+ œ°³jq#ãó€^RIHt^RIHt^RIHt]!RR7]!RR]444t]!RR7]!RR ]444t]!R R7]!R R ]444t]!R R7]!RR]444t .ROt R#)é)Ústrict)ÚPreTrainedConfig)Úauto_docstringznvidia/parakeet-ctc-1.1b)Ú checkpointcó aa€]tRt^toRtRtR.tRt^t^t Rt Rt Rt Rt ^ t^tRt^Pt^t^tR tR tR tR tR tR tRtR tV3R ltV3RltRtVtV;t #)ÚParakeetEncoderConfiga convolution_bias (`bool`, *optional*, defaults to `True`): Whether to use bias in convolutions of the conformer's convolution module. conv_kernel_size (`int`, *optional*, defaults to 9): The kernel size of the convolution layers in the Conformer block. subsampling_factor (`int`, *optional*, defaults to 8): The factor by which the input sequence is subsampled. subsampling_conv_channels (`int`, *optional*, defaults to 256): The number of channels in the subsampling convolution layers. num_mel_bins (`int`, *optional*, defaults to 80): Number of mel features. subsampling_conv_kernel_size (`int`, *optional*, defaults to 3): 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. scale_input (`bool`, *optional*, defaults to `True`): Whether to scale the input embeddings. Example: ```python >>> from transformers import ParakeetEncoderModel, ParakeetEncoderConfig >>> # Initializing a `ParakeetEncoder` configuration >>> configuration = ParakeetEncoderConfig() >>> # Initializing a model from the configuration >>> model = ParakeetEncoderModel(configuration) >>> # Accessing the model configuration >>> configuration = model.config ``` Úparakeet_encoderÚpast_key_valueséiÚsiluTégš™™™™™¹?giˆg{®Gáz”?c óJ<€VPVn\SV` !R/VBR#©N©)Únum_attention_headsÚnum_key_value_headsÚsuperÚ __post_init__©ÚselfÚkwargsÚ __class__s&,€Út/home/luca/screen-checking/.venv/lib/python3.14/site-packages/transformers/models/parakeet/configuration_parakeet.pyrÚ#ParakeetEncoderConfig.__post_init__Ts!ø€Ø#'×#;Ñ#;ˆÔ Ü ‰ÒÑ' Ô'ócóf<€V^8„dQh/S[;R&S[;R&S[;R&S[;R&S[;R&S[;R&S[;R&S[;R&S[;R &S[;R &S[;R &S[;R &S[;R &S[S[,;R&S[S[,;R&S[S[,;R&S[S[,;R&S[S[,;R&S[;R&S[;R&S[;R&#)éÚ hidden_sizeÚnum_hidden_layersrÚintermediate_sizeÚ hidden_actÚattention_biasÚconvolution_biasÚconv_kernel_sizeÚsubsampling_factorÚsubsampling_conv_channelsÚ num_mel_binsÚsubsampling_conv_kernel_sizeÚsubsampling_conv_strideÚdropoutÚdropout_positionsÚ layerdropÚactivation_dropoutÚattention_dropoutÚmax_position_embeddingsÚ scale_inputÚinitializer_range)ÚintÚstrÚboolÚfloat)ÚformatÚ __classdict__s"€rÚ __annotate__Ú"ParakeetEncoderConfig.__annotate__sTø‡‚ñRÑñSñTÑñUñVÑ ñWñXÑ!ñYñZÑñ[ñ\Ññ]ñ^Ñ!ñ_ñ`ÑñañbÑñcñd #Ñ(ñeñfÑñgñh#&Ñ)ñiñj!Ñ$ñkñl‘S[Ññmñn™s•{Ñ(ñoñp‘s{Ñ ñqñr¡ Ñ)ñsñt™s•{Ñ(ñuñv!Ñ'ñwñxÑñyñzÑ#ò{r)r)!