o Zhq:@sdZddlZddlmZmZddlmZddlmZe e Z GdddeZ Gd d d eZ Gd d d eZGd ddeZdS)z#BARK model generation configurationN)DictOptional)GenerationConfig)loggingcsBeZdZdZ             dfd d ZZS)BarkSemanticGenerationConfigZsemantic'TF?@';?33333H@Nc sZtjd|| |||||||d || |_| |_||_| |_| |_||_||_||_ dS)a Class that holds a generation configuration for [`BarkSemanticModel`]. This configuration inherit from [`GenerationConfig`] and can be used to control the model generation. Read the documentation from [`GenerationConfig`] for more information. Args: eos_token_id (`int`, *optional*, defaults to 10_000): The id of the *end-of-sequence* token. renormalize_logits (`bool`, *optional*, defaults to `True`): Whether to renormalize the logits after applying all the logits processors (including the custom ones). It's highly recommended to set this flag to `True` as the search algorithms suppose the score logits are normalized but some logit processors break the normalization. max_new_tokens (`int`, *optional*, defaults to 768): The maximum numbers of tokens to generate, ignoring the number of tokens in the prompt. output_scores (`bool`, *optional*, defaults to `False`): Whether or not to return the prediction scores. See `scores` under returned tensors for more details. return_dict_in_generate (`bool`, *optional*, defaults to `False`): Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple. output_hidden_states (`bool`, *optional*, defaults to `False`): Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for more details. output_attentions (`bool`, *optional*, defaults to `False`): Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned tensors for more details. temperature (`float`, *optional*, defaults to 1.0): The value used to modulate the next token probabilities. do_sample (`bool`, *optional*, defaults to `False`): Whether or not to use sampling ; use greedy decoding otherwise. text_encoding_offset (`int`, *optional*, defaults to 10_048): Text encoding offset. text_pad_token (`int`, *optional*, defaults to 129_595): Text pad token. semantic_infer_token (`int`, *optional*, defaults to 129_599): Semantic infer token. semantic_vocab_size (`int`, *optional*, defaults to 10_000): Semantic vocab size. max_input_semantic_length (`int`, *optional*, defaults to 256): Max length of semantic input vector. semantic_rate_hz (`float`, *optional*, defaults to 49.9): Semantic rate in Hertz. min_eos_p (`float`, *optional*): Minimum threshold of the probability of the EOS token for it to be sampled. This is an early stopping strategy to mitigate potential unwanted generations at the end of a prompt. The original implementation suggests a default value of 0.2. ) temperature do_sample eos_token_idrenormalize_logitsmax_new_tokens output_scoresreturn_dict_in_generateoutput_hidden_statesoutput_attentionsN) super__init__text_encoding_offsettext_pad_tokenZsemantic_pad_tokensemantic_infer_tokensemantic_vocab_sizemax_input_semantic_lengthsemantic_rate_hz min_eos_p)selfrrrrrrrrrrrrrr r!r"kwargs __class__re/var/www/auris/lib/python3.10/site-packages/transformers/models/bark/generation_configuration_bark.pyrs*A   z%BarkSemanticGenerationConfig.__init__)rTr FFFFr Fr r r rrrN)__name__ __module__ __qualname__ model_typer __classcell__rrr%r'rs&rcsHeZdZdZ            dd ed effdd ZZS)BarkCoarseGenerationConfigZcoarse_acousticsTFr /K/rv<max_coarse_historysliding_window_lenc sPtjd|||||||d|||_| |_| |_| |_| |_| |_||_dS)as Class that holds a generation configuration for [`BarkCoarseModel`]. This configuration inherit from [`GenerationConfig`] and can be used to control the model generation. Read the documentation from [`GenerationConfig`] for more information. Args: renormalize_logits (`bool`, *optional*, defaults to `True`): Whether to renormalize the logits after applying all the logits processors (including the custom ones). It's highly recommended to set this flag to `True` as the search algorithms suppose the score logits are normalized but some logit processors break the normalization. output_scores (`bool`, *optional*, defaults to `False`): Whether or not to return the prediction scores. See `scores` under returned tensors for more details. return_dict_in_generate (`bool`, *optional*, defaults to `False`): Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple. output_hidden_states (`bool`, *optional*, defaults to `False`): Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for more details. output_attentions (`bool`, *optional*, defaults to `False`): Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned tensors for more