bi ddlZddlmZmZmZmZmZmZddlZ ddl Z ddl m Z m Z mZmZddlmZmZddlmZmZmZddlmZddlmZdd lmZdd lmZmZm Z m!Z!m"Z"m#Z#dd l$m%Z%d d l&m'Z'ddl(m)Z)er ddl*m+cm,Z-dZ.ndZ.e j^e0Z1dZ2 dde3de3de4de4fdZ5 d dee3deee6e jnfdeee3deee4fdZ8Gdde'eeeZ9y)!N)AnyCallableDictListOptionalUnion) CLIPTextModel CLIPTokenizerT5EncoderModelT5TokenizerFast)PipelineImageInputVaeImageProcessor)FluxLoraLoaderMixinFromSingleFileMixinTextualInversionLoaderMixin) AutoencoderKL)FluxTransformer2DModel)FlowMatchEulerDiscreteScheduler)USE_PEFT_BACKENDis_torch_xla_availableloggingreplace_example_docstringscale_lora_layersunscale_lora_layers) randn_tensor)DiffusionPipeline)FluxPipelineOutputTFa Examples: ```py >>> import torch >>> from controlnet_aux import CannyDetector >>> from diffusers import FluxControlPipeline >>> from diffusers.utils import load_image >>> pipe = FluxControlPipeline.from_pretrained( ... "black-forest-labs/FLUX.1-Canny-dev", torch_dtype=torch.bfloat16 ... ).to("cuda") >>> prompt = "A robot made of exotic candies and chocolates of different kinds. The background is filled with confetti and celebratory gifts." >>> control_image = load_image( ... "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/robot.png" ... ) >>> processor = CannyDetector() >>> control_image = processor( ... control_image, low_threshold=50, high_threshold=200, detect_resolution=1024, image_resolution=1024 ... ) >>> image = pipe( ... prompt=prompt, ... control_image=control_image, ... height=1024, ... width=1024, ... num_inference_steps=50, ... guidance_scale=30.0, ... ).images[0] >>> image.save("output.png") ``` base_seq_len max_seq_len base_shift max_shiftc<||z ||z z }|||zz }||z|z}|SN) image_seq_lenr!r"r#r$mbmus i/home/cdr/jupyterlab/.venv/lib/python3.12/site-packages/diffusers/pipelines/flux/pipeline_flux_control.pycalculate_shiftr-Vs; Z K,$>?AQ%%A  Q B Inum_inference_stepsdevice timestepssigmasc | | td|dttj|jj j v}|std|jd|jd ||d||j}t|}||fS|dttj|jj j v}|std|jd|jd ||d||j}t|}||fS|j|fd |i||j}||fS) a Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`. Args: scheduler (`SchedulerMixin`): The scheduler to get timesteps from. num_inference_steps (`int`): The number of diffusion steps used when generating samples with a pre-trained model. If used, `timesteps` must be `None`. device (`str` or `torch.device`, *optional*): The device to which the timesteps should be moved to. If `None`, the timesteps are not moved. timesteps (`List[int]`, *optional*): Custom timesteps used to override the timestep spacing strategy of the scheduler. If `timesteps` is passed, `num_inference_steps` and `sigmas` must be `None`. sigmas (`List[float]`, *optional*): Custom sigmas used to override the timestep spacing strategy of the scheduler. If `sigmas` is passed, `num_inference_steps` and `timesteps` must be `None`. Returns: `Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the second element is the number of inference steps. zYOnly one of `timesteps` or `sigmas` can be passed. Please choose one to set custom valuesr1zThe current scheduler class zx's `set_timesteps` does not support custom timestep schedules. Please check whether you are using the correct scheduler.)r1r0r2zv's `set_timesteps` does not support custom sigmas schedules. Please check whether you are using the correct