o wi@sddlmZddlmZmZddlZddlmZddlm Z m Z ddl m Z ddl mZddlmZdd lmZmZdd lmZdd lmZdd lmZmZmZmZd dlmZd dlm Z m!Z!m"Z"m#Z#m$Z$m%Z%m&Z&m'Z'ddl(m)Z)m*Z*ddl+m,Z,e-e.Z/eeddGdddeZ0eddeGdddeZ1Gddde&Z2Gddde'Z3Gddde$Z4Gd d!d!e!Z5Gd"d#d#e"Z6eGd$d%d%e%Z7Gd&d'd'ej8Z9ed(dGd)d*d*e#e Z:Gd+d,d,ej8Z;eGd-d.d.e%ZdS)3) dataclass)OptionalUnionN)Cache DynamicCache)GenerationMixin)create_causal_mask)FlashAttentionKwargs)BaseModelOutputWithPastCausalLMOutputWithPast)PreTrainedModel)Unpack) ModelOutputauto_docstringcan_return_tuplelogging) AutoModel)KwargsForCausalLMLlamaAttentionLlamaDecoderLayerLlamaForCausalLMLlamaMLP LlamaModel LlamaRMSNormLlamaRotaryEmbedding) CsmConfigCsmDepthDecoderConfig)CsmGenerationMixinz: Base class for the model autoregressive outputs. ) custom_introc@seZdZUdZdZeejed<dZ ejed<dZ ee e ejed<dZ ee ejdfed<dZ ee ejdfed<dZeejed <dZejed <dZee e ejed <dZee ejdfed <dZee ejdfed <dZeejed<dS)CsmOutputWithPasta# loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided): Language modeling loss (for next-token prediction). logits (`torch.FloatTensor` of shape `(batch_size, sequence_length, config.vocab_size)`): Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax). past_key_values (`tuple(tuple(torch.FloatTensor))`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`): Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of shape `(batch_size, num_heads, sequence_length, embed_size_per_head)`) Contains pre-computed hidden-states (key and values in the self-attention blocks) that can be used (see `past_key_values` input) to speed up sequential decoding. depth_decoder_loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided): Language modeling loss (for next-token prediction) of the depth decoder model. depth_decoder_logits (`torch.FloatTensor` of shape `(batch_size, sequence_length, config.vocab_size)`): Prediction scores of the depth decoder (scores for each vocabulary token before SoftMax). depth_decoder_past_key_values (`tuple(tuple(torch.FloatTensor))`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`): Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of shape `(batch_size, num_heads, sequence_length, embed_size_per_head)`) depth_decoder_hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`): Tuple of `torch.FloatTensor` (one for the output of the embeddings, if the model has an embedding layer, + one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`. Hidden-states of the model at the output of each layer plus the optional initial embedding outputs. depth_decoder_attentions (`tuple(torch.FloatTensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`): Tuple of `torch.FloatTensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length, sequence_length)`. backbone_loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided): Language modeling loss (for next-token prediction) of the backbone model. Nlosslogitspast_key_values. hidden_states attentionsdepth_decoder_lossdepth_decoder_logitsdepth_decoder_past_key_valuesdepth_decoder_hidden_statesdepth_decoder_attentions backbone_loss)__name__ __module__ __qualname____doc__r#rtorch FloatTensor__annotations__r$r%tupler&r'r(r)r*r+r,r-r6r6`/home/ubuntu/sommelier/.venv/lib/python3.10/site-packages/transformers/models/csm/modular_csm.pyr"0s r"z[ The bare Csm Model outputting raw hidden-states without any specific head on top. c@sDeZdZeZdZdZdgZdgZdZ dZ dZ dZ dZ dZddZdS)CsmPreTrainedModelmodelTCsmDecoderLayerr%cCs|jj}t|tjr"|jjjd|d|jdur |jj dSdSt|tj rC|jjjd|d|j durA|jj|j  dSdSt|t ra|j }t|dD] }|jj|jd|dqQdSt|tro|jjddSdS)Ng)meanstdrg?)configinitializer_range isinstancennLinearweightdatanormal_biaszero_ Embedding padding_idxCsmCodebooksHead num_codebooksrange CsmRMSNormfill_)selfmoduler<rJir6r6r7 _init_weightsws&      z CsmPreTrainedModel._init_weightsN)r.r/r0r config_classbase_model_prefixsupports_gradient_checkpointing_no_split_modules_skip_keys_device_placement_supports_flash_attn_2_supports_sdpa_supports_cache_class_supports_quantized_cache_supports_static_cache_supports_attention_backendrQr6r6r6r7r8bs r8c@ eZdZdS)rLNr.r/r0r6r6r6r7rLrLc@r])CsmRotaryEmbeddingNr^r6r6r6r7r`r_r`c@r])CsmMLPNr^r6r6r6r7rar_rac@r]) CsmAttentionNr^r6r6r6r7rbr_rbc@r])r:Nr^r6r6r6r7r:r_r:cseZdZeZfddZee          ddej de ej de ej de ej de e d e ej d e ed e ed e ed e ej deedeeeffddZZS)CsmDepthDecoderModelcs>t|t|j|j|j|_tj|j|j dd|_ dSNF)rE) super__init__r@rGrJ vocab_sizebackbone_hidden_size embed_tokensrA hidden_sizeinputs_embeds_projectorrNr= __class__r6r7rfs zCsmDepthDecoderModel.__init__N input_idsbackbone_last_hidden_stateattention_mask position_idsr% inputs_embeds use_cacheoutput_attentionsoutput_hidden_statescache_positionflash_attn_kwargsreturnc  Ks\|durtjstdd}|dur|n|jj}| dur | n|jj} |dur*|n|jj}|du|duAr:t d|j rI|j rI|rItdd}|rR|durRt }| dur|dur^| nd} |duri|jdn|jd} |duru|jn|j}tj| | | |d} |durtj| ddd }||j}|||}| ddk}|dur||dddf<n tjs|rtd ||}t|j||| ||d }|}| d}|||}| rd nd}|rd nd}|jd|jjD]'}| r||f7}||f|||||| |d | }|d}|r||df7}q||}| r!||f7}t||r(|nd||dS)aJ backbone_last_hidden_state (`torch.FloatTensor` of shape `(batch_size, backbone_hidden_size)`, *optional*): The last hidden state of the backbone model. Such input is required when the first codebook token (the one generated by the backbone model) is provided in the `input_ids` argument. NzCustom `position_ids` were provided but will be ignored. CSM depth decoder automatically determines position_ids from `cache_position` and as it requires them to be identical across the batch, the provided position_ids will be ignored.z;You must specify exactly one of input_ids or inputs_embeds.zX`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`.Frrdevice)minzvWhen the first codebook token is provided, `backbone_last_hidden_state` should also be provided for correct inference.)r= input_embedsrqrwr%rrr6)rqrrpast_key_valuerurtrwposition_embeddings)last_hidden_stater%r&r')r2compiler is_compilinglogger warning_oncer=rurvrt ValueErrorgradient_checkpointingtrainingrget_seq_lengthshaper{arangeclamprgriwarningrkr unsqueeze rotary_emblayersnum_hidden_layersnormr )rNrorprqrrr%rsrtrurvrwrxpast_seen_tokensinputs_seq_lengthr{ codebook_idxsoffsetinput_ids_are_first_codebook causal_maskr&rall_hidden_statesall_self_attns decoder_layer layer_outputsr6r6r7forwards              zCsmDepthDecoderModel.forward) NNNNNNNNNN)r.r/r0rrRrfrrr2 LongTensorrr3Tensorrboolrr rr5r r __classcell__r6r6rmr7rcsP      rccs&eZdZfddZdddZZS)rIcs0t||_tt|jd|||_dS)Nr)rerfrJr@ Parameterr2emptyrB)rNrjrJrgrmr6r7rf"s  zCsmCodebooksHead.__init__Ncsf|durjd}|jt|n |d}|j|fddtjdDtjddS)Nrc s2g|]}tjdd|ddf|jqSN)r@ functionallinearT).0 codebook_idxcodebook_weightr&r6r7 /s$z,CsmCodebooksHead.forward..rdim)rrBr2rrKstack)rNr&rw seq_lengthrr6rr7r's    zCsmCodebooksHead.forwardrr.r/r0rfrrr6r6rmr7rI!s rIa$ The CsmDepthDecoder Model transformer, with a [`CsmCodebooksHead`] on top, which can be seen a position-specific language modeling head, allowing to use a different linear layer for each codebook (e.g. position 0 is the first codebook and uses the first codebook head, etc.) c!s:eZdZdZdZdZfddZddZddZ    dde j d e e d e e j d e e j d e e j f fd d Zee            dde j de e j d e e jde e j d e ee ee j fd e e j de e j de ede ede ed e e j deee jfdeedeeeffddZZS)CsmDepthDecoderForCausalLMNcs2t||`t|j|j|j|_t||_ dSr) rerflm_headrIrjrJrgcodebooks_headrcr9rlrmr6r7rfDs z#CsmDepthDecoderForCausalLM.