+ jr[^RIt^RIHt^RIt^RIHt^RIHtHtH t H t H t ^RI H t ^RIHt^RIHt^RIHt^R IHt^R IHtHtHt^R IHtHt^R IHt^R IHtH t H!t!^RI"H#t#^RI$H%t%^RI&H't'H(t(H)t)H*t*^RI+H,t,H-t-^RI.H/t/H0t0H1t1H2t2H3t3^RI4H5t5]!Pl!]74t8!RR]5]4t9!RR]RR7t:] !RR]44t;] !RR7]!RR]-444t<] !RR7]!R R!],444t=!R"R#]P|4t?!R$R%](4t@!R&R']'4tA!R(R)]04tB!R*R+]/4tC!R,R-]34tD!R.R/]14tE] !R0R17!R2R3]D44tF!R4R5]24tG.R6OtHR#)7N)Callable)strict) Tokenizerdecoders normalizerspre_tokenizers processors)Unigram)nn) AudioInput)create_bidirectional_mask)ALL_ATTENTION_FUNCTIONS)ProcessingKwargsProcessorMixinUnpack)PreTokenizedInput TextInput)TokenizersBackend)TransformersKwargsauto_docstringlogging)merge_with_config_defaults)capture_outputs)LlamaAttentionLlamaRotaryEmbeddingapply_rotary_pos_embeager_attention_forward)ParakeetCTCConfigParakeetEncoderConfig)ParakeetEncoderBlock ParakeetEncoderConvolutionModuleParakeetEncoderModelOutputParakeetForCTCParakeetPreTrainedModel) T5Tokenizerc@a]tRt^0toRRltRV3RlRlltRtVtR#) LasrTokenizerNc WPnVeVU u.uFp R\V 49gKV NK p p \V 4^8d(T\V4U u.uF p RV R2NK up , pMJV^8d#V\V 48wd\ RV RV R24hM \V4U u.uF p RV R2NK p p T pVeWpnMh\V4R3\V4R3\V4R3R.Vn\V^, RR4F$p VP P RV R2R34K& \\VP ^RR 74Vn Ve&\P!V4VPn \P!\P!4\P !RR R R 7.4VPn\$P !RR R R 7VPn\(P*!RR VRVRVRVRV/V B\,P.!RR..RORVP03.R7VPnR#uup iuup iuup i)Nz zBoth 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▁F)unk_id byte_fallbackalwaysT) replacementprepend_schemesplit eos_token unk_token pad_token extra_idsadditional_special_tokens$A)singlepairspecial_tokens)r+g)r7r8z$Br8) _extra_idsstrlenrange ValueError _vocab_scoresappendrr _tokenizerr Precompiled normalizerrSequenceWhitespaceSplit Metaspace pre_tokenizerrdecoderr__init__r TemplateProcessing eos_token_idpost_processor) selfr2r3r4_spm_precompiled_charsmapr5r6vocab vocab_filekwargsx extra_tokensis &&&&&&&&&, f/home/luca/screen-checking/.venv/lib/python3.14/site-packages/transformers/models/lasr/modular_lasr.pyrMLasrTokenizer.__init__1s|$ % 0'@['@!LTWXYTZDZAA'@L[< 1$)yIY-ZIYA 1#Q.?IY-ZZ)Q9L0A#A &yk1RSlRmn 8=Y7GH7G!j1-7GLH(4 %  !& Y%Y%Y% "D  9q="b1""))Zs!+-**+* &k\-ZIsII I Ic f<V^8dQhRS[S[S[,,RS[RS[R,RS[RS[/#) token_idsskip_special_tokensclean_up_tokenization_spacesN group_tokensreturn)intlistboolr?)format __classdict__s"rY __annotate__LasrTokenizer.__annotate__~sI  c? " '+Tk      c .