bi8 ddlZddlZddlZddlmZddlZddlmZm Z m Z m Z m Z m Z mZddlZddlmZmZmZddlmZmZddlmZddlmZddlmZmZdd lmZdd l m!Z!m"Z"m#Z#m$Z$m%Z%m&Z&m'Z'm(Z(m)Z)dd l*m+Z+m,Z,m-Z-d d l.m/Z/d dl0m1Z1ddl2m3Z3e%r ddl4m5cm6Z7dZ8ndZ8e&jre:Z;e#rddlZ>dZ? dde e@de eeAejfde e e@de e eCfdZDGdde/eZEy)N)AnyCallableDictListOptionalTupleUnion)Gemma2PreTrainedModelGemmaTokenizerGemmaTokenizerFast)MultiPipelineCallbacksPipelineCallback)PixArtImageProcessor)SanaLoraLoaderMixin) AutoencoderDCSanaTransformer2DModel)DPMSolverMultistepScheduler) BACKENDS_MAPPINGUSE_PEFT_BACKENDis_bs4_availableis_ftfy_availableis_torch_xla_availableloggingreplace_example_docstringscale_lora_layersunscale_lora_layers) get_deviceis_torch_version randn_tensor)DiffusionPipeline)ASPECT_RATIO_1024_BIN)SanaPipelineOutputTF) BeautifulSoupa Examples: ```py >>> import torch >>> from diffusers import SanaSprintPipeline >>> pipe = SanaSprintPipeline.from_pretrained( ... "Efficient-Large-Model/Sana_Sprint_1.6B_1024px_diffusers", torch_dtype=torch.bfloat16 ... ) >>> pipe.to("cuda") >>> image = pipe(prompt="a tiny astronaut hatching from an egg on the moon")[0] >>> image[0].save("output.png") ``` num_inference_stepsdevice timestepssigmasc | | td|dttj|jj j v}|std|jd|jd ||d||j}t|}||fS|dttj|jj j v}|std|jd|jd ||d||j}t|}||fS|j|fd |i||j}||fS) a Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`. Args: scheduler (`SchedulerMixin`): The scheduler to get timesteps from. num_inference_steps (`int`): The number of diffusion steps used when generating samples with a pre-trained model. If used, `timesteps` must be `None`. device (`str` or `torch.device`, *optional*): The device to which the timesteps should be moved to. If `None`, the timesteps are not moved. timesteps (`List[int]`, *optional*): Custom timesteps used to override the timestep spacing strategy of the scheduler. If `timesteps` is passed, `num_inference_steps` and `sigmas` must be `None`. sigmas (`List[float]`, *optional*): Custom sigmas used to override the timestep spacing strategy of the scheduler. If `sigmas` is passed, `num_inference_steps` and `timesteps` must be `None`. Returns: `Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the second element is the number of inference steps. zYOnly one of `timesteps` or `sigmas` can be passed. Please choose one to set custom valuesr)zThe current scheduler class zx's `set_timesteps` does not support custom timestep schedules. Please check whether you are using the correct scheduler.)r)r(r*zv's `set_timesteps` does not support custom sigmas schedules. Please check whether you are using the correct scheduler.)r*r(r() ValueErrorsetinspect signature set_timesteps parameterskeys __class__r)len) schedulerr'r(r)r*kwargsaccepts_timesteps accept_sigmass h/home/cdr/jupyterlab/.venv/lib/python3.12/site-packages/diffusers/pipelines/sana/pipeline_sana_sprint.pyretrieve_timestepsr;Qs>!3tuu'3w/@/@AXAX/Y/d/d/i/i/k+ll .y/B/B.CDab  M)FMfM'' !)n ) ))   C(9(9):Q:Q(R(](](b(b(d$ee .y/B/B.CD_`  GvfGG'' !)n ) ))  3MFMfM'' ) ))c5eZdZdZej dZdZddgZde e e fde de d ed ef fd Zd Zd ZdZdZ dBde eeefdej0dej2dededeeef dZ dCde