__init__cCtdNzNot needed for CsmAttributeErrorrNr6r6r7get_output_embeddingsJz0CsmDepthDecoderForCausalLM.get_output_embeddingscCrrrrNnew_embeddingsr6r6r7set_output_embeddingsMrz0CsmDepthDecoderForCausalLM.set_output_embeddingsror%rqrsrwc  sHtj|||||fi|}|dddk}|s|d|d|S)Nrwrrprr)reprepare_inputs_for_generationpop) rNror%rqrsrwkwargs model_inputsis_first_generation_steprmr6r7rPs   z8CsmDepthDecoderForCausalLM.prepare_inputs_for_generationrrprrlabelsrtrurvlogits_to_keeprryc  Ks| dur| n|jj} | dur| n|jj} |jd|||||||| | | d | }|d}t| trA| dkr:tdd}n t| d}n| }||dd|ddf| durW| |nd}|}d}|dur}|dddf}|j d|d|jj |d| }t |||j |j |jdS) a backbone_last_hidden_state (`torch.FloatTensor` of shape `(batch_size, backbone_hidden_size)`, *optional*): The last hidden state of the backbone model. Such input is required when the first codebook token (the one generated by the backbone model) is provided in the `input_ids` argument. labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*): Labels for computing the masked language modeling loss. Indices should either be in `[0, ..., config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`. N) rorprqrrr%rsrtrurvrwrr.)r$rrg shift_labels)r#r$r%r&r'r6)r=rurvr9r?intslicer contiguous loss_functionrgr r%r&r')rNrorprqrrr%rsrrtrurvrwrroutputsr& slice_indicesr$r#rr6r6r7rfsT   & z"CsmDepthDecoderForCausalLM.forwardNNNN) NNNNNNNNNNNr)r.r/r0_tied_weights_keys_tp_plan_pp_planrfrrr2rrrr3rrrrrlistrrrrr5r rrr6r6rmr7r8s       rcs$eZdZfddZddZZS)CsmBackboneModelEmbeddingscsDtt|j|j|j|_|jdt |j|jdddS)Naudio_tokens_offsetsF) persistent) rerfr@rGrJrgrjembed_audio_tokensregister_bufferr2rrlrmr6r7rfs  z#CsmBackboneModelEmbeddings.__init__cCs |||j}|jdd}|S)Nrr)rrsum)rNror}r6r6r7rs z"CsmBackboneModelEmbeddings.forwardrr6r6rmr7rs rcs0eZdZfddZeefddZZS)CsmBackboneModelcst|t||_dSr)rerfrrirlrmr6r7rfs zCsmBackboneModel.__init__c stjdi|S)a& input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length, num_codebooks) or (batch_size, sequence_length)`): 1. (batch_size, sequence_length): corresponds to the input sequence prepared with the processor from the text prompt. Such input requires `input_values` to be provided so that audio can be encoded in codebook tokens and then merged with the text tokens. 2. (batch_size, sequence_length, num_codebooks): codebook tokens generated during the autoregressive decoding. Such input is not meant to be used by end users. Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and [`PreTrainedTokenizer.__call__`] for details. [What are input IDs?](../glossary#input-ids) Nr6)rer)rN super_kwargsrmr6r7rszCsmBackboneModel.forward)r.r/r0rfrrrrr6r6rmr7rs  rz The Csm model consists of two llama-like auto-regressive transformer models: a backbone model that predicts the first codebook token and a depth decoder that predicts the other codebook tokens. c#seZdZddgZfddZddZddZd d Zd d Zd dZ e fddZ fddZ    d*de ejde ejde ejde ejde ejf ddZ    d*dejde ede ejde ejde ejf fdd Zee             !d+dejde ejde ejde ejd"e ejde eeeejfde ejde ejd#e ed$e ed%e ede ejd&eeejfd'eedeeeffd(d)ZZ S),CsmForConditionalGenerationz5backbone_model.embed_tokens.embed_audio_tokens.weightz'depth_decoder.model.embed_tokens.weightcspt||j|_tj|j|jdd|_t|j|j|_ t ||_ t |j|_t|j|_|dSrd)rerfrgr@rArjrrGtext_vocab_sizeembed_text_tokensr _from_configbackbone_modelrdepth_decoder_config depth_decoderr from_config codec_config codec_model post_initrlrmr6r7rfs   z$CsmForConditionalGeneration.