wi|VddlZddlmZmZmZmZddlZddlmZ ddl m Z m Z ddl mZddlmZmZmZddlmZdd lmZdd lmZd d lmZmZd dlmZejeZ GddeZ!dS)N)ListOptionalTupleUnion) functional)CLIPTextModelWithProjection CLIPTokenizer)CLIPTextModelOutput)PriorTransformerUNet2DConditionModel UNet2DModel)UnCLIPScheduler)logging) randn_tensor)DiffusionPipelineImagePipelineOutput)UnCLIPTextProjModelc |eZdZUdZdgZeed<eed<eed<e ed<e ed<e ed<e ed<e ed <e ed <e ed <d Z dedede de dede de d e d e d e ffd ZdZ d(deeeefdeejfdZej d)deeeeefdededededeeejeejfd eejd!eejd"eejdeeeefdeejd#ed$ed%eed&efd'ZxZS)*UnCLIPPipelineaE Pipeline for text-to-image generation using unCLIP. This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods implemented for all pipelines (downloading, saving, running on a particular device, etc.). Args: text_encoder ([`~transformers.CLIPTextModelWithProjection`]): Frozen text-encoder. tokenizer ([`~transformers.CLIPTokenizer`]): A `CLIPTokenizer` to tokenize text. prior ([`PriorTransformer`]): The canonical unCLIP prior to approximate the image embedding from the text embedding. text_proj ([`UnCLIPTextProjModel`]): Utility class to prepare and combine the embeddings before they are passed to the decoder. decoder ([`UNet2DConditionModel`]): The decoder to invert the image embedding into an image. super_res_first ([`UNet2DModel`]): Super resolution UNet. Used in all but the last step of the super resolution diffusion process. super_res_last ([`UNet2DModel`]): Super resolution UNet. Used in the last step of the super resolution diffusion process. prior_scheduler ([`UnCLIPScheduler`]): Scheduler used in the prior denoising process (a modified [`DDPMScheduler`]). decoder_scheduler ([`UnCLIPScheduler`]): Scheduler used in the decoder denoising process (a modified [`DDPMScheduler`]). super_res_scheduler ([`UnCLIPScheduler`]): Scheduler used in the super resolution denoising process (a modified [`DDPMScheduler`]). priordecoder text_proj text_encoder tokenizersuper_res_firstsuper_res_lastprior_schedulerdecoder_schedulersuper_res_schedulerzAtext_encoder->text_proj->decoder->super_res_first->super_res_lastc t|||||||||| |  dS)N) rrrrrrrr r!r")super__init__register_modules) selfrrrrrrrr r!r" __class__s z/root/.openclaw/workspace/chatterbox_venv_py311/lib/python3.11/site-packages/diffusers/pipelines/unclip/pipeline_unclip.pyr%zUnCLIPPipeline.__init__Qs`  %+)+/ 3  c|t||||}n:|j|krtd|jd|||}||jz}|S)N) generatordevicedtypezUnexpected latents shape, got z , expected )rshape ValueErrortoinit_noise_sigma)r'r/r.r-r,latents schedulers r)prepare_latentszUnCLIPPipeline.prepare_latentsmsr ?"5IfTYZZZGG}%% !c'-!c!c\a!c!cdddjj((GI66r*Ntext_model_outputtext_attention_maskc|wt|trt|nd}||d|jjdd}|j} |j|} ||ddj} | j d| j dkrtj | | su|j | dd|jjdz df} td |jjd | | ddd|jjf} || |} | j}| j}n%|d j d }|d |d}}|} ||d }||d }| |d } |rdd g|z}||d|jjdd}|j|}||j|}|j}|j}|j d}|d|}|||z|}|j d}|d|d}|||z|d}||d }tj||g}tj||g}tj|| g} ||| fS)Nr max_lengthTpt)paddingr9 truncationreturn_tensorslongest)r;r=z\The following part of your input was truncated because CLIP can only handle sequences up to z tokens: rdim) isinstancelistlenrmodel_max_length input_idsattention_maskboolr1r/torchequal batch_decodeloggerwarningr text_embedslast_hidden_staterepeat_interleaverepeatviewcat)r'promptr-num_images_per_promptdo_classifier_free_guidancer6r7 batch_size text_inputstext_input_ids text_maskuntruncated_ids removed_texttext_encoder_output prompt_embedstext_enc_hid_states uncond_tokens uncond_inputuncond_text_mask*negative_prompt_embeds_text_encoder_outputnegative_prompt_embedsuncond_text_enc_hid_statesseq_lens r)_encode_promptzUnCLIPPipeline._encode_promptxs  $(264(@(@GVaJ..