eeefdedeej0deej<deej<dededeeedeefdZ dZ! dDd Z"dEd!Z#d"Z$dFd#Z%e&d$Z'e&d%Z(e&d&Z)e&d'Z*ejVe,e-dd(dd)d*d+dd,d,d-ddddd.d/dd/dddgdgd0fde eeefd1ed2eed3ed4ed5edeed6ed7ed8ed9ee ej\eej\fdeej<deej<deej<d:eed;ededee1eee/gdfd?eededeed@e e2e3ff0dAZ4xZ5S)GSanaSprintPipelinezn Pipeline for text-to-image generation using [SANA-Sprint](https://huggingface.co/papers/2503.09641). u5[#®•©™&@·º½¾¿¡§~\)\(\]\[\}\{\|\\/\*]{1,}ztext_encoder->transformer->vaelatents prompt_embeds tokenizer text_encodervae transformerr6c t||j|||||t|dr;|j/dt |jj jdz znd|_t|j|_ y)N)rArBrCrDr6rCr!r$ )vae_scale_factor) super__init__register_moduleshasattrrCr5configencoder_block_out_channelsrGrimage_processor)selfrArBrCrDr6r4s r:rIzSanaSprintPipeline.__init__s  lQ\hq  tU#(< #dhhoo@@AAE F  4TEZEZ[r<c8|jjy)z Enable sliced VAE decoding. When this option is enabled, the VAE will split the input tensor in slices to compute decoding in several steps. This is useful to save some memory and allow larger batch sizes. N)rCenable_slicingrOs r:enable_vae_slicingz%SanaSprintPipeline.enable_vae_slicing !r<c8|jjy)z Disable sliced VAE decoding. If `enable_vae_slicing` was previously enabled, this method will go back to computing decoding in one step. N)rCdisable_slicingrRs r:disable_vae_slicingz&SanaSprintPipeline.disable_vae_slicings   "r<c8|jjy)a Enable tiled VAE decoding. When this option is enabled, the VAE will split the input tensor into tiles to compute decoding and encoding in several steps. This is useful for saving a large amount of memory and to allow processing larger images. N)rC enable_tilingrRs r:enable_vae_tilingz$SanaSprintPipeline.enable_vae_tilings  r<c8|jjy)z Disable tiled VAE decoding. If `enable_vae_tiling` was previously enabled, this method will go back to computing decoding in one step. N)rCdisable_tilingrRs r:disable_vae_tilingz%SanaSprintPipeline.disable_vae_tilingrTr<F,Npromptr(dtype clean_captionmax_sequence_lengthcomplex_human_instructionc8t|tr|gn|}t|ddd|j_|j ||}|s|}nOdj |}|D cgc]} || z }} t|jj|} | |zdz }|j|d|ddd } | j} | j} | j|} |j| j|| }|d j|| }|| fScc} w)a Encodes the prompt into text encoder hidden states. Args: prompt (`str` or `List[str]`, *optional*): prompt to be encoded device: (`torch.device`, *optional*): torch device to place the resulting embeddings on clean_caption (`bool`, defaults to `False`): If `True`, the function will preprocess and clean the provided caption before encoding. max_sequence_length (`int`, defaults to 300): Maximum sequence length to use for the prompt. complex_human_instruction (`list[str]`, defaults to `complex_human_instruction`): If `complex_human_instruction` is not empty, the function will use the complex Human instruction for the prompt. rANright)ra r! max_lengthTpt)paddingrg truncationadd_special_tokensreturn_tensors)attention_maskr)r`r() isinstancestrgetattrrA padding_side_text_preprocessingjoinr5encode input_idsrmtorB)rOr_r(r`rarbrcmax_length_all chi_promptpnum_chi_prompt_tokens text_inputstext_input_idsprompt_attention_maskr@s r:_get_gemma_prompt_embedsz+SanaSprintPipeline._get_gemma_prompt_embedss;0(4&& 4d + 7*1DNN '))& )N)0N#<=J.45j1n5F5$'(=(=j(I$J !25HH1LNnn  %# % %.. + : : 5 8 8 @)).