scheduler.)r2r0r0r') ValueErrorsetinspect signature set_timesteps parameterskeys __class__r1len) schedulerr/r0r1r2kwargsaccepts_timesteps accept_sigmass r,retrieve_timestepsrAds>!3tuu'3w/@/@AXAX/Y/d/d/i/i/k+ll .y/B/B.CDab  M)FMfM'' !)n ) ))   C(9(9):Q:Q(R(](](b(b(d$ee .y/B/B.CD_`  GvfGG'' !)n ) ))  3MFMfM'' ) ))r.c)eZdZdZdZgZddgZdedede de d e d e d e ffd Z d:deeeefdededeej*deej,f dZ d;deeeefdedeej*fdZ dd#Z% d?d$Z&e'd%Z(e'd&Z)e'd'Z*e'd(Z+ejXe-e.d d d d d d)d d*dd d d d d+d,d d dgdfdeeeefdeeeeefd-e/d.eed/eed0ed1eeed2edeed3eeej`eej`fdeej2deej2deej2d4eed5e1d6ee2ee3fd7ee4eee2gd fd8eedef&d9Z5xZ6S)@FluxControlPipelinea The Flux pipeline for controllable text-to-image generation with image conditions. Reference: https://bfl.ai/flux-1-tools Args: transformer ([`FluxTransformer2DModel`]): Conditional Transformer (MMDiT) architecture to denoise the encoded image latents. scheduler ([`FlowMatchEulerDiscreteScheduler`]): A scheduler to be used in combination with `transformer` to denoise the encoded image latents. vae ([`AutoencoderKL`]): Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations. text_encoder ([`CLIPTextModel`]): [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModel), specifically the [clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14) variant. text_encoder_2 ([`T5EncoderModel`]): [T5](https://huggingface.co/docs/transformers/en/model_doc/t5#transformers.T5EncoderModel), specifically the [google/t5-v1_1-xxl](https://huggingface.co/google/t5-v1_1-xxl) variant. tokenizer (`CLIPTokenizer`): Tokenizer of class [CLIPTokenizer](https://huggingface.co/docs/transformers/en/model_doc/clip#transformers.CLIPTokenizer). tokenizer_2 (`T5TokenizerFast`): Second Tokenizer of class [T5TokenizerFast](https://huggingface.co/docs/transformers/en/model_doc/t5#transformers.T5TokenizerFast). z.text_encoder->text_encoder_2->transformer->vaelatents prompt_embedsr=vae text_encoder tokenizertext_encoder_2 tokenizer_2 transformerc  t||j|||||||t|ddr/dt |j j jdz znd|_t|ddr |j j jnd|_ t|jdz|j|_ t|dr"|j|jjnd |_d |_y) N)rFrGrIrHrJrKr=rFrr)vae_scale_factorvae_latent_channelsrHM)super__init__register_modulesgetattrr<rFconfigblock_out_channelsrOlatent_channelsrPrimage_processorhasattrrHmodel_max_lengthtokenizer_max_lengthdefault_sample_size) selfr=rFrGrHrIrJrKr;s r,rTzFluxControlPipeline.__init__s  %)##  W^^bdikoVpc$((//*L*L&MPQ&Q RvwFMdTY[_F`488??#B#Bfh  1!22Q6DLdLd 07t[/IdnnNhDNN + +np !$' r.Nrpromptnum_images_per_promptmax_sequence_lengthr0dtypec |xs |j}|xs|jj}t|tr|gn|}t |}t|t r|j||j}|j|d|dddd}|j}|j|ddj} | jd|jdk\rbtj|| sL|jj| dd|jd z df} tj!d |d | |j#|j%|d d } |j"j}| j%||} | j\} } } | j'd |d } | j)||z| d} | S)N max_lengthTFpt)paddingrf truncation return_lengthreturn_overflowing_tokensreturn_tensorslongestrhrlrzXThe following part of your input was truncated because `max_sequence_length` is set to tokens: output_hidden_statesrrdr0)_execution_devicerGrd isinstancestrr<rmaybe_convert_promptrJ input_idsshapetorchequal batch_decoder]loggerwarningrItorepeatview)r_rarbrcr0rd batch_size text_inputstext_input_idsuntruncated_ids removed_textrE_seq_lens r,_get_t5_prompt_embedsz)FluxControlPipeline._get_t5_prompt_embedss14110**00'4&&[ d7 8..vt7G7GHF&&  *&+' %..**69UY*Zdd   $(<(>66q$JcJcfgJgjlJlGl7mnL NN--.i ~G )).