Ú__name__Ú __module__Ú __qualname__Ú__firstlineno__Ú__doc__Ú model_typeÚkeys_to_ignore_at_inferencerrrr r!r"r#r$r%r&r'r(r)r*r+r,r-r.r/r0r1rÚ__annotate_func__Ú__static_attributes__Ú__classdictcell__Ú __classcell__©rr7s@@rrrsŸù‡€ñ!ðF$€JØ#4Ð"5Ðà€KØÐØ ÐØ!ÐØ€JØ€NØ!ÐØÐØÐØ%(ÐØ€LØ()Ð Ø#$ÐØ€GØ%(ÐØ €IØ&)ÐØ%(ÐØ#'ÐØ€KØ#Ðõ(÷…rrcóbaa€]tRt^YtoRtRtR]/tRtRt Rt Rt Rt V3R lt V3R ltR tVtV;t#) ÚParakeetCTCConfigaÈ 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 [`ParakeetForCTC`]. 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 [`ParakeetForCTC`]. encoder_config (`Union[dict, ParakeetEncoderConfig]`, *optional*): The config object or dictionary of the encoder. Example: ```python >>> from transformers import ParakeetForCTC, ParakeetCTCConfig >>> # Initializing a Parakeet configuration >>> configuration = ParakeetCTCConfig() >>> # Initializing a model from the configuration >>> model = ParakeetForCTC(configuration) >>> # Accessing the model configuration >>> configuration = model.config ``` Ú parakeet_ctcÚencoder_configiÚmeanTNr c ó<€\VP\4'd\R/VPBVnMVPf\4VnVPPVn\ SV`!R/VBR#r©Ú isinstancerIÚdictrr1rrrs&,€rrÚParakeetCTCConfig.__post_init__}ógø€Ü d×)Ñ)¬4× 0Ò 0Ü"7Ñ"N¸$×:MÑ:MÑ"NˆDÕ Ø × Ñ Ò (Ü"7Ó"9ˆDÔ Ø!%×!4Ñ!4×!FÑ!FˆÔÜ ‰ÒÑ' Ô'rcó‚<€V^8„dQh/S[;R&S[;R&S[;R&S[S[,R,;R&S[R,;R&#)rÚ vocab_sizeÚctc_loss_reductionÚctc_zero_infinityNrIÚ pad_token_id)r2r3r4rNr)r6r7s"€rr8ÚParakeetCTCConfig.__annotate__Ys[ø‡‚ñ<Ññ=ñ>Ñ$ñ?ñ@Ñ"ñAñBÑ+Õ+¨dÕ2Ñ9ñCñD˜•*Ñ#òEr©rIr1)r:r;r<r=r>r?rÚ sub_configsrRrSrTrIrUrrArBrCrDrEs@@rrGrGYsDù‡€ñð0 €JØ#Ð%:Ð;€Kà€JØ$ÐØ"ÐØ59€NØ#€Lõ(÷I…rrGznvidia/parakeet-rnnt-0.6bcóraa€]tRt^†toRtRtR]/tRtRt ^t Rt ^ t Rt ^tRtR tV3R ltV3R ltR tVtV;t#) ÚParakeetRNNTConfiga¾ decoder_hidden_size (`int`, *optional*, defaults to 640): Hidden size of the LSTM prediction network and joint network. num_decoder_layers (`int`, *optional*, defaults to 2): Number of LSTM layers in the prediction network. max_symbols_per_step (`int`, *optional*, defaults to 10): Maximum number of symbols to emit per encoder time step during greedy decoding. encoder_config (`Union[dict, ParakeetEncoderConfig]`, *optional*): The config object or dictionary of the encoder. blank_token_id (`int`, *optional*, defaults to 8192): Blank token id. Different from `pad_token_id` for RNN-T. Example: ```python >>> from transformers import ParakeetForRNNT, ParakeetRNNTConfig >>> # Initializing a Parakeet RNN-T configuration >>> configuration = ParakeetRNNTConfig() >>> # Initializing a model from the configuration >>> model = ParakeetForRNNT(configuration) >>> # Accessing the model configuration >>> configuration = model.config ``` Ú parakeet_rnntrIi i€ÚreluNi Tc ó<€\VP\4'd\R/VPBVnMVPf\4VnVPPVn\ SV`!R/VBR#rrLrs&,€rrÚ ParakeetRNNTConfig.