qiYPdZddlZddlZddlmZddlmZmZGddeZ dgZ y) zXcodec model configurationN)PreTrainedConfig)CONFIG_MAPPING AutoConfigcDeZdZdZdZeedZdgddgdddZiZd d d d d gd d gd d gd dd d d d f de e d zde de de e de e de e de de de d zde ffd Z e de fdZe de fdZe de fdZe de fdZe de fdZe de fdZxZS) XcodecConfiga This is the configuration class to store the configuration of an [`XcodecModel`]. It is used to instantiate a Xcodec model according to the specified arguments, defining the model architecture. Instantiating a configuration with the defaults will yield a similar configuration to that of the [Manel/X-Codec](https://huggingface.co/Manel/X-Codec) architecture. Configuration objects inherit from [`PreTrainedConfig`] and can be used to control the model outputs. Read the documentation from [`PreTrainedConfig`] for more information. Args: target_bandwidths (`List[float]`, *optional*, defaults to `[0.5, 1, 1.5, 2, 4]`): The range of different bandwidths (in kbps) the model can encode audio with. sample_rate (`int`, *optional*, defaults to 16000): The sampling rate at which the audio waveform should be digitalized, in hertz (Hz). kernel_size (`int`, *optional*, defaults to 3): Kernel size for the initial semantic convolution. channel_ratios (`List[float]`, *optional*, defaults to `[1, 1]`): Expansion factors for the number of output channels in each semantic block. strides (`List[int]`, *optional*, defaults to `[1, 1]`): Strides for each semantic encoder block. block_dilations (`List[int]`, *optional*, defaults to `[1, 1]`): Dilation factors for the residual units in semantic blocks. unit_kernel_size (`int`, *optional*, defaults to 3): Kernel size inside each ResidualUnit in semantic blocks. codebook_size (`int`, *optional*, defaults to 1024): Number of entries in each residual quantizer's codebook. codebook_dim (`int`, *optional*): Dimensionality of each codebook vector. Defaults to sum of hidden size of acoustic and semantic models. initializer_range (`float`, *optional*, defaults to 0.02): Standard deviation of the truncated normal initializer for all weight matrices. acoustic_model_config (`Union[Dict, DacConfig]`, *optional*): An instance of the configuration for the acoustic (DAC) model. semantic_model_config (`Union[Dict, HubertConfig, WavLMConfig]`, *optional*): An instance of the configuration object for the semantic (HuBERT) model. Example: ```python >>> from transformers import XcodecModel, XcodecConfig >>> # Initializing configuration >>> configuration = XcodecConfig() >>> # Initializing a model (with random weights) from the configuration >>> model = XcodecModel(configuration) >>> # Accessing the model configuration >>> configuration = model.config ```xcodec)acoustic_model_configsemantic_model_config@)ri)encoder_hidden_sizedownsampling_ratiosdecoder_hidden_sizeupsampling_ratios hidden_sizeNi>rg{Gz?target_bandwidths sample_rate kernel_sizechannel_ratiosstridesblock_dilationsunit_kernel_size codebook_size codebook_diminitializer_rangec  t| tr5| jdd| d<t| ddii|j| } n| tddi|j} | |_t| tr5| jdd| d<t| ddii|j | } n| tddi|j } | |_|gd}||_||_ ||_ ||_ ||_ ||_ ||_||_| |_| -|j j"|jj"z} | |_t'|Pdi| y)N model_typedachubert)g?rg?rr) isinstancedictgetr%_default_acoustic_model_config_kwargsr %_default_semantic_model_config_kwargsr rrrrrrrrr!rr super__init__)selfrrrrrrrrr r!r r kwargs __class__s a/opt/pipecat/venv/lib/python3.12/site-packages/transformers/models/xcodec/configuration_xcodec.pyr-zXcodecConfig.__init__^s +T 22G2K2KLZ_2` !, /$23H3V$W%YT??YCXY% !# *$25$9$gDrr:rCr9s r1num_quantizerszXcodecConfig.num_quantizerss14$00444K^K^9^_``r2)__name__ __module__ __qualname____doc__r#r sub_configsr*r+listfloatr>r-propertyr:r<r8rCrrG __classcell__)r0s@r1r r s0 dJ",!+K " ,#)-)-/)15 '(!fV&'V !!#'#'""2#;-2#2# 2# U 2# c 2#c2#2#2#Dj2#!2#h=C==6c66LCLL888_S__aaar2r ) rKr6numpyr?configuration_utilsrautorrr __all__r&r2r1rUs2! 3-Pa#Paf  r2