3j \SSKrSSKJr SSKrSSKJr SSKJrJrJ r J r J r SSK J r SSKJr SSKJrJr SS KJr SS KJrJr SS KJrJrJr SS KJrJr SS KJr SSK J!r!J"r"J#r#J$r$ SSK%J&r& SSK'J(r( SSK)J*r*J+r+J,r,J-r- SSK.J/r/J0r0 SSK1J2r2J3r3J4r4J5r5J6r6 SSK7J8r8 \$Rr"\:5r;"SS\8\5r<"SS\SS9r=\""SS\55r>\""SS9\"S S!\0555r?\""SS9\"S"S#\/555r@"S$S%\R5rB"S&S'\+5rC"S(S)\*5rD"S*S+\35rE"S,S-\25rF"S.S/\65rG"S0S1\45rH\""S2S39"S4S5\G55rI"S6S7\55rJ/S8QrKg)9N)Callable)strict) Tokenizerdecoders normalizerspre_tokenizers processors)Unigram)nn) AudioInputmake_list_of_audio)create_bidirectional_mask)ALL_ATTENTION_FUNCTIONSPreTrainedModel)ProcessingKwargsProcessorMixinUnpack)PreTokenizedInput TextInput)TokenizersBackend)TransformersKwargsauto_docstringcan_return_tuplelogging)merge_with_config_defaults)capture_outputs)LlamaAttentionLlamaRotaryEmbeddingapply_rotary_pos_embeager_attention_forward)ParakeetCTCConfigParakeetEncoderConfig)ParakeetEncoderBlock ParakeetEncoderConvolutionModuleParakeetEncoderModelOutputParakeetForCTCParakeetPreTrainedModel) T5Tokenizerc h\rSrSrS SjrS S\\\-S\S\S-S\S\4 S jjr S r g) LasrTokenizer0Nc nXPlUb~UV s/sHn S[U 5;dMU PM n n [U 5S:a$U[U5V s/sH n SU S3PM sn - nOIUS:a!U[U 5:wa[ SUSUS35eO![U5V s/sH n SU S3PM n n U nUbXplOd[U5S4[U5S4[U5S4S /Ul[US- S S 5H$n UR R SU S3S45 M& [[UR S S S 95Ul Ub%[R"U5URl [R"[R"5[R "SSSS9/5URl[$R "SSSS9URl[(R*"SUUUUUS.U D6 [,R."SS//SQSUR04/S9URlgs sn fs sn fs sn f)Nz rzBoth extra_ids (z!) and additional_special_tokens (zm) are provided to LasrTokenizer. In this case the additional_special_tokens must include the extra_ids tokens)▁gr F)unk_id byte_fallbackr2alwaysT) replacementprepend_schemesplit) eos_token unk_token pad_token extra_idsadditional_special_tokens$A)r?r@z$Br@)singlepairspecial_tokens) _extra_idsstrlenrange ValueError _vocab_scoresappendrr _tokenizerr Precompiled normalizerrSequenceWhitespaceSplit Metaspace pre_tokenizerrdecoderr__init__r TemplateProcessing eos_token_idpost_processor) selfr:r;r<_spm_precompiled_charsmapr=r>vocab vocab_filekwargsx extra_tokensis _/home/wildlama/miniconda3/lib/python3.13/site-packages/transformers/models/lasr/modular_lasr.pyrTLasrTokenizer.__init__1s_$ % 0'@['@!LTWXYTZDZA'@L[< 1$)yIY-ZIYA 1#Q.?IY-ZZ)Q9L0A#A &yk1RSlRmn 8=Y7GH7G!j1-7GLH(4 %  !& Y%Y%Y% "D  9q="b1""))Zs!+-**+* &k\-ZIsH(H( H-H2 token_idsskip_special_tokensclean_up_tokenization_spaces group_tokensreturnc [U[5(aU/nU(a([R"U5Vs/sHofSPM nnUVs/sHowUR:wdMUPM nn[ R "U4UUUS.UD6$s snfs snf)Nr)rbrcrd) isinstanceint itertoolsgroupby pad_token_idr_decode)rXrbrcrdrer\ token_grouptokens r`rmLasrTokenizer._decode~s i % %" I ;D;L;LY;WX;WKQ;WIX)2P ud>O>O5OU P ((   3)E     YQsB B #B )rErLrJ)r@zzNdNNN)FNT) __name__ __module__ __qualname____firstlineno__rTrilistboolrFrm__static_attributes__rDr`r,r,0st"&"&K `%*48!  