*;*;F*CZ_)` &33 %((t/@/@/F/Fv(V &,,Q0EF %**:8M+MrR r.prompt_2pooled_prompt_embeds lora_scalec l|xs |j}|gt|trW||_|jt rt |j||jt rt |j|t|tr|gn|}|D|xs|}t|tr|gn|}|j|||}|j||||}|j,t|trt rt|j||j,t|trt rt|j||j|jjn|jj} tj|j ddj#|| } ||| fS)a Args: prompt (`str` or `List[str]`, *optional*): prompt to be encoded prompt_2 (`str` or `List[str]`, *optional*): The prompt or prompts to be sent to the `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is used in all text-encoders device: (`torch.device`): torch device num_images_per_prompt (`int`): number of images that should be generated per prompt 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. pooled_prompt_embeds (`torch.FloatTensor`, *optional*): Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not provided, pooled text embeddings will be generated from `prompt` input argument. lora_scale (`float`, *optional*): A lora scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded. )rar0rb)rarbrcr0rr r0rd)rtrur _lora_scalerGrrrIrvrrrrdrKrzzerosryr) r_rarr0rbrErrcrrdtext_idss r, encode_promptz!FluxControlPipeline.encode_promptDs@1411  !j7J&K)D   ,1A!$"3"3Z@"".3C!$"5"5zB'4&&  )6H%/#%>zHH$(#?#?&;$@$ !66&;$7 7M    ($ 349I#D$5$5zB    *$ 349I#D$7$7D+/+<+<+H!!''dN^N^NdNd;;}2215q9<z3FluxControlPipeline.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.z Cannot forward both `prompt_2`: 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 z4`prompt_2` has to be of type `str` or `list` but is zIf `prompt_embeds` are provided, `pooled_prompt_embeds` also have to be passed. Make sure to generate `pooled_prompt_embeds` from the same text encoder that was used to generate `prompt_embeds`.r`z8`max_sequence_length` cannot be greater than 512 but is ) rOr}r~allr4rrurvlisttype) r_rarheightwidthrEr"callback_on_step_end_tensor_inputsrcrs ` r, check_inputsz FluxControlPipeline.check_inputssD T**Q. /1 4AVAVYZAZ8[_`8` NN?@U@UXY@Y?ZZcdjckkpqvpwx`a  . 9#F 7YF C DTEaEaDbbn|^pHvwbckokGkGbGpqpHoIJ   -";08N}o^00  !m&?28*RSTVZSZ>[#[ #3F#;ell5>QRVXYRY>Z#Z RbRhRhO 57O+33 "%: :>r.c|j|||dzd|dzd}|jdddddd}|j||dz|dzz|dz}|S)Nrrrr )rpermuter)rDrnum_channels_latentsrrs r, _pack_latentsz!FluxControlPipeline._pack_latentssl,,z+?1aQVZ[Q[]^_//!Q1a3//*v{uz.JL`cdLder.c|j\}}}dt||dzzz}dt||dzzz}|j||dz|dz|dzdd}|jdddddd}|j ||dz||}|S)Nrrrr rr)ryintrrr)rDrrrOr num_patcheschannelss r,_unpack_latentsz#FluxControlPipeline._unpack_latentss-4MM) Kc&k&6&:;<SZ$4q$89:,,z6Q; HPQMSTVWX//!Q1a3//*h5.A65Qr.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)rFenable_slicingr_s r,enable_vae_slicingz&FluxControlPipeline.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)rFdisable_slicingrs r,disable_vae_slicingz'FluxControlPipeline.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)rF enable_tilingrs r,enable_vae_tilingz%FluxControlPipeline.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)rFdisable_tilingrs r,disable_vae_tilingz&FluxControlPipeline.disable_vae_tilingrr.c dt||jdzzz}dt||jdzzz}||||f} |0|j||dz|dz||} |j||| fSt |t r)t ||k7rtdt |d|dt| |||}|j|||||}|j||dz|dz||} || fS)Nrrz/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.) generatorr0rd) rrOrrrurr<r4rr) r_rrrrrdr0rrDryrs r,prepare_latentsz#FluxControlPipeline.prepare_latents s8c&kd&;&;a&?