iddlZddlZddlZddlmZddlmZmZm Z m Z m Z ddl Z ddl Z ddlmZmZmZddlmZddlmZddlmZmZmZddlmZmZddlmZdd lm Z dd l!m"Z"dd l#m$Z$dd l%m&Z&m'Z'dd l(m)Z)dZ*ddgddgddgddgddgddgddgdZ+ddgddgddgddgddgdd gddgd dgddgddgddgddgddgd! Z,id"dd#gd$d%d&gd'd%d(gd)d*d+gd,d*d-gd.d/d0gd1d2d3gd4d2d5gd6d7d8gd9d:d;gdgd?d@d@gdAd>d=gdBdCdDgdEd;d:gdFd8d7gdGdHdIgd5d2gd3d2gd-d*gd+d*gd(d%gd&d%gd#dgdJZ-e+e,e-dKZ.e&j^e0Z1GdLdMZ2dNZ3GdOdPe eeeZ4y)QN)CallableDictListOptionalUnion) AutoTokenizerGemmaTokenizerFastT5TokenizerFast)T5GemmaEncoder)PixArtImageProcessor)FromSingleFileMixinLoraLoaderMixinTextualInversionLoaderMixin) AutoencoderDC AutoencoderKL)PRXTransformer2DModel)DiffusionPipeline)PRXPipelineOutput)FlowMatchEulerDiscreteScheduler)loggingreplace_example_docstring) randn_tensori`i@i )z0.460.60.781.01.291.67z2.2iiiiii`ii@ii ) z0.50.57r0.68rz0.88r 1.13r!1.46r"1.75z2.0z0.49iz0.52iiz0.53i`r#ii@z0.59i z0.62i iz0.67i@ir$iriiz0.83ii`z0.94ii r z1.06r%i@iz1.21r!z1.37ii`)r&z1.5z1.71r'z1.87z1.91z2.05)rrr(c&eZdZdZdZdedefdZy)TextPreprocessorz+Text preprocessing utility for PRXPipeline.c8tjd|_y)zInitialize text preprocessor.u6[#®•©™&@·º½¾¿¡§~\)\(\]\[\}\{\|\\\/\*]{1,}N)recompilebad_punct_regexselfs n/home/obispo/Crisostomo_bridge/mision_env/lib/python3.12/site-packages/diffusers/pipelines/prx/pipeline_prx.py__init__zTextPreprocessor.__init__os!zz  textreturnc t|}tj|}|jj }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 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 |jd|}t j dd|}t jd }tt j||d!kDrt j |d|}tj|}tjtj|}|j}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,d-|}t j d.d/|}t j d0d|}|jt j d1d2|}t j d3d|}t j d4d|}t j d5d|}|jS)6z5Clean text using comprehensive text processing logic.zpersonzo\b((?:https?|www):(?:\/{1,3}|[a-zA-Z0-9%])|[a-zA-Z0-9.\-]+[.](?:com|co|ru|net|org|edu|gov|it)[\w/-]*\b\/?(?!@))z @[\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 [`´«»¨]"z['']'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 (?:\-|\_)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+$)strul unquote_plusstriplowerr,subr.r-lenfindallftfyfix_texthtmlunescape)r0r4regex2s r1 clean_textzTextPreprocessor.clean_texts4yt$zz|!!#vvj(D1vv ~   vvmR.vv)2t4vv)2t4vv)2t4vv)2t4vv)2t4vv)2t4vv)2t4vv L   vv'd3vvgsD)vvj"d+vvgr4(vv;S$Gvvnb$/vvfc4(vvmR.vvlB-vvmR.vvI2tTvvmT40vvk4.vvd**D$7vvk4.L) rzz&$' (1 ,66&#t,D}}T"}}T]]401zz|vv2B=vv2B=vv,b$7vv:BEvv3R>vv3R>vvRTVX\]vv'T2vvCT4Pvv6DAvvmUD1vvovt4vvfc4( vv0%>vv&T2vv)35vvj"d+zz|r3N)__name__ __module__ __qualname____doc__r2r>rKr3r1r*r*ls 5 $gsgsgr3r*a Examples: ```py >>> import torch >>> from diffusers import PRXPipeline >>> # Load pipeline with from_pretrained >>> pipe = PRXPipeline.from_pretrained("Photoroom/prx-512-t2i-sft") >>> pipe.to("cuda") >>> prompt = "A digital painting of a rusty, vintage tram on a sandy beach" >>> image = pipe(prompt, num_inference_steps=28, guidance_scale=5.0).images[0] >>> image.save("prx_output.png") ``` c+eZdZdZdZddgZdgZdefdede