*;*;F*CTi)j %a(++%+G 333)6s! Dr$num_images_per_promptr} lora_scalec | |j}|j|jj} nd} | ?t|tr/| |_|jt rt|j| t|ddd|j_ |} dgtt| dzdz} |,|j||| |||\}}|dd| f}|dd| f}|j\} }}|jd|d}|j!| |z|d}|j!| d}|j|d}|j,t|trt rt#|j| ||fS)a Encodes the prompt into text encoder hidden states. Args: prompt (`str` or `List[str]`, *optional*): prompt to be encoded num_images_per_prompt (`int`, *optional*, defaults to 1): number of images that should be generated per prompt device: (`torch.device`, *optional*): torch device to place the resulting embeddings on 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. clean_caption (`bool`, defaults to `False`): If `True`, the function will preprocess and clean the provided caption before encoding. max_sequence_length (`int`, defaults to 300): Maximum sequence length to use for the prompt. complex_human_instruction (`list[str]`, defaults to `complex_human_instruction`): If `complex_human_instruction` is not empty, the function will use the complex Human instruction for the prompt. NrArerr$)r_r(r`rarbrc)_execution_devicerBr`rnr _lora_scalerrrprArqlistranger~shaperepeatviewr)rOr_rr(r@r}rarbrcrr`rg select_indexbs_embedseq_len_s r: encode_promptz SanaSprintPipeline.encode_promptsD >++F    (%%++EE  !j7J&K)D   ,1A!$"3"3Z@ 4d + 7*1DNN ') sT% a";<<  373P3P+$7*C 4Q4 0M0*!\/:M$9!\/$J !,22'1%,,Q0EqI %**86K+KWVXY 5 : :8R H 5 < <=RTU V    ($ 349I#D$5$5zB333r<cVdttj|jjj j v}i}|r||d<dttj|jjj j v}|r||d<|S)Neta generator)r.r/r0r6stepr2r3)rOrr accepts_etaextra_step_kwargsaccepts_generators r:prepare_extra_step_kwargsz,SanaSprintPipeline.prepare_extra_step_kwargs[s s7#4#4T^^5H5H#I#T#T#Y#Y#[\\  '* e $(3w/@/@ATAT/U/`/`/e/e/g+hh -6 k *  r<c h|dzdk7s|dzdk7rtd|d|d|Ltfd|Ds8tdjd|D cgc]} | jvs| c} || td |d | d | | td |7t|ts't|t std t || | td|t||dzk7r td| | td| | td||dk7r tdyycc} w)NrFrz8`height` and `width` have to be divisible by 32 but are z and .c3:K|]}|jvywN)_callback_tensor_inputs).0krOs r: z2SanaSprintPipeline.check_inputs..|s#F 23A-- -F sz2`callback_on_step_end_tensor_inputs` has to be in z , but found zCannot forward both `prompt`: z and `prompt_embeds`: z2. Please make sure to only forward one of the two.zeProvide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined.z2`prompt` has to be of type `str` or `list` but is zEMust provide `prompt_attention_mask` when specifying `prompt_embeds`.r$zWIf providing custom timesteps, `timesteps` must be of length `num_inference_steps + 1`.zFIf providing custom timesteps, `max_timesteps` should not be provided.z5Should provide either `timesteps` or `max_timesteps`.r!zNIntermediate timesteps for SCM is not supported when num_inference_steps != 2.)r-allrrnrortyper5) rOr_heightwidthr'r) max_timestepsintermediate_timesteps"callback_on_step_end_tensor_inputsr@r}rs ` r: check_inputszSanaSprintPipeline.check_inputsls B;! urzQWX^W__dejdkklmn n - 9#F 7YF C DTEaEaDbbn|^pHvwbckokGkGbGpqpHoIJ   -";08N}o^00 ^ 5w  FC)@TZ\`IaQRVW]R^Q_`a a  $)>)Fde e  S^7JQ7N%Nvw w  ]%>ef f  !6TU U ! -2E2Jmn n3K -7pHs D/%D/crOtsEtjtddj dtjddrOt sEtjtddj dtjddt |ttfs|g}dtffd }|Dcgc] }|| c}Scc}w) Nbs4rzSetting `clean_caption=True`z#Setting `clean_caption` to False...Fftfytextcr$j|}j|}|S|jj}|Sr)_clean_captionlowerstrip)rrarOs r:processz7SanaSprintPipeline._text_preprocessing..processsH**40**40Kzz|))+Kr<) rloggerwarningrformatrrntuplerro)rOrrarts` ` r:rrz&SanaSprintPipeline._text_preprocessings !1!3 NN+E226==>\] ^ NN@ A!M !2!4 NN+F3B7>>?]