$>:# )K)2N#27799<>$>:# *L ,:??AADDVLL 9=9J9J 1`. decoder_guidance_scale (`float`, *optional*, defaults to 4.0): A higher guidance scale value encourages the model to generate images closely linked to the text `prompt` at the expense of lower image quality. Guidance scale is enabled when `guidance_scale > 1`. text_model_output (`CLIPTextModelOutput`, *optional*): Pre-defined [`CLIPTextModel`] outputs that can be derived from the text encoder. Pre-defined text outputs can be passed for tasks like text embedding interpolations. Make sure to also pass `text_attention_mask` in this case. `prompt` can the be left `None`. text_attention_mask (`torch.Tensor`, *optional*): Pre-defined CLIP text attention mask that can be derived from the tokenizer. Pre-defined text attention masks are necessary when passing `text_model_output`. output_type (`str`, *optional*, defaults to `"pil"`): The output format of the generated image. Choose between `PIL.Image` or `np.array`. return_dict (`bool`, *optional*, defaults to `True`): Whether or not to return a [`~pipelines.ImagePipelineOutput`] instead of a plain tuple. Returns: [`~pipelines.ImagePipelineOutput`] or `tuple`: If `return_dict` is `True`, [`~pipelines.ImagePipelineOutput`] is returned, otherwise a `tuple` is returned where the first element is a list with the generated images. Nrz2`prompt` has to be of type `str` or `list` but is rg?)r-r)timestepproj_embeddingencoder_hidden_statesrH)rysampler, prev_timestep)image_embeddingsr_text_encoder_hidden_statesrWmps)valueT)r|ryr{ class_labelsrHr@)r}r,r?)size antialiasbicubicF)rmode align_corners)r|ryg?r rm)images)5rCstrrDrEr0typer/_execution_devicerhr set_timesteps timestepsrconfig embedding_dimr5r. enumerate progress_barrJrTpredicted_image_embeddingchunkstep prev_samplepost_process_latentsrintFpadclip_extra_context_tokensrIr!r in_channels sample_sizer|splitclampr"r interpolateinspect signature parametersrmaybe_free_model_hookscpupermutefloatnumpy numpy_to_pilr)2r'rUrVrnrorpr,rqrrrsr6r7rtrurvrwrXr-rWr_r`r[prior_timesteps_tensorritlatent_model_inputr predicted_image_embedding_uncondpredicted_image_embedding_textr}r~additive_clip_time_embeddingsdecoder_text_maskdecoder_timesteps_tensornum_channels_latentsheightwidth noise_prednoise_pred_uncondnoise_pred_text_predicted_variance image_smallsuper_res_timesteps_tensorchannelsimage_upscaledinterpolate_antialiasunetimages2 r)__call__zUnCLIPPipeline.__call__s!J  &#&& f FD)) f [[  !dVZ[aVbVb!d!deee*1-3A6J'"77 &:S&@&`DZ]`D`#8<8K8K F13NPacv9 9 5 *I **+DV*TTT!%!5!? )7 ,,  '         d//0FGGHH  DAqC^!qM?Q+>!?!?!?dq (, ",&9( )3)) ( &+ SlSrSrstSuSuP02P,LOc25UUP-)1u.4Q777 $ 6q1u =  055)$#+ 6  M 77 FF ( >B^^-'':(C >L> > :: ;%  "uy11I !i$.2Z\]1^fg h h h  1 6 6uz B B   !i$.2Z\]1^fj k k k  ,,-HQW,XXX#'#9#C #|2>$0 #/.. -vu =  %     "   d//0HIIJJ  DAqE`!uO+'>?Q?WXY?Z`a'>'b'b$!16E6K6KL^LdefLgmn6K6o6o3!3.1G?]nKn1oo "Y 4F'GQOOO 1u06q999 $ 8Q ? #499AmW`: O*//A66%  ../LU[.\\\%)%=%G"'.:a?%,8$+7 00 65 1       $    ;%  ];fe_MMMNN$& !g/ >>III59%k2]#)5/ QVZoNd//0JKKLL  DAq.4Q7!;;;*+!&,=~+NTU!V!V!V )  1u28;;; $ :1q5 A !% 8 = =A0 Yb!>!!  " ##%%% c! Aq!! ##Aq!Q//5577==?? %  %%e,,E 8O"%0000r*)NN)NrririrjNNNNNNrkrlrmT) __name__ __module__ __qualname____doc___exclude_from_cpu_offloadr __annotations__r rrr rrmodel_cpu_offload_seqr%r5rrr rrJTensorrhno_gradrrr GeneratorrrIr __classcell__)r(s@r)rr"s<")   !!!!""""----    $$$$&&&&((((_  & 2  !  '  % $ ) + -      8   "JN6:Y=Y= $E*=u*D$EF Y=&el3Y=Y=Y=Y=vU]__37%&)++--.MQ042648IM6:&)(+%* !Y1Y1sDI~./Y1 #Y1$' Y1 &) Y1 (+ Y1E%/43H"HIJY1  -Y1"%,/Y1$EL1Y1$E*=u*D$EFY1&el3Y1$Y1!&Y1c]Y1 !Y1Y1Y1_Y1Y1Y1Y1Y1r*r)"rtypingrrrrrJtorch.nnrr transformersrr &transformers.models.clip.modeling_clipr modelsr r r schedulersrutilsrutils.torch_utilsrpipeline_utilsrrrr get_loggerrrMrr*r)rsJ//////////// $$$$$$CCCCCCCCFFFFFFIIIIIIIIII))))))------CCCCCCCC******  H % %K1K1K1K1K1&K1K1K1K1K1r*