c? " '+Tk       ryr,c4\rSrSrSSSS.SSSS.S S 0S .rS rg )LasrProcessorKwargsi>longestT) sampling_ratepaddingreturn_attention_maskrightF)r padding_sideadd_special_tokensreturn_tensorspt) audio_kwargs text_kwargs common_kwargsrDN)rrrsrtru _defaultsrxrDryr`r{r{s3# %) #"' +D1 Iryr{F)totalc ^\rSrSrU4Sjr\S S\S\\-\ \-\ \-S-S\ S-S\ \ 4Sjj5r \S 5rS rU=r$) LasrProcessorc$>[TU]X5 gN)superrT)rXfeature_extractor tokenizer __class__s r`rTLasrProcessor.__init__s *6ryNaudiotextr~r\c [U5nUR"[4SURR0UD6nUc [ R SUSSS35 O#X5SS:wa[SUSUSSS35eUbUR"U40USD6nUbUR"U40US D6nUcW$WS WS 'U$) a sampling_rate (`int`, *optional*): The sampling rate of the input audio in Hz. This should match the sampling rate expected by the feature extractor (defaults to 16000 Hz). If provided, it will be validated against the processor's expected sampling rate, and an error will be raised if they don't match. If not provided, a warning will be issued and the default sampling rate will be assumed. tokenizer_init_kwargszUYou've provided audio without specifying the sampling rate. It will be assumed to be rr~z$, which can result in silent errors.z The sampling rate of the audio (z5) does not match the sampling rate of the processor (zD). Please provide resampled the audio to the expected sampling rate.r input_idslabels) r _merge_kwargsr{r init_kwargslogger warning_oncerIr)rXrrr~r\ output_kwargsinputs encodingss r`__call__LasrProcessor.__call__sG#5)**  "&.."<"<     ghuwEiFGViWhXX|} N;OL L2=/AvxEFTxUVexfwggkl   ++ES]>5RSF  tL}]/KLI <M(5F8 Mryc:URRnUS/-$)Nr)rmodel_input_names)rXfeature_extractor_input_namess r`rLasrProcessor.model_input_namess!(,(>(>(P(P%,z99ryrDNN)rrrsrtrurTrr rrrvrirr{rpropertyrrx __classcell__rs@r`rrs7bf$( ((++d9o=EV@WWZ^^(Tz ( ,- ((T::ryrz google/medasr) checkpointc,\rSrSr%SrSr\\S'Sr\\S'Sr \\S'S r \ \S 'S r \ \S 'S r \\S 'Sr\\S'Sr\\S'Sr\\S'Sr\\S'Sr\\\\S4-\S'Sr\\\\S4-\S'Sr\\S'Sr\S-\S'\"5r\"5rSrg) LasrEncoderConfiga convolution_bias (`bool`, *optional*, defaults to `False`): Whether to use bias in convolutions of the conformer's convolution module. conv_kernel_size (`int`, *optional*, defaults to 32): The kernel size of the convolution layers in the Conformer block. subsampling_conv_channels (`int`, *optional*, defaults to 256): The number of channels in the subsampling convolution layers. subsampling_conv_kernel_size (`int`, *optional*, defaults to 5): The kernel size of the subsampling convolution layers. subsampling_conv_stride (`int`, *optional*, defaults to 2): The stride of the subsampling convolution layers. dropout_positions (`float`, *optional*, defaults to 0.0): The dropout ratio for the positions in the input sequence. feed_forward_residual_weights (`tuple[float, float]`, *optional*, defaults to `[1.5, 0.5]`): The residual weights for the feed forward layers. conv_residual_weights (`tuple[float, float]`, *optional*, defaults to `[2.0, 1.0]`): The residual weights for the convolution layers. batch_norm_momentum (`float`, *optional*, defaults to 0.01): The momentum for the batch normalization layers Example: ```python >>> from transformers import LasrEncoderModel, LasrEncoderConfig >>> # Initializing a `LasrEncoder` configuration >>> configuration = LasrEncoderConfig() >>> # Initializing a model from the configuration >>> model = LasrEncoderModel(configuration) >>> # Accessing the model configuration >>> configuration = model.config ``` This configuration class is based on the LasrEncoder architecture from Google Health AI. You can find more details and pre-trained models at [google/medasr](https://huggingface.co/google/medasr).  hidden_sizenum_hidden_layersiintermediate_sizeFattention_biasconvolution_bias conv_kernel_sizesubsampling_conv_kernel_size num_mel_binsi'max_position_embeddingsgư>layer_norm_eps)g?g?.feed_forward_residual_weights)g@g?conv_residual_weightsg{Gz?batch_norm_momentumNrope_parametersrD)rrrsrtru__doc__rri__annotations__rrrrwrrrrrrfloatrrvtuplerrrdictAttributeErrorsubsampling_factor scale_inputrxrDryr`rrs$LKs!s! ND "d"c() #)L##(S( NE EO!4;ucz1B#BO=G4;ucz)::G!%%#'OTD['') "KryrcF\rSrSr%SrSr\\S'Sr\\S'\ S5r Sr g ) LasrCTCConfigia ctc_loss_reduction (`str`, *optional*, defaults to `"mean"`): Specifies the reduction to apply to the output of `torch.nn.CTCLoss`. Only relevant when training an instance of [`LasrForCTC`]. ctc_zero_infinity (`bool`, *optional*, defaults to `True`): Whether to zero infinite losses and the associated gradients of `torch.nn.CTCLoss`. Infinite losses mainly occur when the inputs are too short to be aligned to the targets. Only relevant when training an instance of [`LasrForCTC`]. Example: ```python >>> from transformers import LasrForCTC, LasrCTCConfig >>> # Initializing a Lasr configuration >>> configuration = LasrCTCConfig() >>> # Initializing a model from the configuration >>> model = LasrForCTC(configuration) >>> # Accessing the model configuration >>> configuration = model.config ``` This configuration class is based on the Lasr CTC architecture from Google Health AI. You can find more details and pre-trained models at [google/medasr](https://huggingface.co/google/medasr). r vocab_sizerrlc4URRS-$)Nr)encoder_configsubsampling_conv_stride)rXs r`inputs_to_logits_ratio$LasrCTCConfig.inputs_to_logits_ratio6s""::A==ryrDN) rrrsrtrurrrirrlrrrxrDryr`rrs/.JL# >>ryrcj^\rSrSrS\4U4SjjrS\RS\R4SjrSr U=r $)LasrEncoderSubsamplingi;configc&>[TU]5 [R"URUR 5Ul[R"UR UR URURS9Ul [R"UR URURURS9Ul [R"URUR 5Ul [R"5Ulg)N) kernel_sizestride)rrTr Linearrrdense_0Conv1drrconv_0subsampling_conv_channelsconv_1dense_1ReLUact_fn)rXrrs r`rTLasrEncoderSubsampling.