^ _ NN@ A!M$ .6D # %))q )))sC!c t|}tj|}|jj }t j dd|}t j dd|}t j dd|}t|dj}t j dd|}t j d d|}t j d d|}t j d d|}t j d d|}t j d d|}t j dd|}t j dd|}t j dd|}t j dd|}t j dd|}t j dd|}t j dd|}t j dd|}t j dd|}t j dd|}t j dd|}t j dd|}t j dd|}t j dd|}t j d d|}t j d!d|}t j |jd|}t j d"d|}t jd#}tt j||d$kDrt j |d|}tj|}tj tj |}t j d%d|}t j d&d|}t j d'd|}t j d(d|}t j d)d|}t j d*d|}t j d+d|}t j d,d|}t j d-d|}t j d.d|}t j d/d0|}t j d1d2|}t j d3d|}|jt j d4d5|}t j d6d|}t j d7d|}t j d8d|}|jS)9Nzpersonzk\b((?:https?:(?:\/{1,3}|[a-zA-Z0-9%])|[a-zA-Z0-9.\-]+[.](?:com|co|ru|net|org|edu|gov|it)[\w/-]*\b\/?(?!@)))zh\b((?:www:(?:\/{1,3}|[a-zA-Z0-9%])|[a-zA-Z0-9.\-]+[.](?:com|co|ru|net|org|edu|gov|it)[\w/-]*\b\/?(?!@)))z html.parser)featuresz @[\w\d]+\bz[\u31c0-\u31ef]+z[\u31f0-\u31ff]+z[\u3200-\u32ff]+z[\u3300-\u33ff]+z[\u3400-\u4dbf]+z[\u4dc0-\u4dff]+z[\u4e00-\u9fff]+z|[\u002D\u058A\u05BE\u1400\u1806\u2010-\u2015\u2E17\u2E1A\u2E3A\u2E3B\u2E40\u301C\u3030\u30A0\uFE31\uFE32\uFE58\uFE63\uFF0D]+-u[`´«»“”¨]"u[‘’]'z"?z&z"\d{1,3}\.\d{1,3}\.\d{1,3}\.\d{1,3} z \d:\d\d\s+$z\\nz #\d{1,3}\bz #\d{5,}\bz \b\d{6,}\bz0[\S]+\.(?:png|jpg|jpeg|bmp|webp|eps|pdf|apk|mp4)z [\"\']{2,}z[\.]{2,}z\s+\.\s+z (?:\-|\_)r z\b[a-zA-Z]{1,3}\d{3,15}\bz\b[a-zA-Z]+\d+[a-zA-Z]+\bz\b\d+[a-zA-Z]+\d+\bz!(worldwide\s+)?(free\s+)?shippingz(free\s)?download(\sfree)?z\bclick\b\s(?:for|on)\s\w+z9\b(?:png|jpg|jpeg|bmp|webp|eps|pdf|apk|mp4)(\simage[s]?)?z\bpage\s+\d+\bz*\b\d*[a-zA-Z]+\d+[a-zA-Z]+\d+[a-zA-Z\d]*\bu\b\d+\.?\d*[xх×]\d+\.?\d*\bz \b\s+\:\s+z: z (\D[,\./])\bz\1 z\s+z^[\"\']([\w\W]+)[\"\']$z\1z ^[\'\_,\-\:;]z[\'\_,\-\:\-\+]$z^\.\S+$)roul unquote_plusrrresubr&rbad_punct_regexcompiler5findallrfix_texthtmlunescape)rOcaptionregex2s r:rz!SanaSprintPipeline._clean_captionsg,//'*--/'')&&Xw7&& z    && w     -@EE&&G4&&,b':&&,b':&&,b':&&,b':&&,b':&&,b':&&,b':&& L   &&-sG<&&c73&&R1&&"g.&&>WM&&W5&&g.&&G4&&r73&&G4&&LbRYZ&&g6&&dG4&&--tW=&&dG4L) rzz&'* +a /ffVS'2G--(-- g 67&&5r7C&&5r7C&&/W=&&=r7K&&6GD&&6GD&&UWY[bc&&*B8&&FgV&&92wG&&w7&&&':&&g. &&3UGD&&)38&&,c7;&&R1}}r<c $||j||S||t||jzt||jzf} t|tr)t ||k7rt dt |d|dt| |||}|S)Nr(r`z/You have passed a list of generators of length z+, but requested an effective batch size of z@. Make sure the batch size matches the length of the generators.)rr(r`)rvintrGrnrr5r-r ) rO batch_sizenum_channels_latentsrrr`r(rr?rs r:prepare_latentsz"SanaSprintPipeline.prepare_latents+s  ::V5:9 9  K400 0 J$// /   i &3y>Z+GA#i.AQR&<'gi  u &PUVr<c|jSr)_guidance_scalerRs r:guidance_scalez!SanaSprintPipeline.guidance_scale>s###r<c|jSr)_attention_kwargsrRs r:attention_kwargsz#SanaSprintPipeline.attention_kwargsBs%%%r<c|jSr)_num_timestepsrRs r: num_timestepsz SanaSprintPipeline.num_timestepsFs"""r<c|jSr) _interruptrRs