@ASZD$9$9A$=>?165A  #==j&TU+W\`aWacikpq ::V5:9;KK K i &3y>Z+GA#i.AQR&<'gi  u &PUV$$Wj:NPVX]^99*fPQkSX\]S]_eglm(((r.c 0t|tjrn|jj |||}|j d} | dk(r|} n|} |j | d}|j||}|r| stj|gdz}|S)N)rrrrdimrr) rurzTensorrZ preprocessryrepeat_interleavercat) r_imagerrrrbr0rddo_classifier_free_guidance guess_modeimage_batch_size repeat_bys r, prepare_imagez!FluxControlPipeline.prepare_image0s eU\\ * ((33E&PU3VE ;;q> q "I.I'' q'9e4 &zIIugk*E r.c|jSr&)_guidance_scalers r,guidance_scalez"FluxControlPipeline.guidance_scaleRs###r.c|jSr&)_joint_attention_kwargsrs r,joint_attention_kwargsz*FluxControlPipeline.joint_attention_kwargsVs+++r.c|jSr&)_num_timestepsrs r, num_timestepsz!FluxControlPipeline.num_timestepsZs"""r.c|jSr&) _interruptrs r, interruptzFluxControlPipeline.interrupt^s r.g @pilT control_imagerrr/r2rr output_type return_dictrcallback_on_step_endrc  |xs|j|jz}|xs|j|jz}|j||||| | ||||_||_d|_|t |trd}n-|t |tr t|}n| jd}|j}|j|jjddnd}|j||| | || ||\} } }|jj j"dz}|j%||||| z| ||j&j( }|j*d k(r|j&j-|j.j1| }||j&j j2z |j&j j4z}|jd d\}}|j7||| z|||}|j9|| z|||| j(|| | \} }|t;j<d d|z |n|}| jd}t?||j@j jdd|j@j jdd|j@j jdd|j@j jdd}tC|j@||||\}}tEt|||j@jFzz d}t||_$|jj jJrGtMjNdg||tLjP} | jS| jd} nd} |jU|5}!tW|D]\}"}#|jXrtMjZ| |gd }$|#jS| jdj]| j(}%|j|$|%dz | | | |||jd d}&| j(}'|j@j_|&|#| dd} | j(|'k7r9tLj`jbjer| j]|'} |Hi}(|D]})tg|)|(|)<|||"|#|(}*|*jid| } |*jid| } |"t|dz k(s'|"dz|kDr/|"dz|j@jFzdk(r|!jktlstojp ddd|dk(r| }+n|js| |||j} | |j&j j4z |j&j j2z} |j&ju| dd}+|jvjy|+| }+|j{|s|+fSt}|+!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. prompt_2 (`str` or `List[str]`, *optional*): The prompt or prompts to be sent to `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is will be used instead control_image (`torch.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.Tensor]`, `List[PIL.Image.Image]`, `List[np.ndarray]`,: `List[List[torch.Tensor]]`, `List[List[np.ndarray]]` or `List[List[PIL.Image.Image]]`): The ControlNet input condition to provide guidance to the `unet` for generation. If the type is specified as `torch.Tensor`, it is passed to ControlNet as is. `PIL.Image.Image` can also be accepted as an image. The dimensions of the output image defaults to `image`'s dimensions. If height and/or width are passed, `image` is resized accordingly. If multiple ControlNets are specified in `init`, images must be passed as a list such that each element of the list can be correctly batched for input to a single ControlNet. height (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor): The height in pixels of the generated image. This is set to 1024 by default for the best results. width (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor): The width in pixels of the generated image. This is set to 1024 by default for the best results. num_inference_steps (`int`, *optional*, defaults to 50): The number of denoising steps. More denoising steps usually lead to a higher quality image at the expense of slower inference. sigmas (`List[float]`, *optional*): Custom sigmas to use for the denoising process with schedulers which support a `sigmas` argument in their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed will be used. guidance_scale (`float`, *optional*, defaults to 3.5): Embedded guidance 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 classifier-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. 