i9ddlZddlZddlmZmZddlmZddlm Z m Z m Z ddl m Z ddlmZmZerddlZej$eZGdd e d ZGd d e Zd gZy)N) AudioInputmake_list_of_audio) BatchFeature)ProcessingKwargsProcessorMixinUnpack) TextInput)is_torch_availableloggingc*eZdZddidddddddd Zy ) GlmAsrProcessorKwargspaddingTi> max_length) sampling_ratereturn_attention_maskrptleft)return_tensors padding_side) text_kwargs audio_kwargs common_kwargsN)__name__ __module__ __qualname__ _defaults}/var/www/vps2.regionflexible.com/Desarrollo/venv/lib/python3.12/site-packages/transformers/models/glmasr/processing_glmasr.pyrr(s2 t #%)# #"  IrrF)totalc eZdZdZ dfd ZddZdZdZ dd ee ezd e dzd e dzd e e def d Zede efdZ dd ee eze zdee ezdzd e e defdZdddZdZd edefdZxZS)GlmAsrProcessora Constructs an GlmAsr processor which wraps an GlmAsr feature extractor and an GlmAsr tokenizer into a single processor. [`GlmAsrProcessor`] offers all the functionalities of [`WhisperFeatureExtractor`] and [`Qwen2TokenizerFast`]. See the [`~GlmAsrProcessor.__call__`] for more information. Args: feature_extractor ([`WhisperFeatureExtractor`]): The feature extractor is a required input. tokenizer ([`Qwen2TokenizerFast`]): The tokenizer is a required input. chat_template (`Optional[str]`, *optional*): The Jinja template to use for formatting the conversation. If not provided, the tokenizer's default chat template will be used. audio_token (`Optional[str]`, *optional*, defaults to `"<|pad|>`"): Special token used to represent audio inputs in the chat template. default_transcription_prompt (`str`, *optional*, defaults to `"Please transcribe this audio into text"`): Default prompt to use for transcription tasks when applying transcription requests. max_audio_len (`int`, *optional*, defaults to 655): Maximum length of audio sequences in seconds. Audio longer than this will be truncated. 655 gives approximately 8192 tokens, corresponding to the maximum sequence length of the text model. Nc||_|j||_||_||_t ||||y)N) chat_template) audio_tokenconvert_tokens_to_idsaudio_token_iddefault_transcription_prompt max_audio_lensuper__init__)selffeature_extractor tokenizerr%r&r)r* __class__s r r,zGlmAsrProcessor.__init__RsE''==kJ,H)* *I]Srreturncdd}dD]\}}}|d|zz|dz z dz |zdz}||z |zdz}|S)N))rr4)r4rr5r4r)r- audio_lengths merge_factorr kernel_sizestride num_tokenss r _get_audio_token_lengthz'GlmAsrProcessor._get_audio_token_lengthasc ,B ` (G[&*Q[8K!OLqPU[[^__M `$l2|CaG rctjtj|jd|Dcgc]}|jc}}|j |}t j t j|j}t|D]*\}} |j|j| z||||<,|Scc}w)N) torchstacksplitsumr;recompileescaper& enumeratesub) r-text padding_maskper_sample_windowssr6audio_tokens_lengthsaudio_token_patterni audio_lengths r _expand_audio_tokensz$GlmAsrProcessor._expand_audio_tokensis ekk,BRBRSUBVXj6k$lQUUW$lm #;;MJ jj43C3C)DE()=> XOA|)--d.>.>.MtTUwWDG X %msCc ||jk(SN)r()r- input_idss r _get_audio_tokens_maskz&GlmAsrProcessor._get_audio_tokens_maskqsD////rFrGaudio output_labelskwargsc |jtfd|jji|}|d}|d}|j d}|dk7r"t |j jdt|tr|g}n3t|ttfrtd|Ds t di} |t|}t|t|k7r$t d t|d t|d t|d |j j"z} t|j$|j j"z} g} g} |D]}t|j&d }t)d|| zdz | z}|| kDr` placeholders in the text based on the post-pool frame counts of the audio windows, then tokenizes the provided strings as-is, and extracts log-mel features with [`WhisperFeatureExtractor`]. If `audio` is `None`, no audio processing is performed and the text is tokenized as-is (LM-only behavior). Args: text (`str` or `list[str]`): Input sequence or batch of sequences. audio (`np.ndarray` or `list[np.ndarray]`): Input audio or batch of audios as NumPy arrays. If provided, there must be as many `text` inputs as `audio` inputs. output_labels (bool, *optional*, default=False): Whether to return labels for training. Returns: [`BatchFeature`]: A dictionary with tokenized text (`input_ids`, `attention_mask`) and audio features (`input_features`, `input_features_mask`). tokenizer_init_kwargsrrrrz% only supports `return_tensors='pt'`.c3<K|]}t|tywrQ isinstancestr).0ts r z+GlmAsrProcessor.