r: interruptzSanaSprintPipeline.interruptJs r<r!gH.!?g?g@igpilT)zGiven a user prompt, generate an 'Enhanced prompt' that provides detailed visual descriptions suitable for image generation. Evaluate the level of detail in the user prompt:z- If the prompt is simple, focus on adding specifics about colors, shapes, sizes, textures, and spatial relationships to create vivid and concrete scenes.zo- If the prompt is already detailed, refine and enhance the existing details slightly without overcomplicating.z8Here are examples of how to transform or refine prompts:z- User Prompt: A cat sleeping -> Enhanced: A small, fluffy white cat curled up in a round shape, sleeping peacefully on a warm sunny windowsill, surrounded by pots of blooming red flowers.z- User Prompt: A busy city street -> Enhanced: A bustling city street scene at dusk, featuring glowing street lamps, a diverse crowd of people in colorful clothing, and a double-decker bus passing by towering glass skyscrapers.zPlease generate only the enhanced description for the prompt below and avoid including any additional commentary or evaluations:z User Prompt: r'r)rrrrrrr output_type return_dictuse_resolution_binningrcallback_on_step_endrreturnc  t|ttfr |j}|rZ|jj j dk(rt}n td|| }}|jj|| |\}} |j||| |||||| | ||_ ||_ d|_|t|trd}n-|t|t r t#|}n| j$d}|j&}|j(|j(j+d dnd}|j-|||| ||||| \} }t/|j0|||d|| \}}t3|j0d r|j0j5d|jj j6}|j9||z||| t:j<|| | } | |j0j j>z} t;j@dg||t:j< }|jC| j$djE| jF}||jj jHz}|jK| | } |dd}tMt#|||j0jNzz d}!t#||_(|jjF}"|jS|5}#tU|D]Y\}$}%|jVr|%jC| j$d}&| |j0j j>z }'t;jX|&t;jZ|&t;jX|&zz }(|(j]dddd})|'t;j^|)dzd|)z dzzz}*|j |*jE|"| jE|"|||(d|j(d}+dd|)zz |*zdd|)zz d|)dzzz|+zzt;j^|)dzd|)z dzzz }+|+ja|j0j j>z}+|j0jb|+|&| fi| ddi\} },|Hi}-|D]}.te|.|-|.<|||$|%|-}/|/jgd| } |/jgd| } |$t#|dz k(s'|$dz|!kDr/|$dz|j0jNzdk(r|#jitjsFtmjn\ ddd,|j0j j>z } |dk(r| }0n| jE|jpjF} tst:tut:jv}1tyddrt:jzn |1jz}2 |jpj}| |jpj j~z dd}0|r|jj0}0|dk(s|jj0|}0|j|s0fSt0S#1swYMxYw#|2$r"}3tj|3dYd}3~3d}3~3wwxYw)u Function invoked when calling the pipeline for generation. Args: prompt (`str` or `List[str]`, *optional*): The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`. instead. num_inference_steps (`int`, *optional*, defaults to 20): The number of denoising steps. More denoising steps usually lead to a higher quality image at the expense of slower inference. max_timesteps (`float`, *optional*, defaults to 1.57080): The maximum timestep value used in the SCM scheduler. intermediate_timesteps (`float`, *optional*, defaults to 1.3): The intermediate timestep value used in SCM scheduler (only used when num_inference_steps=2). timesteps (`List[int]`, *optional*): Custom timesteps to use for the denoising process with schedulers which support a `timesteps` argument in their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed will be used. Must be in descending order. guidance_scale (`float`, *optional*, defaults to 4.5): Embedded guiddance scale is enabled by setting `guidance_scale` > 1. Higher `guidance_scale` encourages a model to generate images more aligned with `prompt` at the