__init__cCs|jjSrrrirr6r6r7get_input_embeddingsrz0CsmForConditionalGeneration.get_input_embeddingscCs ||j_dSrr)rNvaluer6r6r7set_input_embeddingss z0CsmForConditionalGeneration.set_input_embeddingscCs|jSrrrr6r6r7rsz1CsmForConditionalGeneration.get_output_embeddingscCs ||_dSrrrr6r6r7rs z1CsmForConditionalGeneration.set_output_embeddingscCs(|jjr||jjj|jjjdSdSr)r=tie_codebooks_embeddings_tie_or_clone_weightsrrirrr9rr6r6r7 _tie_weightss z(CsmForConditionalGeneration._tie_weightscs|ddrtj|i|\}}n tj|i|}dtfddt|jD}t|jjddi||D] }t |j|q?d|vrR||fS|S)Noutput_loading_infoFdepth_decoder_cs(i|]\}}|r|d|qSr) startswith)rattrrprefix prefix_lenr6r7 s  z?CsmForConditionalGeneration.from_pretrained.._from_model_config) getrefrom_pretrainedlenvarsgeneration_configitemsrupdatedelattr)clsargsrr9 loading_infodepth_decoder_attrsrrmrr7rs   z+CsmForConditionalGeneration.from_pretrainedcsVd}|jj}|dd|D] \}}t|j|||qtj|i|dS)Nrtransformers_version)rr to_diff_dictrrsetattrresave_pretrained)rNrrrrrrrmr6r7r s   z+CsmForConditionalGeneration.save_pretrainedNro input_valuesinput_values_cutoffsrrycsF||}|durtj|d}||dk}||dk}tj||jd t |d}|| dk}t jg}t ||D]?\} } | | dk} t| jddD]+} | | } | | d} | d| | f}|j| d}|jdd}||dqUqBtdd |Dtfd d |D}|j|}Wdn1swY|jj}||k}|j|}||||<tjdd|jjf|jtjd |jj}|j|d}||jj k}|!|"d||<|dur| d!dd|jj}||||<|||<|d kj#dd}d||d|dddf<|}||dS)a Merges the input_ids and input_values to produce a single inputs_embeds tensor: 1 - Infers the codec model on the input_values to retreive codebook token. 2 - Embeds codebook tokens and places them at the correct positions in the inputs_embeds tensor. 3 - If labels are provided, expands them to match codebook dimensions and position the target codebook tokens in the inputs_embeds tensor. Args: input_ids (`torch.Tensor` of shape `(batch_size, sequence_length)`): The input ids to embed. input_values (`torch.Tensor` of shape `(batch_size, channels, audio_sequence_length)`): The audio input values to embed. input_values_cutoffs (`torch.Tensor` of shape `(batch_size, max_num_audio)`): The cutoffs of the audio input values relative to its batch index, padded with -1 when no audio. Nrrrrzr.css|]}|jdVqdS)rN)rrelr6r6r7 ZszQCsmForConditionalGeneration._merge_input_ids_with_input_values..c s,g|]}tj|ddd|jdfqS)r)r@rpadrr max_audio_framesr6r7r\s,zRCsmForConditionalGeneration._merge_input_ids_with_input_values..)r{dtypeiTas_tuple)rsr)$rr@rr diffr2rmaxr{expandrrno_gradziprKrrencode audio_codes transposeappendrget_audio_codes_maskr=audio_token_idrrionesrJlongcodebook_eos_token_idsqueezeaudio_eos_token_idrepeatrnonzero)rNrorrrrs audio_lengthsinput_values_maskaudio_tokens_listbatch_input_valuesbatch_input_values_cutoffsrP start_idxend_idx audio_batch codec_outputs codebook_idsbatched_audio_token_idsaudio_codes_maskraudio_token_mask audio_embedsaudio_eos_frame_idsaudio_eos_embedsaudio_eos_token_masklabels_expanded depth_decoder_ignore_frames_idxsr6r r7"_merge_input_ids_with_input_values*s\            z>CsmForConditionalGeneration._merge_input_ids_with_input_valuesr%rqrsrwc  stjd |||||d|}|dur>|jdkr>|ddur>|j||d|d|dd}||d|ddd|S) N)ror%rqrsrwrrsrrr)rorrr)rsrror6)rerndimrr8r) rNror%rqrsrwrr merged_inputsrmr6r7r|s(  z9CsmForConditionalGeneration.prepare_inputs_for_generationrrrrtrurvrrc Ks| dur| n|jj} | dur| n|jj} |dur/|jdkr/|||||}|d}|d}d}|jd|||||| | | | d |}|d}t| trPt| dn| }| |dd|ddf}d}d}d}d}|dur|dddddf}|j d|||jj d|}|ddddddfd kj d d }||d d|jj df}tjj|d dd}|jdd}||d|ddddf}||}|j||| | | d|d}|j}||}t|||||j|j|j|dur|jnd|dur|jnd|dur|jnd|dur |jd Sdd S)a input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length, num_codebooks) or (batch_size, sequence_length)`): 1. (batch_size, sequence_length): corresponds to the input sequence prepared with the processor from the text prompt. Such input requires `input_values` to be provided so that audio can be encoded in codebook tokens and then merged with the text tokens. 