__init__<s yy!4!4f6H6HI ii      ;;11   ii     , ,;;11   yy!A!A6CUCUV ggi ryinput_featuresrfc,URURU55nURSS5nURURU55nURUR U55nURSS5nUR U5$)Nr/r)rr transposerrr)rXr hidden_statess r`forwardLasrEncoderSubsampling.forwardNsz DLL$@A %//15  DKK $>?  DKK $>? %//15 ||M**ry)rrrrr) rrrsrtrurrTtorchTensorrrxrrs@r`rr;s0 0 $+ell+u||++ryrc\rSrSrSrg)LasrEncoderRotaryEmbeddingiWrDNrrrsrtrurxrDryr`rrWsryrc ^\rSrSrS\S\4U4SjjrS S\RS\ \R\R4S-S\RS-S \ \ S \ \R\R44 S jjr S r U=r$)LasrEncoderAttentioniZr layer_idxc2>[TU]X5 SUlg)NF)rrT is_causalrXrrrs r`rTLasrEncoderAttention.__init__[s +ryNrposition_embeddingsattention_maskr\rfc URSSn/UQSPURP7nURU5RU5R SS5nUR U5RU5R SS5nUR U5RU5R SS5n Uup[XxX5upx[R"URR[5n U "UUUU U4UR(dSO URURS.UD6upU R "/UQSP76R#5n UR%U 5n X4$)Nr3r/rr1)dropoutscaling)shapehead_dimq_projviewrk_projv_projr!r get_interfacer_attn_implementationr"trainingattention_dropoutrreshape contiguouso_proj)rXrrrr\ input_shape hidden_shape query_states key_states value_statescossinattention_interface attn_output attn_weightss r`rLasrEncoderAttention.forward_s\$))#2. 88b8$--8 {{=166|DNNqRST [[/44\BLLQPQR {{=166|DNNqRST &#7 RU#[ (?(M(M KK , ,.E) %8      % $}}C$2H2HLL %  % ! "));;;;FFH kk+. ((ry)rr)rrrsrtrurrirTrrrrrrrxrrs@r`rrZs0SIM.2 ")||")#5<<#=>E") t+ ") +, ") u||U\\) * ")")ryrc4^\rSrSrSS\4U4SjjjrSrU=r$)LasrEncoderConvolutionModuleirc>[TU]X5 SUl[R"UR UR S9Ulg)Nsame)momentum)rrTrr BatchNorm1drrnorm)rXr module_configrs r`rT%LasrEncoderConvolutionModule.__init__s5 / NN6#5#5@Z@Z[ ry)rrr)rrrsrtrurrTrxrrs@r`rrs\0\\ryrc ^\rSrSrS\S\4U4SjjrS S\RS\RS-S\RS-S \ \ S \R4 S jjr S r U=r $)LasrEncoderBlockirrc>>[TU]X5 URUlURUl[R "UR URSS9Ul[R "UR URSS9Ul [R "UR URSS9Ul [R "UR URSS9Ul [R "UR URSS9Ul g)NF)bias) rrTrrr LayerNormrrnorm_feed_forward1 norm_self_att norm_convnorm_feed_forward2norm_outrs r`rTLasrEncoderBlock.__init__s +-3-Q-Q*%+%A%A""$,,v/A/A6CXCX_d"e\\&*<*S>SZ_`f&8&8&:O:OV[\"$,,v/A/A6CXCX_d"e V%7%79N9NUZ[ ryNrrrr\rfc &UnURURU55nURSU-URSU--nURU5nUR"SUUUS.UD6upxX-nUR UR U5US9n URSU-URSU --nUnURURU55nURSU-URSU--nURU5nU$)Nrr/)rrr)rrD) feed_forward1r rr! self_attnconvr"r feed_forward2r#r$) rXrrrr\residualnormalized_hidden_statesr_ conv_outputs r`rLasrEncoderBlock.forwardsD!**4+B+B=+QR  . .q 1H [TU]U5 SUlURUlURUlUR Ul[ U5Ul[U5Ul [R"[UR5Vs/sHn[X5PM sn5Ul[R "UR"UR$SS9UlUR)5 gs snf)NF)epsr)rrTgradient_checkpointingrdropout_positions layerdropr subsamplerr rotary_embr ModuleListrHrrlayersrrrout_norm post_initrs r`rTLasrEncoder.__init__s  &+#~~ !'