d e d e e e efdee eefd eef fd Zed ZedZedZdZ d4dededededej4dej6deej8deej:fdZ d5de ee efdeej6de!dededeejDd eejDd!eejFd"eejFfd#Z$d$e edej6fd%Z% d6de edej6de!defd&Z& d7de ee efdeded'e'd(ee edeejDd eejDfd)Z(ejRe*e+ddddd*dd+dddddddd,ddddgfde ee efdedeedeed-ed.e ed'e'deedee ej8e ej8fdeej:deejDd eejDd!eejFd"eejFd/eed0e!d1e!d2ee,eee-gdfd(e ef&d3Z.xZ/S)8 PRXPipelinea+ Pipeline for text-to-image generation using PRX Transformer. This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.) Args: transformer ([`PRXTransformer2DModel`]): The PRX transformer model to denoise the encoded image latents. scheduler ([`FlowMatchEulerDiscreteScheduler`]): A scheduler to be used in combination with `transformer` to denoise the encoded image latents. text_encoder ([`T5GemmaEncoder`]): Text encoder model for encoding prompts. tokenizer ([`T5TokenizerFast` or `GemmaTokenizerFast`]): Tokenizer for the text encoder. vae ([`AutoencoderKL`] or [`AutoencoderDC`]): Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations. Supports both AutoencoderKL (8x compression) and AutoencoderDC (32x compression). ztext_encoder->transformer->vaelatents prompt_embedsvaeN transformer scheduler text_encoder tokenizerdefault_sample_sizec,t|t tdt |_||_d|_|j||||||j|j |t|j|_ yd|_ y)NzgPRXTransformer2DModel is not available. Please ensure the transformer_prx module is properly installed.?)rVrWrXrYrU)rZ)vae_scale_factor) superr2r ImportErrorr*text_preprocessorrZ_guidance_scaleregister_modulesregister_to_configr r]image_processor)r0rVrWrXrYrUrZ __class__s r1r2zPRXPipeline.__init__s  (y "2!3#6 " #%   D4L4LM ?#7I^I^#_D #'D r3c|jyt|jdr|jjSdt|jjj dz zS)Nspatial_compression_ratio)rUhasattrrhrDconfigblock_out_channelsr/s r1r]zPRXPipeline.vae_scale_factor=sQ 88  4888 98855 5TXX__??@1DE Er3c |jdkDS)zECheck if classifier-free guidance is enabled based on guidance scale.r\rar/s r1do_classifier_free_guidancez'PRXPipeline.do_classifier_free_guidanceFs##c))r3c|jSNror/s r1guidance_scalezPRXPipeline.guidance_scaleKs###r3cDt|jdd}||StS)zDetermine the default resolution based on the loaded VAE and config. Returns: int: The default sample size (height/width) to use for generation. rZN)getattrrlDEFAULT_RESOLUTION)r0default_from_configs r1get_default_resolutionz"PRXPipeline.get_default_resolutionOs* &dkk3H$O  *& &!!r3 batch_sizenum_channels_latentsheightwidthdtypedevice generatorc |-|j} || z|| z} } ||| | f} t| |||}|S|j|}|S)z2Prepare initial latents for the diffusion process.)rr~r})r]rto) r0ryrzr{r|r}r~rrSspatial_compression latent_height latent_widthshapes r1prepare_latentszPRXPipeline.prepare_latents[sl ?"