3j SSKrSSKJrJr SSKrSSKrSSKrSSKJ r J r SSK J r SSK JrJr SSKJr SSKJrJrJr SS KJr S S KJr S S KJr S SKJr \"5(a SSKJs J r! Sr"OSr"\RF"\$5r%Sr&S\'S\'S\(4Sjr)S!S\'S-S\*\RV-S-S\,\'S-S\,\(S-4Sjjr-S"S\R\S\R^S-S\*4Sjjr0"SS \\ 5r1g)#N)AnyCallable)Qwen2TokenizerFastQwen3ForCausalLM)Flux2LoraLoaderMixin)AutoencoderKLFlux2Flux2Transformer2DModel)FlowMatchEulerDiscreteScheduler)is_torch_xla_availableloggingreplace_example_docstring) randn_tensor)DiffusionPipeline)Flux2ImageProcessor)Flux2PipelineOutputTFa Examples: ```py >>> import torch >>> from diffusers import Flux2KleinPipeline >>> pipe = Flux2KleinPipeline.from_pretrained( ... "black-forest-labs/FLUX.2-klein-base-9B", torch_dtype=torch.bfloat16 ... ) >>> pipe.to("cuda") >>> prompt = "A cat holding a sign that says hello world" >>> # Depending on the variant being used, the pipeline call will slightly vary. >>> # Refer to the pipeline documentation for more details. >>> image = pipe(prompt, num_inference_steps=50, guidance_scale=4.0).images[0] >>> image.save("flux2_output.png") ``` image_seq_len num_stepsreturncSup#SupEUS:aX@-U-n[U5$X@-U-nX -U-nXx- S- n USU -- n X-U -n[U5$)N)gT ?gŒ_?)g w:/&?gDw:?igg@gi@)float) rra1b1a2b2mum_200m_10abs h/home/wildlama/miniconda3/lib/python3.13/site-packages/diffusers/pipelines/flux2/pipeline_flux2_klein.pycompute_empirical_mur$?sy 'FB #FBt  " $Ry   #E  "D A  A  B 9num_inference_stepsdevice timestepssigmasc UbUb [S5eUbS[[R"UR5R R 55;nU(d[SURS35eUR"S X2S.UD6 URn[U5nX14$UbS[[R"UR5R R 55;nU(d[SURS35eUR"S XBS.UD6 URn[U5nX14$UR"U4S U0UD6 URnX14$) 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 r#retrieve_timestepsr9Rs}>!3tuu'3w/@/@AXAX/Y/d/d/i/i/k+ll .y/B/B.CDab  M)MfM'' !)n  ))   C(9(9):Q:Q(R(](](b(b(d$ee .y/B/B.CD_`  GvGG'' !)n  ))  3MFMfM''  ))r%encoder_output generator sample_modec[US5(a!US:XaURRU5$[US5(a US:XaURR5$[US5(a UR$[ S5e)N latent_distsampleargmaxlatentsz3Could not access latents of provided encoder_output)hasattrr>r?moderAAttributeError)r:r;r<s r#retrieve_latentsrEs}~}--+2I))00;;  / /K84K))..00  + +%%%RSSr%c*^\rSrSrSrSrSS/rSES\S\S\ S \ S \ S \ 4 U4S jjjr \SFS\ S \ S\\\-S\R$S -S\R&S -S\S\\4Sjj5r\SGS\R,S\R,S -4Sjj5r\S\R,4Sj5r\SHS\\R,S\4Sjj5r\S5r\S5r\S5r\SIS\R,S \R,S!\S -S"\S -S#\\R,4 S$jj5rSJS\\\-S\R&S -S&\S\R,S -S\S'\\4 S(jjrS)\R,S*\R@4S+jr!SGS*\R@S\R,S -4S,jjr"S-\\R,S*\R@4S.jr#SKS/jr$\%S05r&\%S15r'\%S25r(\%S35r)\%S45r*\%S55r+\RX"5\-"\.5S S S S S6S S7S%S S S S S8S9S S S/SS4S)\\/R`R`\/R`R`-S -S\\\-S!\S -S"\S -S:\S;\\1S -S<\1S&\S*\R@\\R@-S -S\R,S -S\R,S -S=\\\-S -S>\S?