expense of lower image quality. Guidance-distilled models approximates true classifer-free guidance for `guidance_scale` > 1. Refer to the [paper](https://huggingface.co/papers/2210.03142) to learn more. num_images_per_prompt (`int`, *optional*, defaults to 1): The number of images to generate per prompt. height (`int`, *optional*, defaults to self.unet.config.sample_size): The height in pixels of the generated image. width (`int`, *optional*, defaults to self.unet.config.sample_size): The width in pixels of the generated image. eta (`float`, *optional*, defaults to 0.0): Corresponds to parameter eta (η) in the DDIM paper: https://huggingface.co/papers/2010.02502. Only applies to [`schedulers.DDIMScheduler`], will be ignored for others. generator (`torch.Generator` or `List[torch.Generator]`, *optional*): One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make generation deterministic. latents (`torch.Tensor`, *optional*): Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image generation. Can be used to tweak the same generation with different prompts. If not provided, a latents tensor will ge generated by sampling using the supplied random `generator`. 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. prompt_attention_mask (`torch.Tensor`, *optional*): Pre-generated attention mask for text embeddings. output_type (`str`, *optional*, defaults to `"pil"`): The output format of the generate image. Choose between [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`. return_dict (`bool`, *optional*, defaults to `True`): Whether or not to return a [`~pipelines.stable_diffusion.IFPipelineOutput`] instead of a plain tuple. attention_kwargs: A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under `self.processor` in [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py). clean_caption (`bool`, *optional*, defaults to `True`): Whether or not to clean the caption before creating embeddings. Requires `beautifulsoup4` and `ftfy` to be installed. If the dependencies are not installed, the embeddings will be created from the raw prompt. use_resolution_binning (`bool` defaults to `True`): If set to `True`, the requested height and width are first mapped to the closest resolutions using `ASPECT_RATIO_1024_BIN`. After the produced latents are decoded into images, they are resized back to the requested resolution. Useful for generating non-square images. callback_on_step_end (`Callable`, *optional*): A function that calls at the end of each denoising steps during the inference. The function is called with the following arguments: `callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict)`. `callback_kwargs` will include a list of all tensors as specified by `callback_on_step_end_tensor_inputs`. callback_on_step_end_tensor_inputs (`List`, *optional*): The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the `._callback_tensor_inputs` attribute of your pipeline class. max_sequence_length (`int` defaults to `300`): Maximum sequence length to use with the `prompt`. complex_human_instruction (`List[str]`, *optional*): Instructions