!9!9))084V<mmBGH`H`Ba bBaY f 0Ba b   V%7%7V=R=RY^_   csC3Nrroutput_attention_maskr\rfc URU5nURU[R"URSUR S9R S55upg[RRXPRURS9n[RRX`RURS9n[RRXpRURS9nSnUbURX%RSS9nUn[URUUS9nURHSn Sn UR(a'[R "/5n XR":aS n U (aMFU "U4UXg4S .UD6nMU UR%U5n['UU(aUbUR)5S 9$SS 9$) a output_attention_mask (`bool`, *optional*): Whether to return the output attention mask. Example: ```python >>> from transformers import AutoProcessor, LasrEncoder >>> from datasets import load_dataset, Audio >>> model_id = "google/medasr" >>> processor = AutoProcessor.from_pretrained(model_id) >>> encoder = ParakeetEncoder.from_pretrained(model_id) >>> ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation") >>> ds = ds.cast_column("audio", Audio(sampling_rate=processor.feature_extractor.sampling_rate)) >>> inputs = processor(ds[0]["audio"]["array"]) >>> encoder_outputs = encoder(**inputs) >>> print(encoder_outputs.last_hidden_state.shape) ``` r/)devicer)prN) target_length)r inputs_embedsrFT)rr)last_hidden_stater)rGrHrarangerrP unsqueezer functionalrrrE_get_output_attention_maskrrrJrandrFrKr=ri) rXrrrNr\rr r  output_mask encoder_layerto_dropdropout_probabilitys r`rLasrEncoder.forwardsF7 ?? 5<< (;(;A(>}G[G[\ffghi  --m||VZVcVc-d mm##C+A+ADMM#Zmm##C+A+ADMM#Z  %99.XkXklmXn9oK(N2;;')  "[[MG}}&+jjn#&7"G7 -!!#1), ! ! )  m4 %+0E+Ja;??,  gk  ry)rrErDrFrJrKrHrGr)rrrsrtrurrbase_model_prefixrTrrrrrrrwrrr=rrxrrs@r`r@r@s !0"/3-1 H  H  t+H  $d{ H +, H H  H ryr@c(^\rSrSrU4SjrSrU=r$) LasrForCTCi<c 8>[5R"S0UD6$)a Example: ```python >>> from transformers import AutoProcessor, LasrForCTC >>> from datasets import load_dataset, Audio >>> model_id = "google/medasr" >>> processor = AutoProcessor.from_pretrained(model_id) >>> model = LasrForCTC.from_pretrained(model_id) >>> ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation") >>> ds = ds.cast_column("audio", Audio(sampling_rate=processor.feature_extractor.sampling_rate)) >>> inputs = processor(ds[0]["audio"]["array"], text=ds[0]["text"]) >>> predicted_ids = model.generate(**inputs) >>> transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True) >>> print(transcription) ``` rD)rgenerate) super_kwargsrs r`rcLasrForCTC.generate=s,w/,//ryrD)rrrsrtrurcrxrrs@r`rara<s 00ryra)rar@r1rrrr,)Lrjcollections.abcrrhuggingface_hub.dataclassesr tokenizersrrrrr tokenizers.modelsr r audio_utilsr r masking_utilsrmodeling_utilsrrprocessing_utilsrrrtokenization_utils_baserrtokenization_utils_tokenizersrutilsrrrr utils.genericrutils.output_capturingrllama.modeling_llamarr r!r"parakeet.configuration_parakeetr#r$parakeet.modeling_parakeetr%r&r'r(r)t5.tokenization_t5r* get_loggerrrrr,r{rrrModulerrrrrr1r=r@ra__all__rDryr`rzs$ .SS%96FHHC>RR75vvV-   H %d K!2d N *%  2:N2:2:j?+7#-7#,7#t?+>%>,>@+RYY+8')T\#C\.+.b1& 7  a %a  a H004 ry