bi(ddlmZmZmZmZmZddlZddlmZm Z ddl m Z ddl m Z ddlmZdd lmZmZe r ddlmcmZd Znd ZGd d eZy))DictListOptionalTupleUnionN) AutoencoderKLDiTTransformer2DModel)KarrasDiffusionSchedulers)is_torch_xla_available) randn_tensor)DiffusionPipelineImagePipelineOutputTFceZdZdZdZ ddedededee e e fffd Z de e ee fd ee fd Zej" dd ee d ed ee ej&eej&fde dee ded e eeffdZxZS) DiTPipelinea Pipeline for image generation based on a Transformer backbone instead of a UNet. 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.). Parameters: transformer ([`DiTTransformer2DModel`]): A class conditioned `DiTTransformer2DModel` to denoise the encoded image latents. Initially published as [`Transformer2DModel`](https://huggingface.co/facebook/DiT-XL-2-256/blob/main/transformer/config.json#L2) in the config, but the mismatch can be ignored. vae ([`AutoencoderKL`]): Variational Auto-Encoder (VAE) model to encode and decode images to and from latent representations. scheduler ([`DDIMScheduler`]): A scheduler to be used in combination with `transformer` to denoise the encoded image latents. ztransformer->vae transformervae schedulerid2labelct||j|||i|_||j D]O\}}|j dD]6}t ||j|jj<8Qtt|jj |_yy)N)rrr,) super__init__register_moduleslabelsitemssplitintlstriprstripdictsorted) selfrrrrkeyvaluelabel __class__s _/home/cdr/jupyterlab/.venv/lib/python3.12/site-packages/diffusers/pipelines/dit/pipeline_dit.pyrzDiTPipeline.__init__<s  +3)T  &nn. D U"[[-DE;>s8DKK  5 5 78D Dvdkk&7&7&9:;DK r'returnct|ts t|}|D]+}||jvst|d|jd|Dcgc]}|j|c}Scc}w)a0 Map label strings from ImageNet to corresponding class ids. Parameters: label (`str` or `dict` of `str`): Label strings to be mapped to class ids. Returns: `list` of `int`: Class ids to be processed by pipeline. zK does not exist. Please make sure to select one of the following labels: .) isinstancelistr ValueError)r$r'ls r) get_label_idszDiTPipeline.get_label_idsNsw%&KE A # cefjfqfqerrst  )..1 A...sA) class_labelsguidance_scale generatornum_inference_steps output_type return_dictc t|}|jjj}|jjj} t || ||f||j |jj} |dkDrtj| gdzn| } tj||j jd}tjdg|z|j } |dkDrtj|| gdn|} |jj||j|jjD]}|dkDr*| dt| dz}tj||gd } |jj!| |} |}tj"|s| j$j&d k(}| j$j&d k(}t)|t*r%|s|rtj,ntj.}n$|s|rtj0ntj2}tj|g|| j$ }n6t|j4dk(r|dj7| j$}|j9| j4d}|j| || j:}|dkDrz|ddd| f|dd| df}}tj<|t|dzd \}}||||z zz}tj||gd }tj||gd }|jjj>dz| k(rtj<|| d \}}n|}|jjA||| jB} tDsytGjH|dkDr| jKdd \} }n| } d|jLjjNz | z} |jLjQ| j:}|dz dzjSdd}|jUjWddddj+jY}|dk(r|j[|}|j]|s|fSt_|S)a> The call function to the pipeline for generation. Args: class_labels (List[int]): List of ImageNet class labels for the images to be generated. 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`. generator (`torch.Generator`, *optional*): A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make generation deterministic. num_inference_steps (`int`, *optional*, defaults to 250): The number of denoising steps. More denoising steps usually lead to a higher quality image at the expense of slower inference. 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 [`ImagePipelineOutput`] instead of a plain tuple. Examples: ```py >>> from diffusers import DiTPipeline, DPMSolverMultistepScheduler >>> import torch >>> pipe = DiTPipeline.from_pretrained("facebook/DiT-XL-2-256", torch_dtype=torch.float16) >>> pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config) >>> pipe = pipe.to("cuda") >>> # pick words from Imagenet class labels >>> pipe.labels # to print all available words >>> # pick words that exist in ImageNet >>> words = ["white shark", "umbrella"] >>> class_ids = pipe.get_label_ids(words) >>> generator = torch.manual_seed(33) >>> output = pipe(class_labels=class_ids, num_inference_steps=25, generator=generator) >>> image = output.images[0] # label 'white shark' ``` 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 )shaper5devicedtyper)r;irN)dimmpsnpu)r<r;)timestepr3g?rpil)images)0lenrconfig sample_size in_channelsr _execution_devicer<torchcattensorreshaper set_timesteps progress_bar timestepsscale_model_input is_tensorr;typer.floatfloat32float64int32int64r:toexpandsampler out_channelsstep prev_sample XLA_AVAILABLExm mark_stepchunkrscaling_factordecodeclampcpupermutenumpy numpy_to_pilmaybe_free_model_hooksr)r$r3r4r5r6r7r8 batch_size latent_sizelatent_channelslatentslatent_model_input class_nullclass_labels_inputthalfrPis_mpsis_npur< noise_predepsrestcond_eps uncond_epshalf_eps model_output_sampless r)__call__zDiTPipeline.__call__gsJx& &&--99 **11== [I))""((   :H!9KUYYy1}5QX||L9O9OPXXY[\ \\4&:"5d>T>TU IWZ[I[UYY j'A1Eam $$%89""4>>#;#;<. A!)*HC0B,Cq,HI%*YYd|%C"!%!A!ABTVW!X I??9-,22775@+22775@i/.4EMMU]]E,2fEKK5;;E!LL)EJ\JcJcd Y__%*%dO../A/H/HI !(();)A)A!)DEI))"YEW*f  !&q*:?*:':;Z?K[H[=\T',{{3CA 1'M$*%(Z:O(PPii8 4!<"YYT{: &&33q8OK"'++j/q"Q a) "&!4!4\1FX!Y!e!e  ]. ` A +11!1;JGQ(Gdhhoo444w>((//'*11Q;$++Aq1++-''1a399;AAC % ''0G ##%: "'22r*)N)g@N2rCT)__name__ __module__ __qualname____doc__model_cpu_offload_seqr r r rrrstrrrrr2rJno_gradrT Generatorboolrrr __classcell__)r(s@r)rr(s"/.2 <*<<- < 4S>* <$/5d3i#8/T#Y/2U]]_!$MQ#%%* S33iS3S3E%//43H"HIJ S3 ! S3 c] S3S3 "E) *S3S3r*r)typingrrrrrrJmodelsr r schedulersr utilsr utils.torch_utilsr pipeline_utilsrrtorch_xla.core.xla_modelcore xla_modelr`r_rr*r)rsF*65 :3+-C))MMS3#S3r*