for complex human attention: https://github.com/NVlabs/Sana/blob/main/configs/sana_app_config/Sana_1600M_app.yaml#L55. Examples: Returns: [`~pipelines.sana.pipeline_output.SanaPipelineOutput`] or `tuple`: If `return_dict` is `True`, [`~pipelines.sana.pipeline_output.SanaPipelineOutput`] is returned, otherwise a `tuple` is returned where the first element is a list with the generated images rFzInvalid sample size)ratios) r_rrr'r)rrrr@r}FNr$rscale)rr(r@r}rarbrcr)r*rrset_begin_indexrr)totalr!)r`)encoder_hidden_statesencoder_attention_maskguidancetimesteprrrr?r@latentz>=z2.5.0)rz. Try to use VAE tiling for large images. For example: pipe.vae.enable_tiling(tile_sample_min_width=512, tile_sample_min_height=512))r)images)Frnrr tensor_inputsrDrL sample_sizer#r-rNclassify_height_width_binrrrrrorr5rrrgetrr;r6rKr in_channelsrtorchfloat32 sigma_datafullexpandrvr`guidance_embeds_scalermaxorderr progress_bar enumeratersincosrsqrtfloatrlocalspopupdate XLA_AVAILABLExm mark_steprCrprcudarOutOfMemoryErrordecodescaling_factorwarningswarnresize_and_crop_tensor postprocessmaybe_free_model_hooksr%)4rOr_r'r)rrrrrrrrr?r@r}rrrarrrrrbrcaspect_ratio_bin orig_height orig_widthrr(rlatent_channelsrrnum_warmup_stepstransformer_dtyperirrlatents_model_input scm_timestepscm_timestep_expandedlatent_model_input noise_preddenoisedcallback_kwargsrcallback_outputsimagetorch_accelerator_module oom_errores4 r:__call__zSanaSprintPipeline.__call__Nsr *-=?U,V W1E1S1S . "&&22b8#8  !677&,eK 00JJ6SXaqJrMFE  3'#9/Q'"7   .!1  *VS"9J  Jvt$<VJ&,,Q/J''AEAVAVAbT**..w=hl    "7'"7' 3&?!    !*< NN   '#9* & & 4>>#4 5 NN * *1 -**11==&& . .    MM     DNN11<<<::qc>& V??7==#3477 8K8KLd..55KKK!::9cJcrN s9~0CdnnFZFZ0ZZ\]^!)n ,,22   %8  94 #\!),3 #1>>88GMM!$45&-0E0E0P0P&P#$yy2eii6IEIIV^L_6_` (4(9(9"aA(F%%85::)1,4I0Ia/OO<&" "--&))0A)B*7*:*:AR*:*S+@%) %%)%:%:. 2226HH1444q;PRS;S7SSWaabJJ4a71?T;TYZ:ZZ[\ (--/$..2G2G2R2RR %8DNN$7$7'%5F%TY%!(3&(O?9-3Xa[*9';D!Q'X$.229gFG$4$8$8-$XMI**A9I/IqSTuX\XfXfXlXlNlpqNq '') LLNg3 #4 #lT^^22=== ( "Ejj0G'.ujlEJJ'O $$D'2&&->>   $((//2P2P(P^cdefg&,,CCE:Wbch&((44U 4TE ##%8O!//k4 #4 #F  cde s,IZ5Z5A[5Z?[)[$$[))Fr^N)r$NNNFr^NN)NNN)Fr)6__name__ __module__ __qualname____doc__rrrmodel_cpu_offload_seqrr r r r rrrrIrSrWrZr]rorrr(r`boolrrr~Tensorrrrrrrrrpropertyrrrrno_gradrEXAMPLE_DOC_STRING Generatorrrrr%rr! __classcell__)r4s@r:r>r>s+ !bjj"abO=(/:\);;<\,\ \ , \ / \*"#!"$#&9=84c49n%84 84{{ 84  84 ! 84$,DI#684z&')-048<##&9=&*S4c49n%S4 #S4& S4   - S4 ( 5 S4S4!S4$,DI#6S4UOS4l!4,0"0of*2pf&$$&&##U]]_12)-#$#&(+ #/0MQ*.048<%* #'+59KO9B #& 0 1m0c49n%m0!m09 m0  m0 !& m0m0 (}m0m0m0m0E%//43H"HIJm0%,,'m0  -m0 ( 5m0 c]!m0"#m0$%m0&!%'m0(#4S>2)m0*'xc40@$0F'GH+m0,-1I-m0.!/m00$(91m0D !5( )Em03m0r<r>)NNNN)Frr/r urllib.parseparserr typingrrrrrrr r transformersr r r callbacksrrrNrloadersrmodelsrr schedulersrutilsrrrrrrrrrutils.torch_utilsrrr pipeline_utilsr""pixart_alpha.pipeline_pixart_alphar#pipeline_outputr%torch_xla.core.xla_modelcore xla_modelrr get_loggerr"rrr&rr+rror(rr;r>r,r<r:r?s  DDD RRA3*;5   LK.F/))MM   H %!(*.15%)$( 8*!#8* U3 ,- .8*S " 8* T%[ ! 8*vq 0*,?q 0r<