details. temperature (`float`, *optional*, defaults to 1.0): The value used to modulate the next token probabilities. do_sample (`bool`, *optional*, defaults to `False`): Whether or not to use sampling ; use greedy decoding otherwise. coarse_semantic_pad_token (`int`, *optional*, defaults to 12_048): Coarse semantic pad token. coarse_rate_hz (`int`, *optional*, defaults to 75): Coarse rate in Hertz. n_coarse_codebooks (`int`, *optional*, defaults to 2): Number of coarse codebooks. coarse_infer_token (`int`, *optional*, defaults to 12_050): Coarse infer token. max_coarse_input_length (`int`, *optional*, defaults to 256): Max length of input coarse vector. max_coarse_history (`int`, *optional*, defaults to 630): Max length of the output of the coarse acoustics model used in the fine generation step. sliding_window_len (`int`, *optional*, defaults to 60): The coarse generation step uses a sliding window to generate raw audio. )rrrrrrrNr) rrcoarse_semantic_pad_tokencoarse_rate_hzn_coarse_codebookscoarse_infer_tokenmax_coarse_input_lengthr4r5)r#rrrrrrrr6r7r8r9r:r4r5r$r%rr'rys$9  z#BarkCoarseGenerationConfig.__init__)TFFFFr Fr.r/r0r1rr2r3)r(r)r*r+intrr,rrr%r'r-vs*r-cs2eZdZdZ    d fdd Zdd ZZS) BarkFineGenerationConfigZfine_acousticsr c s$tj|d||_||_||_dS)aClass that holds a generation configuration for [`BarkFineModel`]. [`BarkFineModel`] is an autoencoder model, so should not usually be used for generation. However, under the hood, it uses `temperature` when used by [`BarkModel`] This configuration inherit from [`GenerationConfig`] and can be used to control the model generation. Read the documentation from [`GenerationConfig`] for more information. Args: temperature (`float`, *optional*): The value used to modulate the next token probabilities. max_fine_history_length (`int`, *optional*, defaults to 512): Max length of the fine history vector. max_fine_input_length (`int`, *optional*, defaults to 1024): Max length of fine input vector. n_fine_codebooks (`int`, *optional*, defaults to 8): Number of codebooks used. )rN)rrmax_fine_history_lengthmax_fine_input_lengthn_fine_codebooks)r#rr@rArBr$r%rr'rs z!BarkFineGenerationConfig.__init__cKsdS)z Overrides GenerationConfig.validate because BarkFineGenerationConfig don't use any parameters outside temperature. Nr)r#r$rrr'validatesz!BarkFineGenerationConfig.validate)r r=r>r?)r(r)r*r+rrCr,rrr%r'r<s r<c@s`eZdZdZ     ddeedeedeefdd Zedede de fd d Z d d Z dS)BarkGenerationConfigZbarkN]r>semantic_configcoarse_acoustics_configfine_acoustics_configcKs|dur i}td|duri}td|dur!i}tdtdi||_tdi||_tdi||_||_||_ dS)a$Class that holds a generation configuration for [`BarkModel`]. The [`BarkModel`] does not have a `generate` method, but uses this class to generate speeches with a nested [`BarkGenerationConfig`] which uses [`BarkSemanticGenerationConfig`], [`BarkCoarseGenerationConfig`], [`BarkFineGenerationConfig`]. This configuration inherit from [`GenerationConfig`] and can be used to control the model generation. Read the documentation from [`GenerationConfig`] for more information. Args: semantic_config (`Dict`, *optional*): Semantic generation configuration. coarse_acoustics_config (`Dict`, *optional*): Coarse generation configuration. fine_acoustics_config (`Dict`, *optional*): Fine generation configuration. sample_rate (`int`, *optional*, defaults to 24_000): Sample rate. codebook_size (`int`, *optional*, defaults to 1024): Vector length for each codebook. NzMsemantic_config is None. initializing the semantic model with default values.zScoarse_acoustics_config is None. initializing the coarse model with default values.zOfine_acoustics_config is None. initializing the fine model with default values.r) loggerinforrFr-rGr<rH sample_rate codebook_size)r#rFrGrHrKrLr$rrr'rs    zBarkGenerationConfig.__init__cKs"|d|||d|S)z Instantiate a [`BarkGenerationConfig`] (or a derived class) from bark sub-models generation configuration. Returns: [`BarkGenerationConfig`]: An instance of a configuration object )rFrGrHNr)to_dict)clsrFrGrHr$rrr'from_sub_model_configs'sz+BarkGenerationConfig.from_sub_model_configscCsFt|j}|j|d<|j|d<|j|d<|jj|d<|S)z Serializes this instance to a Python dictionary. Override the default [`~PretrainedConfig.to_dict`]. Returns: `Dict[str, any]`: Dictionary of all the attributes that make up this configuration instance, rFrGrHr+) copydeepcopy__dict__rFrMrGrHr&r+)r#outputrrr'rM<s  zBarkGenerationConfig.to_dict)NNNrEr>) r(r)r*r+rrr classmethodrr-r<rOrMrrrr'rDs. 1 rD)__doc__rPtypingrrZgeneration.configuration_utilsrutilsrZ get_loggerr(rIrr-r<rDrrrr's   [P+