`ik:dZddlZddlmZddlmZddlmZeje Z GddeZ Gd d eZ Gd d eZ Gd deZdS)z#BARK model generation configurationN)Optional)GenerationConfig)loggingcDeZdZdZ dfd ZxZS)BarkSemanticGenerationConfigsemantic'TF?@';?33333H@Nc tjd|| |||||||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_tokensemantic_pad_tokensemantic_infer_tokensemantic_vocab_sizemax_input_semantic_lengthsemantic_rate_hz min_eos_p)selfrrrrrrrrrrr r"r#r$r%r&kwargs __class__s /home/jaya/work/projects/VOICE-AGENT/VIET/agent-env/lib/python3.11/site-packages/transformers/models/bark/generation_configuration_bark.pyrz%BarkSemanticGenerationConfig.__init__sB  #%1)'$;!5/  %9!,".$8!#6 )B& 0")r Tr FFFFr Fr rrr rrN)__name__ __module__ __qualname__ model_typer __classcell__r)s@r*rrs~J %"#$""%#U#U#U#U#U#U#U#U#U#U#r+rcJeZdZdZ dd ed effd ZxZS)BarkCoarseGenerationConfigcoarse_acousticsTFr /K/rv<max_coarse_historysliding_window_lenc tjd|||||||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_lengthr;r<)r'rrrrrrrr>r?r@rArBr;r<r(r)s r*rz#BarkCoarseGenerationConfig.__init__ysr  #1'$;!5/  *C&,"4"4'>$"4"4r+)TFFFFr Fr5r6r7r8rr9r:)r,r-r.r/intrr0r1s@r*r3r3vs#J  %""(! #"%"$J5J5 J5 J5J5J5J5J5J5J5J5J5J5r+r3c2eZdZdZ dfd ZdZxZS) BarkFineGenerationConfigfine_acousticsr c vt|||_||_||_dS)aClass that holds a generation configuration for [`BarkFineModel`]. [`BarkFineModel`] is an autoencoder model, so should not usually be used for generation. However, under the hood, it uses `temperature` when used by [`BarkModel`] This configuration inherit from [`GenerationConfig`] and can be used to control the model generation. Read the documentation from [`GenerationConfig`] for more information. Args: temperature (`float`, *optional*): The value used to modulate the next token probabilities. max_fine_history_length (`int`, *optional*, defaults to 512): Max length of the fine history vector. max_fine_input_length (`int`, *optional*, defaults to 1024): Max length of fine input vector. n_fine_codebooks (`int`, *optional*, defaults to 8): Number of codebooks used. )rN)rrmax_fine_history_lengthmax_fine_input_lengthn_fine_codebooks)r'rrKrLrMr(r)s r*rz!BarkFineGenerationConfig.__init__s>4 [111'>$%:" 0r+c dS)z Overrides GenerationConfig.validate because BarkFineGenerationConfig don't use any parameters outside temperature. Nr)r'r(s r*validatez!BarkFineGenerationConfig.validates r+)r rGrHrI)r,r-r.r/rrOr0r1s@r*rErEs`!J #" 111111@       r+rEceZdZdZ d deedeedeefdZedede de fd Z d Z dS) BarkGenerationConfigbarkN]rHsemantic_configcoarse_acoustics_configfine_acoustics_configc <|i}td|i}td|i}tdtdi||_t di||_t di||_||_||_ dS)a$Class that holds a generation configuration for [`BarkModel`]. The [`BarkModel`] does not have a `generate` method, but uses this class to generate speeches with a nested [`BarkGenerationConfig`] which uses [`BarkSemanticGenerationConfig`], [`BarkCoarseGenerationConfig`], [`BarkFineGenerationConfig`]. This configuration inherit from [`GenerationConfig`] and can be used to control the model generation. Read the documentation from [`GenerationConfig`] for more information. Args: semantic_config (`Dict`, *optional*): Semantic generation configuration. coarse_acoustics_config (`Dict`, *optional*): Coarse generation configuration. fine_acoustics_config (`Dict`, *optional*): Fine generation configuration. sample_rate (`int`, *optional*, defaults to 24_000): Sample rate. codebook_size (`int`, *optional*, defaults to 1024): Vector length for each codebook. NzMsemantic_config is None. initializing the semantic model with default values.zScoarse_acoustics_config is None. initializing the coarse model with default values.zOfine_acoustics_config is None. initializing the fine model with default values.r) loggerinforrTr3rUrErV sample_rate codebook_size)r'rTrUrVrZr[r(s r*rzBarkGenerationConfig.__init__s<  " O KKg h h h " *&( # KKm n n n ($& ! KKi j j j;NNoNN'A'\'\D['\'\$%=%V%V@U%V%V"&*r+c |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 )rTrUrVr)to_dict)clsrTrUrVr(s r*from_sub_model_configsz+BarkGenerationConfig.from_sub_model_configs'sYs +3355$;$C$C$E$E"7"?"?"A"A      r+ctj|j}|j|d<|j|d<|j|d<|jj|d<|S)z Serializes this instance to a Python dictionary. Override the default [`~PretrainedConfig.to_dict`]. Returns: `dict[str, any]`: Dictionary of all the attributes that make up this configuration instance, rTrUrVr/) copydeepcopy__dict__rTr]rUrVr)r/)r'outputs r*r]zBarkGenerationConfig.to_dict<swt}--$($8$@$@$B$B !,0,H,P,P,R,R()*.*D*L*L*N*N&'#~8| r+)NNNrSrH) r,r-r.r/rdictr classmethodrr3rEr_r]rr+r*rQrQsJ +/2604 /+/+!$/+"*$/+ (~ /+/+/+/+b 5 "<  8    [ (r+rQ)__doc__ratypingrgeneration.configuration_utilsrutilsr get_loggerr,rXrr3rErQrr+r*rls3*) >>>>>>  H % %X#X#X#X#X##3X#X#X#vM5M5M5M5M5!1M5M5M5`( ( ( ( ( /( ( ( VYYYYY+YYYYYr+