bi%ddlZddlmZmZmZddlZddlZddlm Z m Z ddl m Z ddl mZddlmZmZdd lmZmZdd lmZmZmZmZd d lmZerddlZej8eZd ZdZ dZ!GddeZ"y)N)ListOptionalUnion) AutoTokenizerUMT5EncoderModel) FrozenDict)ClassifierFreeGuidance)is_ftfy_availablelogging)ModularPipelineBlocks PipelineState) ComponentSpec ConfigSpec InputParam OutputParam)WanModularPipelinectj|}tjtj|}|j SN)ftfyfix_texthtmlunescapestriptexts c/home/cdr/jupyterlab/.venv/lib/python3.12/site-packages/diffusers/modular_pipelines/wan/encoders.py basic_cleanr %s3 == D ==t, -D ::<cTtjdd|}|j}|S)Nz\s+ )resubrrs rwhitespace_cleanr&+s$ 66&#t $D :: isinstancestrlist ValueErrortype) block_states r check_inputszWanTextEncoderStep.check_inputsgs]    );--s3J{GYGY[_<`QRVWbWiWiRjQklm m=a3 *r!r>max_sequence_lengthdevicec|jj}t|tr|gn|}|Dcgc] }t |}}|j |d|dddd}|j |j}}|jdjdj} |j|j||j|j} | j||} t| | D cgc] \}} |d|  } }} tj| Dcgc]J}tj ||j#||j%dz |j%dgLc}d} | Scc}wcc} }wcc}w) N max_lengthTpt)paddingrV truncationadd_special_tokensreturn_attention_maskreturn_tensorsrr)dim)dtyperT)r3r^rLrMr(r4 input_idsattention_maskgtsumlongtolast_hidden_stateziprHstackcat new_zerossize) componentsr>rSrTr^u text_inputstext_input_idsmaskseq_lensrCvs r_get_t5_prompt_embedsz(WanTextEncoderStep._get_t5_prompt_embedsnsj''--'4&&+12a,q/22 **  *#"&+  +44k6P6P771:>>a>(--/"//0A0A&0I477SY?[mm %((uV(D +.}h+GH41a2AH H ^k lYZUYY1;;':QVVAY'Fq RS T lrs )3I lsE>?F$AF Nnum_videos_per_promptprepare_unconditional_embedsr?rCrGc |xs |j}t|tr|gn|}| t|n|jd} |t j ||||}|r||xsd}t|tr| |gzn|}|:t|t|ur$tdt|dt|d| t|k7r!td|dt|d|d| d t j ||||}|j\} } } |jd |d }|j| |z| d }|r)|jd |d }|j| |z| d }||fS) a Encodes the prompt into text encoder hidden states. Args: prompt (`str` or `List[str]`, *optional*): prompt to be encoded device: (`torch.device`): torch device num_videos_per_prompt (`int`): number of videos that should be generated per prompt prepare_unconditional_embeds (`bool`): whether to use prepare unconditional embeddings or not negative_prompt (`str` or `List[str]`, *optional*): The prompt or prompts not to guide the image generation. If not defined, one has to pass `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is less than `1`). prompt_embeds (`torch.Tensor`, *optional*): Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not provided, text embeddings will be generated from `prompt` input argument. negative_prompt_embeds (`torch.Tensor`, *optional*): Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input argument. max_sequence_length (`int`, defaults to `512`): The maximum number of text tokens to be used for the generation process. rz?`negative_prompt` should be the same type to `prompt`, but got z != .z`negative_prompt`: z has batch size z, but `prompt`: zT. Please make sure that passed `negative_prompt` matches the batch size of `prompt`.r) _execution_devicerLrMlenshaper*rrrP TypeErrorrOrepeatview) rkr>rTrsrtr?rCrGrS batch_sizebs_embedseq_len_s r encode_promptz WanTextEncoderStep.encode_promptsL7:77'4&&$*$6S[Mr?rCrGset_block_state)r0rkrrQs r__call__zWanTextEncoderStep.__call__s**51  +&3=3D3D3S3SVW3W 0'99             4 4  ' '#'    %  . UK05  r!)NrTNNNi)__name__ __module__ __qualname__ model_namepropertyrMr1rrr:rr<rrArrJ staticmethodrRrintrHrTrrrboolrIrno_gradrrrr.r!rr*r*6sJ _S__  T-%8    $z"2 Z(   d;&7  nn c49n%! <*.%&-1)-049=#&H5H5&H5 # H5 '+ H5 "# H5  -H5!) 6H5!H5H5TU]]_!#5!m!P]!!r!r*)#rtypingrrrregexr$rH transformersrrconfiguration_utilsr guidersr utilsr r modular_pipelinerrmodular_pipeline_utilsrrrrrr get_loggerrloggerr r&r(r*r.r!rrsk (( 8--/CWW0   H %   |!.|!r!