tj50ddlZddlmZddlmZddlZddlmZm Z m Z e rddl Z e rddl Z ddlmZmZddlmZddlmZmZmZmZdd lmZmZGd d ed ZGdded ZeGddeZdgZdS)N)Path)Any)auto_docstringis_soundfile_availableis_torch_available) AudioInputmake_list_of_audio) BatchFeature) AudioKwargsProcessingKwargsProcessorMixinUnpack)PreTokenizedInput TextInputc4eZdZUdZeeefdzed<dS)CsmAudioKwargsa encoded_length_kwargs (`dict[str, Any]`, *optional*): Dictionary of keyword arguments used to compute the encoded audio sequence length. This includes parameters such as `kernel_sizes`, `strides`, `dilations`, and `use_causal_conv` that define the convolutional layers used in audio encoding. The encoded length is used to determine how many audio tokens to generate for each audio input in the text sequence. Nencoded_length_kwargs)__name__ __module__ __qualname____doc__dictstrr__annotations__d/var/www/html/CA-Chatbot/venv/lib/python3.11/site-packages/transformers/models/csm/processing_csm.pyrr$s7 S>D000000rrF)totalcNeZdZUeed<ddddgdgdgddd d d d d idZdS)CsmProcessorKwargs audio_kwargsTleftF)padding padding_sideadd_special_tokens)rrr( rr( rr(r)r(r(r(r-r(r(r(r(r(r(r)r()r(r(r(r(r(r(r(r(r(r(r(r(r(r(r() kernel_sizesstrides dilationsuse_causal_convi])r sampling_ratereturn_tensorspt) text_kwargsr" common_kwargsN)rrrrr _defaultsrrrr!r!0s    ""'  !Q P PHHHJJJ#' && #  +D1IIIrr!c eZdZ dfd ZeddZdedeeze eezzde e fdZ e dd eeze eze ezdzdedzd edzd edzde e f d ZedZxZS) CsmProcessorNct|ds'd|_||j|_n|j|_|j|_t|ds'd|_||j|_n|j|_|j|_t |||dS)N audio_tokenz <|AUDIO|>audio_eos_tokenz <|audio_eos|>) chat_template)hasattrr>convert_tokens_to_idsaudio_token_idr?audio_eos_token_idsuper__init__)selffeature_extractor tokenizerr@ __class__s rrFzCsmProcessor.__init__Gs y-00 ;*D "+"A"A$BR"S"SD  (4D "+":D y"344 C#2D &/&E&EdFZ&[&[D # ##,#, - - = #l2]BJ$x;?'CCaGFRUVVJJraudio saving_pathkwargsc tstdt|}t|tt fr|g}nDt|t tfrtd|Dstdt|t|krtd|j tfi|}|d}|d}t||D]m\}}t|tjr8|}t'j|||ndS)Nz/Please install `soundfile` to save audio files.c3NK|] }t|ttfV!dSN) isinstancerr).0ps r z*CsmProcessor.save_audio..s3@q@q`aAPSUY{A[A[@q@q@q@q@q@qrzAInvalid input path. Please provide a string, or a list of stringsz5The number of audio and saving paths must be the samer"r5)r ImportErrorr rbrrlisttupleall ValueErrorlen _merge_kwargsr!rLtorchTensorcpufloatnumpysfwrite) rGr\r]r^ output_kwargsr"r5 audio_valuerds r save_audiozCsmProcessor.save_audiosu&'' QOPP P#5)) kC; / / b&-KK[4-88 bS@q@qep@q@q@q=q=q b`aa a u::[)) ) )TUU U**     %^4 $_5 !%55 4 4NK+u|44 @)oo//5577==?? HQ ] 3 3 3 3 4 4rF?text output_labelsdepth_decoder_labels_ratioc ~ jtfdjji|}|d}|d}|dd} | dkrt jjdt|tr|g}nDt|ttfrtd|Dst d fd |D} d } |t|}t|} t| d kr:| t| kr'|t d t d | d| d||difd|D} | } g}|D]}g}j|vrY| d }j|z}|||jdd}j|vYd|vr.|d|d d}d|v.|||}j|fi|}i}||||ddgg}}d }| D]} | d krP|t-jd |t1jdgX|t-jd|||| zDd|t1jd|||| zDd|| z }j|fi|}|dd||t;d|Dfd|D}t1j|d |d<|r|djk }|j!d }|dkr8t1j"|dtG|d|z z}||}n|}t1j$|djk|dj%kz|dd }d!