\V\4'dV.pV'd,\P!V4Uu.uF qf^,NK ppVUu.uFqwVP8wgKVNK pp\ P !V3RVRVRV/VB#uupiuupi)rr]r^r_) isinstancerb itertoolsgroupby pad_token_idr_decode)rQr]r^r_r`rU token_grouptokens&&&&&, rYroLasrTokenizer._decode~s i % %" I ;D;L;LY;WX;WKQ;WIX)2P ud>O>O5OUU P ((   !4 *F     YQsB B(B)r>rErC)r8zzNdNNN)FNT)__name__ __module__ __qualname____firstlineno__rMro__static_attributes____classdictcell__rfs@rYr'r'0sK Z   rir'c B]tRt^tRRRRRRR/RRRR R R R /R RR//tRtR#)LasrProcessorKwargs audio_kwargs sampling_ratei>paddinglongestreturn_attention_maskT text_kwargs padding_siderightadd_special_tokensF common_kwargsreturn_tensorsptr=N)rtrurvrw _defaultsrxr=rirYr|r|sF U y #T  t G % *D1 Irir|F)totalclaa]tRt^to]tV3Rlt]RV3RlV3Rlll4t] R4t Rt Vt V;t #) LasrProcessorc&<\SV`W4R#N)superrM)rQfeature_extractor tokenizer __class__s&&&rYrMLasrProcessor.__init__s *6ric <V^8dQhRS[RS[S[,S[S[,,S[S[,,R,RS[R,RS[S[,/#)r\audiotextNr~rU)r rrrcrbrr|)rerfs"rYrgLasrProcessor.__annotate__s]!!!++d9o=EV@WWZ^^!Tz ! ,- !ric <VP!\3RVPP/VBpVf)\P RVR,R, R24M4W5R,R,8wd!\ RV RVR,R, R 24h\SV` !RR VR V/VBpVe'VPR 4VR &VPRR4V#)a sampling_rate (`int`, *optional*): The sampling rate of the input audio in Hz. This should match the sampling rate expected by the feature extractor (defaults to 16000 Hz). If provided, it will be validated against the processor's expected sampling rate, and an error will be raised if they don't match. If not provided, a warning will be issued and the default sampling rate will be assumed. tokenizer_init_kwargsNzUYou've provided audio without specifying the sampling rate. It will be assumed to be r}r~z$, 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.rr input_idslabelsattention_maskr=) _merge_kwargsr|r init_kwargslogger warning_oncerBr__call__pop)rQrrr~rU output_kwargs model_inputsrs&&&&, rYrLasrProcessor.__call__s0**  "&.."<"<     ghuwEiFGViWhXX|} N;OL L2=/AvxEFTxUVexfwggkl w'PeP$P-P  %1%5%5k%BL "   -t 4ricBVPPpVR.,#)r)rmodel_input_names)rQfeature_extractor_input_namess& rYrLasrProcessor.model_input_namess!(,(>(>(P(P%,z99rir=NN)rtrurvrwr|valid_processor_kwargsrMrrpropertyrrxry __classcell__rrfs@@rYrrsC07!!!F:::rirz google/medasr) checkpointca]tRt^toRtRt^tRtRtRt ^ t ^t ^t Rt RtR tR tRtRt]!4t]!4tV3R ltR tVtR#) 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). iFi'gư>g{Gz?Nc8<V^8dQh/S[;R&S[;R&S[;R&S[;R&S[;R&S[;R&S[;R&S[;R&S[;R &S[;R &S[S[,S[S[R 3,,;R &S[S[,S[S[R 3,,;R &S[;R&S[R,;R&#)r\ hidden_sizenum_hidden_layersintermediate_sizeattention_biasconvolution_biasconv_kernel_sizesubsampling_conv_kernel_size num_mel_binsmax_position_embeddingslayer_norm_eps.feed_forward_residual_weightsconv_residual_weightsbatch_norm_momentumNrope_parameters)rbrdfloatrctupledict)rerfs"rYrgLasrEncoderConfig.