__call__..s9[QR*Q:L9[zAInvalid input text. Please provide a string, or a list of stringszGot z text but z audios; they must match 1:1.rrr4zAudio duration (z.1fz s) exceeds zs; truncating to first zs.attention_maskinput_features_maskrRilabels)data tensor_type) _merge_kwargsrr/ init_kwargsget ValueErrorr0rr[r\listtupleallrlenintr. chunk_lengthr*shapemaxloggerwarningappendminrangepoprOclonerS pad_token_idr)r-rGrTrUrV call_kwargsrrr audio_inputs window_size max_windowsrI flat_chunksaudio_el n_samplesn_wintime_caprMstartendrH text_inputsrdrcs r __call__zGlmAsrProcessor.__call__ts,:)d(( ! "&.."<"<  "-0 ">2 $)9: T ! 7 788]^_ _ dC 6DTD%=1c9[VZ9[6[`a a  &u-E4yCJ& 4D {*SZLHe!fggl?;d>T>T>a>aabKd00D4J4J4W4WWXK,. ,.K! <q 12 A K 7! ; KL;&NN*9|O7T+TUX*YYdeiewewdxyPQUQcQcPddfg(E"))%0y%+*=>uC&&xc':;< <"2411+NNL'++,<=L2>L. /,,TF4..v6 7<@F6T^^888 9#DN>BBrc|jj}|jj}ttj ||zdgzS)Nrb)r/model_input_namesr.rjdictfromkeys)r- tok_names fea_namess r rz!GlmAsrProcessor.model_input_namessDNN44 **<< DMM)i"7;P:Q"QRSSrpromptc t|tr|g}nt|ttfr |rt d|Dr t|}nstt |}t rU|Dcgc]J}t|tjr,|jjjn|L}}t|}|dk(r td||jg|z}nt|tr|g|z}nt|ttfr}t||k7rtdt|d|dg}|D]L}||j|j!t|tr|j|Ct!dn t!dt#||D cgc](\} } d t| trd | d nd | d d | dgdg*} } } |j$| fdddd|Scc}wcc} } w)a  Prepare inputs for automatic speech recognition without manually writing the default transcription prompt. Args: audio (`str`, `list[str]`, `np.ndarray`, `torch.Tensor`, `list[np.ndarray]`, `list[torch.Tensor]`): Audio to transcribe. Strings are interpreted as local paths or URLs and will be loaded automatically by the chat template loader; NumPy arrays and PyTorch tensors are forwarded directly. prompt (`str` or `list[str]`, *optional*): Custom prompt(s) to include in the user turn. A list must be the same length as the batch. When `None`, each sample uses `"Transcribe the input speech."`. **kwargs: Additional keyword arguments forwarded to [`~GlmAsrProcessor.apply_chat_template`] (for example `text_kwargs`, `audio_kwargs`, ...). Returns: [`BatchFeature`]: Processor outputs ready to be passed to [`GlmAsrForConditionalGeneration.generate`]. c3<K|]}t|tywrQrZ)r]els r r_z>GlmAsrProcessor.apply_transcription_request..s?dXZ 2s@S?dr`rz)`audio` must contain at least one sample.z Received z prompt(s) for z$ audio sample(s); counts must match.z'Each prompt must be a string or `None`.z<`prompt` must be a string, a sequence of strings, or `None`.userrT)typepath)rrTrG)rrG)rolecontentT)tokenizeadd_generation_prompt return_dict)r[r\rjrkrlrr r>Tensordetachcpunumpyrmrir)rt TypeErrorzipapply_chat_template) r-rTrrV audio_itemsr batch_sizepromptsitem prompt_text audio_item conversationss r apply_transcription_requestz+GlmAsrProcessor.apply_transcription_requests2 eS !38'K e} -%C?d^c?d889JFG  $h+G u .6{j( F }OJ>tDYGYZs?c&|j|i|S)z)BC as previous examples used batch_decode)r)r-rrVs r batch_decodezGlmAsrProcessor.batch_decode7st{{D+F++rc@|j}dD]1}|j|s|t|dj}n|jdr|ddj}t|dk\r%|d|dk(r|ddvr|ddj}|S) zi Remove the assistant prefix and surrounding quotes from a decoded transcription string. )z"The spoken content of the audio isz!The transcription of the audio isz!The content of the input audio isN.r=r5r>"'r4)strip startswithrmendswith)r-rGstrippedprefixs r rz2GlmAsrProcessor._strip_assistant_prefix_and_quotes;s ::<  F ""6*#CKM288:    S !}**,H x=A (1+""=(1+Q[B["~++-Hr)Nz<|pad|>z&Please transcribe this audio into texti)r6 torch.Tensorr1r)NFrQ)rrr__doc__r,r;rOrSr rjrboolr rrrpropertyr\rrrrr __classcell__)r0s@r r#r#9s'8%M T0 $(%* ]C$y/)]CD ]Cd{ ]C ./ ]C  ]C~T49TT*.O T#Y+O d3i$&O ./ O  O b*/ ,ssrr#)rBrnp audio_utilsrrfeature_extraction_utilsrprocessing_utilsrrr tokenization_utils_baser utilsr r r> get_loggerrrrrr#__all__rrr rsf, 94HH00   H %,E"XnXv  r