__post_init__±rPrcó¤<€V^8„dQh/S[;R&S[;R&S[;R&S[;R&S[;R&S[S[,R,;R&S[;R&S[;R &S[;R &#) rrRÚdecoder_hidden_sizeÚnum_decoder_layersr!Úmax_symbols_per_stepNrIrUÚblank_token_idÚis_encoder_decoder)r2r3rNrr4)r6r7s"€rr8ÚParakeetRNNTConfig.__annotate__†s—ø‡‚ñBÑñCñDÑ"ñEñFÑñGñHÑñIñJÑ"ñKñLÑ+Õ+¨dÕ2Ñ9ñMñNÑñOñPÑñQñRÑ#òSrrW)r:r;r<r=r>r?rrXrRr`rar!rbrIrUrcrdrrArBrCrDrEs@@rrZrZ†sZù‡€ñð6!€JØ#Ð%:Ð;€Kà€JØ"ÐØÐØ€JØ "ÐØ59€NØ€LØ€NØ#Ðõ(÷W…rrZznvidia/parakeet-tdt-0.6b-v3có8a€]tRt^ºtoRtRtRtV3RltRtVt R#)ÚParakeetTDTConfigaâ A TDT (Token-and-Duration Transducer) extends the base RNN-T configuration [`ParakeetRNNTConfig`] with a `durations` field: the joint network gains a duration head (its output width grows from `vocab_size` to `vocab_size + len(durations)`), and during greedy decoding the encoder frame pointer advances by the predicted duration rather than a fixed single frame. decoder_hidden_size (`int`, *optional*, defaults to 640): Hidden size of the LSTM prediction network and joint network. num_decoder_layers (`int`, *optional*, defaults to 2): Number of LSTM layers in the prediction network. max_symbols_per_step (`int`, *optional*, defaults to 10): Maximum number of symbols to emit per encoder time step during greedy decoding. durations (`list[int]`, *optional*, defaults to `[0, 1, 2, 3, 4]`): Token duration values that can be predicted. Each value represents how many frames a token or blank emission spans. encoder_config (`Union[dict, ParakeetEncoderConfig]`, *optional*): The config object or dictionary of the encoder. blank_token_id (`int`, *optional*, defaults to 8192): Blank token id. Different from `pad_token_id` for TDT. Example: ```python >>> from transformers import ParakeetForTDT, ParakeetTDTConfig >>> # Initializing a Parakeet TDT configuration >>> configuration = ParakeetTDTConfig() >>> # Initializing a model from the configuration >>> model = ParakeetForTDT(configuration) >>> # Accessing the model configuration >>> configuration = model.config ``` Ú parakeet_tdtcóZ<€V^8„dQh/S[S[,S[S[R3,,;R&#)r.Ú durations)Úlistr2Útuple)r6r7s"€rr8ÚParakeetTDTConfig.__annotate__ºs%ø‡‚ñN‘Cy™5¡ c ?Õ*Ñ<òOrrN)réréé) r:r;r<r=r>r?rjrArBrC)r7s@rrgrgºsø‡€ñ!ðF €JØ-<€I÷OƒrrgN)rGrrZrg) Úhuggingface_hub.dataclassesrÚconfiguration_utilsrÚutilsrrrGrZrgÚ__all__rrrÚrusÄðõ/å3Ý#ñÐ5Ô6Øô?(Ð,ó?(óó7ð?(ñDÐ5Ô6Øô((Ð(ó((óó7ð((ñVÐ6Ô7Øô/(Ð)ó/(óó8ð/(ñdÐ8Ô9Øô%=Ð*ó%=óó:ð%=òP d‚r