o iP@s*ddlmZddlZGdddejZdS))AnyNcseZdZUdZejed<dZdZdZ        d!de e e fdBde e e fdBde dBde dBde de de de dede dede ffdd ZfddZedejfdd ZZS)"UltravoxConfiga This is the configuration class to store the configuration of a [`UltravoxForConditionalGeneration`]. It is used to instantiate an Ultravox model according to the specified arguments, defining the model architecture. Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the documentation from [`PretrainedConfig`] for more information. Args: audio_config (`Union[AutoConfig, dict]`, *optional*): Custom audio config or dict. text_config (`Union[AutoConfig, dict]`, *optional*): The config object of the text backbone. audio_model_id (`str`, *optional*): The model ID of the audio backbone. text_model_id (`str`, *optional*): The model ID of the text backbone. ignore_index (`int`, *optional*, defaults to -100): The ignore index for the loss function. audio_token_index (`int`, *optional*, defaults to 32000): The audio token index to encode the audio prompt. stack_factor (`int`, *optional*, defaults to 8): Audio downsampling factor for the multimodal projector. norm_init (`float`, *optional*, defaults to 0.4): The initialization value for the layer normalization. projector_act (`str`, *optional*, defaults to `"swiglu"`): The activation function used by the multimodal projector. projector_ln_mid (`bool`, *optional*, defaults to `False`): Whether to apply layer normalization at the middle of the projector or at the end. Versions v0.4.1 and below use `False`, but v0.5 and above use `True`. wrapped_model_configultravoxz <|audio|>FN}皙?swiglur audio_config text_configaudio_model_id text_model_id ignore_indexaudio_token_index hidden_size stack_factor norm_init projector_actprojector_ln_midnum_projector_layersc  s||_||_||_||_| |_| |_| |_| |_||_|dur2|p"i}t j | dddi||_ ||_ |durOd|_ |p?i}t j | dddi||_tjdi| dS)N model_typellamawhisper)rrrrrrrrr transformersCONFIG_MAPPINGgetrrr super__init__)selfr r rrrrrrrrrrkwargs __class__r^/home/ubuntu/vllm_env/lib/python3.10/site-packages/vllm/transformers_utils/configs/ultravox.pyr 3s4  zUltravoxConfig.__init__csd|dkr|durddlm}||dd|_n|dkr+|dur+ddlm}||dd|_t||S)Nrr) get_configF)trust_remote_coder)vllm.transformers_utils.configr&rr r __setattr__)r!keyvaluer&r#rr%r)`s  zUltravoxConfig.__setattr__returncCs |jS)N)rget_text_config)r!rrr%r ss zUltravoxConfig.text_config) NNNNrrrr r r Fr)__name__ __module__ __qualname____doc__rPretrainedConfig__annotations__r audio_tokenis_compositiondictstrrintfloatboolr r)propertyr __classcell__rrr#r%r s\  #      -r)typingrrr2rrrrr%s