tokensscoretoken token_strsequence)r-r r!r"r#squeezesoftmaxtopk enumerateziptolistnumpycopyrwhere pad_token_iddecodeappend)r$rItop_k target_idsroutputsr%rMprobsvalues predictionsresult single_maski_values _predictionsrowvptokensrS propositions&&&& r postprocessFillMaskPipeline.postprocess}s  !j&6&6q&9E&A$$Q'E!+.q1 )}}Y..2N2N%NY^_gghjk L!+,2&  !#z/*E#jj/ll1o* *3C I[I[I]4^*_ &A&CG21"*//1)"1 ,,.A*+|'4>>3N3N)N OP >>00[0Y&7A{DNNDYDY[\Z]D^`jltu  ;'3 MM# #+`$ !9  rc \V\4'dV.pVPP4p.pVFpVP V4pVfVPVRRR^RR7R,p\ V4^8Xd\PRV R24KaV^,p\PRV RVPPV4 R24VPV4K \\V44p\ V4^8Xd \R 4h\P!V4pV# \d/pELi;i) NFT)add_special_tokensreturn_attention_maskreturn_token_type_ids max_length truncationrzThe specified target token `zd` does not exist in the model vocabulary. We cannot replace it with anything meaningful, ignoring itz:` does not exist in the model vocabulary. Replacing with `z`.z1At least one target must be provided when passed.)r7r>r" get_vocab Exceptiongetlenloggerwarningconvert_ids_to_tokensr`r8set ValueErrorrarray)r$targetsvocabrbtargetid_rs&& rget_target_idsFillMaskPipeline.get_target_idssV gs # #iG NN,,.E F))F#C{ NN',*/*/ # + y>Q&NN6vh?UUl 26(;''+~~'K'KC'P&QQSU   c "56#j/* z?a PQ QXXj) E E sD44 EEc/pVeW4R&/pVeVPV4pWeR&VeWR&VPPf"\RVPP R4hV/V3#)NrCrbrar+z-The tokenizer does not define a `mask_token`.)rr"r#r r.r/)r$rarrCpreprocess_paramspostprocess_paramsrbs&&&& r_sanitize_parameters%FillMaskPipeline._sanitize_parameterss  '4D0 1  ,,W5J/9| ,  */w ' >> ' ' /#TZZ99;j !"&888rc R<V^8dQhRS[RS[RS[S[S[S[3,,/#rrBkwargsr)r>rr8r=)rrs"rrrs%OOsOcOd4S>6JOrc R#NrHr$rBrs&&,r__call__FillMaskPipeline.__call__sLOrc r<V^8dQhRS[S[,RS[RS[S[S[S[S[3,,,/#r)r8r>rr=)rrs"rrrs/[[tCy[C[Dd3PS8nAUr8rr=)rrs"rrrsVsT#Y#$tCQTH~BVY]^bcghkmphpcq^rYsBsrc <\SV`!V3/VBp\V\4'd\ V4^8Xd V^,#V#)a  Fill the masked token in the text(s) given as inputs. Args: inputs (`str` or `list[str]`): One or several texts (or one list of prompts) with masked tokens. targets (`str` or `list[str]`, *optional*): When passed, the model will limit the scores to the passed targets instead of looking up in the whole vocab. If the provided targets are not in the model vocab, they will be tokenized and the first resulting token will be used (with a warning, and that might be slower). top_k (`int`, *optional*): When passed, overrides the number of predictions to return. Return: A list or a list of list of `dict`: Each result comes as list of dictionaries with the following keys: - **sequence** (`str`) -- The corresponding input with the mask token prediction. - **score** (`float`) -- The corresponding probability. - **token** (`int`) -- The predicted token id (to replace the masked one). - **token_str** (`str`) -- The predicted token (to replace the masked one). )superrr7r8r|)r$rBrrc __class__s&&, rrrs@,'"64V4 fd # #F q(81: rrH)NN)N)NNN)__name__ __module__ __qualname____firstlineno___load_processor_load_image_processor_load_feature_extractor_load_tokenizerr&r2r:rErJrqrrrr__static_attributes____classdictcell__ __classcell__)rrs@@rrrsO!#O2h??   'R'R9*OO [[rr)typingrrr[rutilsrrrbaserr r r r get_loggerrr}rrHrrrsh CCVV   H %40Y gxg gr