__annotate__sRSTUV!WX YZ"[\]^#&)_`ab!(cd ef$(;ucz1B#BOgh ;ucz)::Gij%klD['mrir=)g?g?)g@g?)rtrurvrw__doc__rrrrrrrrrrrrrrAttributeErrorsubsampling_factor scale_input__annotate_func__rxryrzs@rYrrsu$LK! N"() L#( NEO!=G!%#'O') "KsrircHa]tRtRtoRtRt^t]R4tV3Rlt Rt Vt R#) 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). rc<VPP^,#r\)encoder_configsubsampling_conv_stride)rQs&rYinputs_to_logits_ratio$LasrCTCConfig.inputs_to_logits_ratio1s""::A==ric2<V^8dQh/S[;R&S[;R&#)r\ vocab_sizern)rb)rerfs"rYrgLasrCTCConfig.__annotate__s4567rir=N) rtrurvrwrrrnrrrrxryrzs@rYrrs/.JL >>=rircPaa]tRtRtoV3RlV3RlltV3RlRltRtVtV;t#)LasrEncoderSubsamplingi6c <V^8dQhRS[/#r\configr)rerfs"rYrg#LasrEncoderSubsampling.__annotate__7s  0 ric4<\SV`4\P!VPVP 4Vn\P!VP VP VPVPR7Vn \P!VP VPVPVPR7Vn \P!VPVP 4Vn \P!4VnR#)) kernel_sizestrideN)rrMr Linearrrdense_0Conv1drrconv_0subsampling_conv_channelsconv_1dense_1ReLUact_fn)rQrrs&&rYrMLasrEncoderSubsampling.__init__7s yy!4!4f6H6HI ii      ;;11   ii     , ,;;11   yy!A!A6CUCUV ggi ricN<V^8dQhRS[PRS[P/#)r\input_featuresratorchTensor)rerfs"rYrgrIs#++ell+u||+ric,VPVPV44pVP^^4pVPVPV44pVPVP V44pVP^^4pVP V4#))rr transposerrr)rQr hidden_statess&& rYforwardLasrEncoderSubsampling.forwardIsz DLL$@A %//15  DKK $>?  DKK $>? %//15 ||M**ri)rrrrr rtrurvrwrMrrxryrrs@@rYrr6s  $+++rirc]tRtRtRtR#)LasrEncoderRotaryEmbeddingiRr=Nrtrurvrwrxr=rirYrrRsrircTaa]tRtRtoV3RlV3RlltRV3RlRlltRtVtV;t#)LasrEncoderAttentioniUc&<V^8dQhRS[RS[/#r\r layer_idxrrb)rerfs"rYrg!LasrEncoderAttention.__annotate__Vs0Sric4<\SV`W4RVnR#)FN)rrM is_causalrQrrrs&&&rYrMLasrEncoderAttention.__init__Vs +ric <V^8dQhRS[PRS[S[PS[P3,R,RS[PR,RS[S[,RS[S[PS[P3,/#)r\rposition_embeddingsNrrUra)rrrrr)rerfs"rYrgrZsy")")||")#5<<#=>E") t+ ") +, ") u||U\\) * ")ric VPRRp.VORNVPN5pVPV4PV4P ^^4pVP V4PV4P ^^4pVP V4PV4P ^^4p Vwr\WxW4wrx\P!VPP\4p T !VVVV V3RVP'gRM VPRVP/VBwrV P !.VORN5!P#4p VP%V 4p W3#)Ndropoutr*scalingr<)shapehead_dimq_projviewrk_projv_projrr get_interfacer_attn_implementationrtrainingattention_dropoutrreshape contiguouso_proj)rQrrrrU input_shape hidden_shape query_states key_states value_statescossinattention_interface attn_output attn_weightss&&&&, rYrLasrEncoderAttention.forwardZse$))#2. 88b8$--8 {{=166|DNNqRST [[/44\BLLQPQR {{=166|DNNqRST &#7 RU#[ (?