&"7"7 --,,(M !5}lSE"5IfTYZGjj(Gr3Tr8rjpromptrpnegative_promptnum_images_per_promptnegative_prompt_embedsprompt_attention_masknegative_prompt_attention_maskc &| |j}|,t|tr|g}|j||||\}}}} |dkDr|j\} } } |j d|d}|j | |z| d}|$|j | d}|j |d}|ra|_|j\} } } |j d|d}|j | |z| d}| $| j | d} | j |d} |||r|nd|r| fSdfS)z\Encode text prompt using standard text encoder and tokenizer, or use precomputed embeddings.Nrj)_execution_device isinstancer>_encode_prompt_standardrrepeatview) r0rr~rprrrTrrrbs_embedseq_len_s r1 encode_promptzPRXPipeline.encode_promptss >++F  &#& ,,VV=XZij iM02HJh !1 $#0#6#6 Hgq)004I1MM)..x:O/OQXZ\]M$0(=(B(B8R(P%(=(D(DEZ\](^%+/E/Q'='C'C$'1)?)F)FqJ_ab)c&)?)D)DXPeEegnpr)s&1=5S5X5XYace5f25S5Z5Z[prs5t2  !&A "t.I *   PT   r3promptsc|Dcgc]}|jj|}}|j|d|jjddd}|dj ||dj j |fScc}w)zTokenize and clean prompts. max_lengthTpt)paddingr truncationreturn_attention_maskreturn_tensors input_idsattention_mask)r`rKrYmodel_max_lengthrbool)r0rr~r4cleanedtokenss r1_tokenize_promptszPRXPipeline._tokenize_promptssGNOt4))44T:OO  ~~66"&   k"%%f-v6F/G/L/L/N/Q/QRX/YYYPs"B c|t|}|rt|tr|g|z}||z}n|}|j||\}}t j 5|j ||dd} ddd|r- j|d\} } |j|d\} } n } |} d} d} | | | | fS#1swYFxYw)zNEncode prompt using standard text encoder and tokenizer with batch processing.T)rroutput_hidden_stateslast_hidden_stateNrdim)rDrr>rtorchno_gradrXsplit)r0rr~rprryprompts_to_encoderr embeddingsuncond_text_embeddingstext_embeddingsuncond_cross_attn_maskcross_attn_masks r1rz#PRXPipeline._encode_prompt_standards[ &/3/#2"3j"@ /& 8  & $($:$:;Lf$U! > ]]_ #**#-%)+" #J # '6@6F6FzWX6F6Y 3 "O6D6J6J:[\6J6] 3 "O(O,O%) "%) "1GI___! # #s B22B;rs"callback_on_step_end_tensor_inputsc Z||td|d|d| | td|7t|ts't|tstdt |||dkDr | tdj }||zd k7s||zd k7rtd |d |d |d |dkrtd||t|tstd||Mt fd|Ds8tdjd|D cgc]} | jvs| c} yycc} w)z,Check that all inputs are in correct format.NzCannot 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 r\zWhen `prompt_embeds` is provided and `guidance_scale > 1.0`, `negative_prompt_embeds` must also be provided for classifier-free guidance.rz-`height` and `width` have to be divisible by z but are z and .z'guidance_scale has to be >= 1.0 but is z=`callback_on_step_end_tensor_inputs` has to be a list but is c3:K|]}|jvywrr)_callback_tensor_inputs).0kr0s r1 z+PRXPipeline.check_inputs..s#F 23A-- -F sz2`callback_on_step_end_tensor_inputs` has to be in z , but found ) ValueErrorrr>listtyper]allr) r0rr{r|rsrrTrrrs ` r1 check_inputszPRXPipeline.check_inputss  -";08N}o^00  >m3w   z&#'>zRXZ^G_QRVW]R^Q_`a a  $#)=BXB`_  #33 ' '1 ,8K0Kq0P?@S?TT]^d]eejkpjqqrs  C F~FVWX X - 9*MoquBvOPrOst  . 9#F 7YF C DTEaEaDbbn|^pHvwbckokGkGbGpqpHoIJ C 9pHs D(D(g@pilnum_inference_steps timesteps output_type return_dictuse_resolution_binningcallback_on_step_endc  |j}|xs|}|xs|}|r|j td|jtvr7tdt tj d|jdt|j}||}}|jj|||\}}|j|||||| | |j|dvrtd|d |t|trd }n-|t|t r t|}n| jd }|j}||_|j!|||j"||| | | | \}}}}|} |} |?|j$j'|| |j$j(}t|}n3|j$j'|||j$j(}t||_|j!|jj,j.}n |j0j,j2}|j5||z||||j6|| | } i}dt9t;j<|j$j>j@j v} | rd|d<|j"r9tCjD||gd }!d}"| |tCjD||gd }"n|}!|}"tGt|||j$jHzz d }#|jK|5}$tM|D]\}%}&|j"rxtCjD| | gd }'|&jO|j$j,jPz jSd jUdjW|}(nQ| }'|&jO|j$j,jPz jSd jW|}(|j1|'|(|!