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.FloatTensor`, *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.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. pooled_prompt_embeds (`torch.FloatTensor`, *optional*): Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not provided, pooled text embeddings will be generated from `prompt` input argument. 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.flux.FluxPipelineOutput`] instead of a plain tuple. joint_attention_kwargs (`dict`, *optional*): 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). 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 512): Maximum sequence length to use with the `prompt`. Examples: Returns: [`~pipelines.flux.FluxPipelineOutput`] or `tuple`: [`~pipelines.flux.FluxPipelineOutput`] if `return_dict` is True, otherwise a `tuple`. When returning a tuple, the first element is a list with the generated images. )rErrrcFNrrscale)rarrErr0rbrcrrM)rrrrrbr0rdr)rrg?base_image_seq_lenmax_image_seq_lenr#?r$ffffff?)r2r+r)totalri) hidden_statestimestepguidancepooled_projectionsencoder_hidden_statestxt_idsimg_idsrr)rrDrElatent)r)images)?r^rOrrrrrurvrr<ryrtrgetrrKrW in_channelsrrFrdndimencode latent_distsample shift_factorscaling_factorrrnplinspacer-r=rAmaxorderrguidance_embedsrzfullfloat32expand progress_bar enumeraterrrstepbackendsmps is_availablelocalspopupdate XLA_AVAILABLExm mark_steprdecoderZ postprocessmaybe_free_model_hooksr ),r_rarrrrr/r2rrbrrDrErrrrrrrcrr0rrrheight_control_imagewidth_control_imagerr(r+r1num_warmup_stepsrritlatent_model_inputr noise_pred latents_dtypecallback_kwargsrcallback_outputsrs, r,__call__zFluxControlPipeline.__call__bsJLK433d6K6KKI11D4I4II     '!5/Q 3   .'=$  *VS"9J  Jvt$<VJ&,,Q/J''?C>Y>Y>eD ' ' + +GT :ko    '!5"7 3!     $//66BBaG**!$99"7((..+     " HHOOM:FFMMXaMbM*TXX__-I-IITXX__MkMkkM8E8K8KAB8O 5 "5 ..22$$# M%)$8$8 . .         % !!TZSaS!&9"9;NOgm a(   NN ! ! % %&:C @ NN ! ! % %&94 @ NN ! ! % %lC 8 NN ! ! % %k4 8  *< NN   * & &s9~0CdnnFZFZ0ZZ\]^!)n    " " 2 2zz1#~fEMMZHw}}Q'78HH  %8  9- #\!),, #1>>%*YY/GQ%O"88GMM!$4588G!--"4%_%';*7$,+/+F+F %.   !( ..--j!WRW-XYZ[==M1~~))668")**]";'3&(O?9-3Xa[*9';D!Q'X$.229gFG$4$8$8-$XMI**A9I/IqSTuX\XfXfXlXlNlpqNq '') LLNY, #- #^ ( "E**7FE4CXCXYG!?!??488??C_C__GHHOOGO?BE((44U 4TE ##%8O!//{- #- #s F(Y7YY()Nrr`NN)rN)NNrNNr`NNNNNr&)FF)7__name__ __module__ __qualname____doc__model_cpu_offload_seq_optional_componentsrrrr r r r rrTrrvrrrrzr0rdrr FloatTensorfloatrr staticmethodrrrrrrrrrpropertyrrrrno_gradrEXAMPLE_DOC_STRINGr Generatorboolrrrr! __classcell__)r;s@r,rCrCsj 4M(/:'2''$ ' ! ' ' '%','H)-%&#&)-'+ /c49n%/ #/! / & /  $ /j&')- *c49n%* #*& *`59)-%&59<@#&&*M=c49n%M=5d3i01M=& M= # M=   1 12 M='u'8'89M=!M=UOM=j!+/ 0od ? ?  "#!""!)Z%* D$$,,##U]]_12)-48,0 $##%(, #/0MQ/359<@%* ;?KO9B #&)R0c49n%R05d3i01R0* R0  R0 } R0!R0e%R0R0 (}R0E%//43H"HIJR0%++,R0  1 12R0'u'8'89R0c]R0 !R0"!)c3h 8#R0$'xc40@$0F'GH%R0&-1I'R0(!)R03R0r.rC)rrrrr"):r6typingrrrrrrnumpyrrz transformersr r r r rZrrloadersrrrmodels.autoencodersrmodels.transformersr schedulersrutilsrrrrrrutils.torch_utilsrpipeline_utilsrpipeline_outputr torch_xla.core.xla_modelcore xla_modelrr get_loggerr#r}r.rr*r-rvr0rArCr'r.r,rAs,== VVD\\099../))MM   H % L       *.15%)$( 8*!#8* U3 ,- .8*S " 8* T%[ ! 8*vW 0 W 0r.