2. (batch_size, sequence_length, num_codebooks): codebook tokens generated during the autoregressive decoding. Such input is not meant to be used by end users. Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and [`PreTrainedTokenizer.__call__`] for details. [What are input IDs?](../glossary#input-ids) input_values_cutoffs (`torch.Tensor` of shape `(batch_size, max_num_audio)`, *optional*): Specify the end positions of audio segments within each batch entry, relative to the concatenated audio input. If a batch entry has fewer segments than the maximum, it is padded with -1. For example, in a batch of 2 sequences where the first contains 2 audio segments of length l1, and the second contains 1 audio segment of length l2, the input_values_cutoffs would be: [[l1, 2 * l1], [l2, -1]]. labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*): Labels for computing the masked language modeling loss. Indices should be in `[config.audio_token_id, -100, -101]`. Requires targeted `input_values` to be provided as audio tokens will be infered from it using the `codec_model`. - `config.audio_token_id` indicates an audio frames (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) Such labels can be prepared using `output_labels=True` when calling [`CsmProcessor`]. logits_to_keep (`int` or `torch.Tensor`, *optional*): Kept for compatibility. Does not support another value than: 1. `0`, which is equivalent to keeping all logits, used in the training regime 2. `1`, which is equivalent to keeping only the last logit, used in the generation regime Example: ```python >>> import torch >>> from transformers import CsmForConditionalGeneration, AutoProcessor >>> from datasets import load_dataset, Audio >>> model_id = "sesame/csm-1b" >>> torch_device = "cuda" if torch.cuda.is_available() else "cpu" >>> processor = AutoProcessor.from_pretrained(model_id) >>> ds = load_dataset("hf-internal-testing/dailytalk-dummy", split="train") >>> # ensure the audio is 24kHz >>> ds = ds.cast_column("audio", Audio(sampling_rate=24000)) >>> conversation = [] >>> # prepare a conversation with text and corresponding audio >>> for text, audio, speaker_id in zip(ds[:4]["text"], ds[:4]["audio"], ds[:4]["speaker_id"]): ... conversation.append( ... { ... "role": f"{speaker_id}", ... "content": [{"type": "text", "text": text}, {"type": "audio", "path": audio["array"]}], ... } ... ) >>> inputs = processor.apply_chat_template( ... conversation, ... tokenize=True, ... return_dict=True, ... output_labels=True, ... ).to(torch_device) >>> model = CsmForConditionalGeneration.from_pretrained(model_id, device_map=torch_device) >>> output = model(**inputs) >>> output.loss.backward() ```Nrrsr) rorqrrr%rsrtrurvrwr)r$rrgrrrr.r)rTr)rorprtrurv return_dictr) r#r-r(r$r%r&r'r)r*r+r,r6)r=rurvr9r8rr?rrrrrgallrJr@rr r$rr#r"r%r&r'r$)rNrorrqrrrr%rsrrtrurvrwrrr:backbone_outputsbackbone_hidden_statesrbackbone_logitsr#r-r(depth_decoder_outputsbackbone_labels train_maskdepth_decoder_input_ids train_idxsbackbone_last_hidden_statesdepth_decoder_labelsr6r6r7rsV   (  z#CsmForConditionalGeneration.forwardr) NNNNNNNNNNNNr)!r.r/r0rrfrrrrr classmethodrrrr2rr8rrr3rrrrrrrrrr5r"rrr6r6rmr7rs     U       r)r8rrcrr)? dataclassesrtypingrrr2torch.nnr@ cache_utilsrr generationr masking_utilsr modeling_flash_attention_utilsr modeling_outputsr r modeling_utilsr processing_utilsrutilsrrrrautorllama.modeling_llamarrrrrrrrconfiguration_csmrrgeneration_csmr get_loggerr.rr"r8rLr`rarbr:rcModulerIrrrr__all__r6r6r6r7sd        (  ,"v h