|"dd })|j"r!|)jYdd \}*}+|*||+|*z zz})|j$j>|)|&| fi|jZ} |$i},|D]}-t]|-|,|-<|||%|&|,|%t|d z k(s+|%d z|#kDs|%d z|j$jHzd k(s|$j_ ddd|dk(s|dk(r|j| }.nta|jj,dd}/ta|jj,dd}0| |/z |0z} |jjc| dd }.|r|jje|.}.|jjg|.|}.|ji|s|.fStk|.S#1swYxYw)a 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. negative_prompt (`str`, *optional*, defaults to `""`): The prompt or prompts not to guide the image generation. Ignored when not using guidance (i.e., ignored if `guidance_scale` is less than `1`). height (`int`, *optional*, defaults to self.transformer.config.sample_size * self.vae_scale_factor): The height in pixels of the generated image. width (`int`, *optional*, defaults to self.transformer.config.sample_size * self.vae_scale_factor): The width in pixels of the generated image. num_inference_steps (`int`, *optional*, defaults to 28): The number of denoising steps. More denoising steps usually lead to a higher quality image at the expense of slower inference. 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 4.0): Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://huggingface.co/papers/2207.12598). `guidance_scale` is defined as `w` of equation 2. of [Imagen Paper](https://huggingface.co/papers/2205.11487). Guidance scale is enabled by setting `guidance_scale > 1`. Higher guidance scale encourages to generate images that are closely linked to the text `prompt`, usually at the expense of lower image quality. num_images_per_prompt (`int`, *optional*, defaults to 1): The number of images to generate per prompt. 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 be generated by sampling using the supplied random `generator`. prompt_embeds (`torch.FloatTensor`, *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.FloatTensor`, *optional*): Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not provided and `guidance_scale > 1`, negative embeddings will be generated from an empty string. prompt_attention_mask (`torch.BoolTensor`, *optional*): Pre-generated attention mask for `prompt_embeds`. If not provided, attention mask will be generated from `prompt` input argument. negative_prompt_attention_mask (`torch.BoolTensor`, *optional*): Pre-generated attention mask for `negative_prompt_embeds`. If not provided and `guidance_scale > 1`, attention mask will be generated from an empty string. 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.prx.PRXPipelineOutput`] instead of a plain tuple. use_resolution_binning (`bool`, *optional*, defaults to `True`): If set to `True`, the requested height and width are first mapped to the closest resolutions using predefined aspect ratio bins. After the produced latents are decoded into images, they are resized back to the requested resolution. Useful for generating non-square images at optimal resolutions. 