(M(M KK , ,.E) %8      % $}}}C$2H2H % LL %  % ! "));;;;FFH kk+. ((ri)rrrrs@@rYrrUs")")")rircBaa]tRtRtoRV3RlV3RllltRtVtV;t#)LasrEncoderConvolutionModuleic <V^8dQhRS[/#rr)rerfs"rYrg)LasrEncoderConvolutionModule.__annotate__s\\0\ric<\SV`W4RVn\P!VP VP R7VnR#)same)momentumN)rrMrr BatchNorm1drrnorm)rQr module_configrs&&&rYrM%LasrEncoderConvolutionModule.__init__s5 / NN6#5#5@Z@Z[ ri)r)rr)rtrurvrwrMrxryrrs@@rYr"r"s\\\rir"cTaa]tRtRtoV3RlV3RlltRV3RlRlltRtVtV;t#)LasrEncoderBlockic&<V^8dQhRS[RS[/#rr)rerfs"rYrgLasrEncoderBlock.__annotate__s \ \0 \S \ric^<\SV`W4VPVnVPVn\P !VP VPRR7Vn\P !VP VPRR7Vn \P !VP VPRR7Vn \P !VP VPRR7Vn \P !VP VPRR7Vn R#)F)biasN) rrMrrr LayerNormrrnorm_feed_forward1 norm_self_att norm_convnorm_feed_forward2norm_outrs&&&rYrMLasrEncoderBlock.__init__s +-3-Q-Q*%+%A%A""$,,v/A/A6CXCX_d"e\\&*<*S>SZ_`f&8&8&:O:OV[\"$,,v/A/A6CXCX_d"e V%7%79N9NUZ[ ric <V^8dQhRS[PRS[PR,RS[PR,RS[S[,RS[P/#)r\rrNrrUra)rrrr)rerfs"rYrgr/s]!!||! t+!#\\D0 ! +, !  !ric TpVPVPV44pVP^,V,VP^,V,,pVPV4pVP!RRVRVRV/VBwrxW,pVP VP V4VR7p VP^,V,VP^,V ,,pTpVPVPV44pVP^,V,VP^,V,,pVPV4pV#)rrrr)rr=) feed_forward1r3rr4 self_attnconvr5r feed_forward2r6r7) rQrrrrUresidualnormalized_hidden_statesr_ conv_outputs &&&&, rYrLasrEncoderBlock.forwardsP!**4+B+B=+QR  . .q 1H >> 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) ``` )device)prN) target_length)r inputs_embedsrFTrr)last_hidden_stater)r_r`raranger ri unsqueezer functionalrrr]_get_output_attention_maskr rrbrandr^rcrSrb) rQrrrgrUrrr output_mask encoder_layerto_dropdropout_probabilitys &&&&, rYrLasrEncoder.forwardsD7 ?? 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  ric&<V^8dQh/S[;R&#rr)rerfs"rYrgrYs  ri)rr]r\r^rbrcr`r_r)rtrurvrwbase_model_prefixrMrrrrrrxryrrs@@rYrVrVsJ""H H  H ;rirVc2aa]tRtRtoV3RltRtVtV;t#) LasrForCTCi6c 8<\4P!R/VB#)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) ``` r=)rgenerate) super_kwargsrs,rYr}LasrForCTC.generate7s,w/,//rir=)rtrurvrwr}rxryrrs@@rYr{r{6s00rir{)r{rVrErrrr')Irlcollections.abcrrhuggingface_hub.dataclassesr tokenizersrrrrr tokenizers.modelsr r audio_utilsr masking_utilsr modeling_utilsrprocessing_utilsrrrtokenization_utils_baserrtokenization_utils_tokenizersrutilsrrr utils.genericrutils.output_capturingrllama.modeling_llamarrrrparakeet.configuration_parakeetrrparakeet.modeling_parakeetr r!r"r#r$t5.tokenization_t5r% get_loggerrtrr'r|rrrModulerrrr"r-rErSrVr{__all__r=rirYrs$ .SS%%65HHC>@@75vvV-   H %d K!2d N *%  -:N-:-:`?+7#-7#,7#t?+>%>,>@+RYY+8')T\#C\.+.b1& 7  ` %`  ` F004 ri