||ddd f|dddff<||d"<tM|| #S)$a output_labels (bool, *optional*, default=False): Whether to return labels for training. Indices will be in `[config.audio_token_id, -100, -101]`. - `config.audio_token_id` indicates an audio frame (considering sequence length elements as frames) - `-100` will be ignored in the loss computation - `-101` indicates the audio frame will be used only for the backbone model (using the first codebook token as labels) depth_decoder_labels_ratio (float, *optional*, default=1.0): The ratio of audio frames to keep for the depth decoder labels. Returns: [`BatchFeature`]: A [`BatchFeature`] with the following fields: - **input_ids** -- List of token ids to be fed to a model. Returned when `text` is not `None`. - **input_values** -- List of audio values to be fed to a model. Returned when `audio` is not `None`. - **attention_mask** -- List of indices specifying which tokens should be attended to by the model (when `return_attention_mask=True` or if *"attention_mask"* is in `self.model_input_names` and if `text` is not `None`). - **labels** -- List of labels for the audio frames. Returned when `output_labels=True`. tokenizer_init_kwargsr8r"r6Nr7z% only supports `return_tensors='pt'`.c3@K|]}t|tVdSra)rbr)rcts rrez(CsmProcessor.__call__..s-9[9[QR*Q:L:L9[9[9[9[9[9[rzAInvalid input text. Please provide a string, or a list of stringscDg|]}|jSr)countr>)rcr~rGs r z)CsmProcessor.__call__..s(CCC1774#344CCCrrz@No audio were provided, but there are audio tokens in the promptz)The number of audio tokens in each text (z7) should be the same as the number of provided audios (z).rcBg|]}j|jdfiS)r[shape)rc audio_arrayrrGs rrz)CsmProcessor.__call__..sE%%%]h(():2)>XXBWXX%%%rz r(return_attention_maskrcg|]D}t|tjr&|n|ESr)rbrmrnrorqrcels rrz)CsmProcessor.__call__.. sQ$&5?r5<4P4P X 0 0 0VXr)axisc(g|]}|jdSrrrs rrz)CsmProcessor.__call__..s%^%^%^rbhrl%^%^%^r)dim padding_maskc30K|]}|jdVdS)rNr)rccut_idxss rrez(CsmProcessor.__call__..s)RR(.,RRRRRRrc ~g|]9}tjj|d|jdz fd:S)rr)value)rmnn functionalpadr)rcrmax_lens rrz)CsmProcessor.__call__..sU$$$#''1gr@R6R2S[]'^^$$$rinput_values_cutoffs input_idsrwiilabels)data tensor_type)'rlr!rI init_kwargsgetrjrJrrbrrgrhrir rksumpopcopyr>appendreplaceupdatenpzerosrmtensor concatenatecumsumrHmaxstackrCnonzerorrandpermintwhererDr ) rGrxr\ryrzr^rtr8r"r6n_audio_in_textn_audionum_audio_tokens_listnum_audio_tokens_list_copy expanded_textsample replace_strnum_audio_tokensexpanded_audio_tokenencodingrconcatenated_audioroffset audio_inputsaudio_frame_idxsn_audio_frames rand_idxsskip_frames_idxsrrrs ` @@r__call__zCsmProcessor.__call__s:+*   "&."<   $M2 $^4 $)94@@ T ! ! 7^^^__ _ dC  b6DDTD%=11 bc9[9[VZ9[9[9[6[6[ b`aa aCCCCdCCC  &u--E%jjG   ! # #33G3G(G(G} !cddd >>>29>>>  $0$4$45Lb$Q$Q !%%%%%lq%%% !*?)C)C)E)E &M - - &&00'A'E'Ea'H'H$+/+;>N+N(&&';<<<#^^D,$66+66 H     4d ; ; ;792 4 F* & &a<<&--bhqkk:::(// bT0B0BCCCC&--*/'9I0I*J"$ )// %^%^U6FU\L\C\=]%^%^%^__ffkmfnng%FF1412DUU UUL   ^T 2 2 2 KK % % %RR=QRRRRRG$$$$ 4$$$ ,1;7KQR+S+S+SD' (  $ $[ 1T5H HQQSS -3A6N)S00!N>::;sSSTWqSqAr=s=s;st #3I#>  #3 [k"d&99d;>OSWSj>jk[!F FJF#AAAqD)+;AAAqD+AA B#DN>BBBBrcv|jj}|jj}d|D}t||zdgzS)Ncg|] }|dk| S)rr)rcnames rrz2CsmProcessor.model_input_names..>s$(r(r(r$[_cq[q[q[q[q[qrr)rImodel_input_namesrHrg)rGtokenizer_input_namesfeature_extractor_input_namess rrzCsmProcessor.model_input_names7sO $ @(,(>(P%)s(r:W(r(r(r%),IIMcLddeeerra)NNNN)NFrw)rrrrF staticmethodr[r rrrgrr!rvrrrboolrprpropertyr __classcell__)rJs@rr<r<Es|  TTTTTT,###\#J 4 44Z$sTz"22 4+, 4 4 4 4D$(%*36 OCOC++d9o=EV@WWZ^^OCD OCd{ OC %*DL OC +, OCOCOC^OCbffXfffffrr<)rMpathlibrtypingrrqrutilsrrrrm soundfilerr audio_utilsr r feature_extraction_utilsr processing_utilsr r rrtokenization_utils_baserrrr!r<__all__rrrrs OOOOOOOOOOLLL99999999444444UUUUUUUUUUUUCCCCCCCC 1 1 1 1 1[ 1 1 1 1)*yfyfyfyfyf>yfyfyfx  r