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, step, timestep, callback_kwargs)`. `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 tensors that are listed in the `._callback_tensor_inputs` attribute. Examples: Returns: [`~pipelines.prx.PRXPipelineOutput`] or `tuple`: [`~pipelines.prx.PRXPipelineOutput`] if `return_dict` is True, otherwise a `tuple. When returning a tuple, the first element is a list with the generated images. NzResolution binning requires a VAE with image_processor, but VAE is not available. Set use_resolution_binning=False or provide a VAE.z@Resolution binning is only supported for default_sample_size in z , but got zC. Set use_resolution_binning=False to disable aspect ratio binning.)ratios)latentrz!VAE is required for output_type='zj' but it is not available. Either provide a VAE or set output_type='latent' or 'pt' to get latent outputs.rjr)rprrrTrrr)rr~)r~etagr)totalriF) hidden_statestimestepencoder_hidden_statesrrrrscaling_factorg{P? shift_factor)r)r)images)6rxrdrrZASPECT_RATIO_BINSrkeysclassify_height_width_binrrUrr>rDrrrarrprW set_timestepsr num_timestepsrllatent_channelsrV in_channelsrr}setinspect signaturestep parametersrcatmaxorder progress_bar enumeratefloatnum_train_timestepsrrrchunk prev_samplelocalsupdaterudecoderesize_and_crop_tensor postprocessmaybe_free_model_hooksr)1r0rrr{r|rrrsrrrSrTrrrrrrrrdefault_resolutionaspect_ratio_bin orig_height orig_widthryr~rrrrrzextra_step_kwargs accepts_etaca_embedca_masknum_warmup_stepsrit latents_int_cont noise_pred noise_uncond noise_textcallback_kwargsrimagerrs1 r1__call__zPRXPipeline.__call__sKH"88:--++ !##+ I''/@@ VW[\m\r\r\tWuVvw#7788{} 11I1IJ '-eK 00JJ6SXaqJrMFE      .  "  88  3C C3K=Abb   *VS"9J  Jvt$<VJ&,,Q/J''-\`[m[m  (,(H(H+"7'#9"7+I\n \ X*@BX( !7  NN ( (9V ( L00I"%i.  NN ( ()rr FloatTensor BoolTensorrrrrrrrEXAMPLE_DOC_STRINGrrr __classcell__)res@r1rRrRs (=(/:!7>B-?!(*!(3!(% !( *B<@EI2 c49n%2 &2 &* 2  2 # 2   1 122 !)):): ;2  ((8(892 )11A1A(B2 h Zc ZELL Z"-1! $`S $` $`&* $`  $`XCG59>B2c49n%22 2  2 -5T#Y,? 2  1 122!)):): ;2hU]]_12)-! $##%# #/0MQ*.59>B<@EI%* '+KO9B )R/c49n%R/R/ R/ } R/ ! R/9R/R/ (}R/E%//43H"HIJR/%,,'R/  1 12R/!)):): ;R/ ((8(89R/)11A1A(BR/ c]!R/"#R/$!%%R/&'xc40@$0F'GH'R/(-1I)R/3R/r3rR)5rHrr, urllib.parseparser?typingrrrrrrFr transformersrr r ,transformers.models.t5gemma.modeling_t5gemmar diffusers.image_processorr diffusers.loadersr rrdiffusers.modelsrr-diffusers.models.transformers.transformer_prxr"diffusers.pipelines.pipeline_utilsr'diffusers.pipelines.prx.pipeline_outputrdiffusers.schedulersrdiffusers.utilsrrdiffusers.utils.torch_utilsrrvASPECT_RATIO_256_BINASPECT_RATIO_512_BINASPECT_RATIO_1024_BINr get_loggerrLloggerr*r rRrPr3r1r s  88  H:__9O@E@5#J : #J : #J #J : : #J : #J #J #J : #J #J #J #J #J :  S$K S$K S$K S$K   S$K   S$K  S$K S$K S$K S$K S$K D$< T3K T3K T3K  T3K!" T3K#$3K #; 3K 3K 3K 3K 3K18      H %||~"f/ f/r3