\ S@\2\\34S -SA\4\\\2/S 4S -SB\\S\S'\\4&SCjj55r5SDr6U=r7$)LFlux2KleinPipelinea The Flux2 Klein pipeline for text-to-image generation. Reference: [https://bfl.ai/blog/flux2-klein-towards-interactive-visual-intelligence](https://bfl.ai/blog/flux2-klein-towards-interactive-visual-intelligence) Args: transformer ([`Flux2Transformer2DModel`]): 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 ([`AutoencoderKLFlux2`]): Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations. text_encoder ([`Qwen3ForCausalLM`]): [Qwen3ForCausalLM](https://huggingface.co/docs/transformers/en/model_doc/qwen3#transformers.Qwen3ForCausalLM) tokenizer (`Qwen2TokenizerFast`): Tokenizer of class [Qwen2TokenizerFast](https://huggingface.co/docs/transformers/en/model_doc/qwen2#transformers.Qwen2TokenizerFast). ztext_encoder->transformer->vaerA prompt_embedsr5vae text_encoder tokenizer transformer is_distilledcH>[TU]5 URUUUUUS9 URUS9 [ USS5(a/S[ UR RR5S- -OSUl [URS-S9Ul SUl S Ul g) N)rJrKrLr5rM)rNrJrr)vae_scale_factor)super__init__register_modulesregister_to_configgetattrr4rJconfigblock_out_channelsrQrimage_processortokenizer_max_lengthdefault_sample_size)selfr5rJrKrLrMrNr3s r#rUFlux2KleinPipeline.__init__s  %#   \:V]^bdikoVpVpc$((//*L*L&MPQ&Q Rvw 3DDYDY\]D]^$'!#& r%NrR promptdtyper'max_sequence_lengthhidden_states_layersc Uc UROUnUc UROUn[U[5(aU/OUn/n/nUHMn SU S./n UR U SSSS9n U"U SSSUS9n UR U S 5 UR U S 5 MO [ R"US S 9RU5n [ R"US S 9RU5nU"U USSS 9n[ R"UVs/sHnURUPM snSS 9nURX4S9nURunnnnURS SSS5RUUUU-5nU$s snf)Nuser)rolecontentFT)tokenizeadd_generation_promptenable_thinkingpt max_length)return_tensorspadding truncationrp input_idsattention_maskrdim)rtruoutput_hidden_states use_cacher)rer'rr)rer' isinstancestrapply_chat_templateappendtorchcattostack hidden_statesshapepermutereshape)rKrLrdrer'rfrg all_input_idsall_attention_masks single_promptmessagestextinputsrtruoutputkout batch_size num_channelsseq_len hidden_dimrIs r#_get_qwen3_prompt_embeds+Flux2KleinPipeline._get_qwen3_prompt_embedss',m ""(.$$F'44&&  #M!'MBCH00&* % 1D #$. F   !4 5  & &v.>'? @#$&IIm366v> #6A>AA&I)!%  kk-..J4P]K^J__`ab b05||As=?Q/RS/R!EI%/RS(011FF2J1 0DA ! A ww Au((ELL,@%,,uBUW\WcWcdeWfgE  # #E * 1!99%51=+55a8T1s E&E cURupp4URXUS-SUS-S5nURSSSSSS5nURXS-US-US-5nU$)Nrrrr)rrrrrArnum_channels_latentsrrs r#_patchify_latents$Flux2KleinPipeline._patchify_latentspsu;B--7 &,,z1aQVZ[Q[]^_//!Q1a3//*Q.FRS UZ^_U_`r%cURupp4URXS-SSX45nURSSSSSS5nURXS-US-US-5nU$)Nrrrrrrrrrrs r#_unpatchify_latents&Flux2KleinPipeline._unpatchify_latentsysn;B--7 &//*u.MqRSU[c//!Q1a3//*u.MvXYz[`cd[der%cnURupp4URXX4-5RSSS5nU$)zg pack latents: (batch_size, num_channels, height, width) -> (batch_size, height * width, num_channels) rrrr)rArrrrs r# _pack_latents Flux2KleinPipeline._pack_latentss;3:--/ &//*FNKSSTUWXZ[\r%rrrrc/n[X5GH7upVURupxUSS2S4R[R5n USS2S4R[R5n UbUO[R "U 5S-n UbUO[R "U 5S-n X-U -n [R "X-U4URURS9nURSU RS5RSU5U5 URXU5RSSS5nURU5 GM: [R"USS9$)z1 using position ids to scatter tokens into place Nrrr'rerrrv)rrrr~int64maxzerosr'rescatter_rrrrr}r)rrrrx_listdataposrchh_idsw_idsrrflat_idsrs r#_unpack_latents_with_ids+Flux2KleinPipeline._unpack_latents_with_idss"QIDJJEA1ILL-E1ILL-E!,%))E2BQ2FA* %0@10DAy5(H++qubk$++TZZPC LLH..q188R@$ G((1$,,Q15C MM# #'&{{6q))r%rnum_images_per_prompttext_encoder_out_layersc xU=(d URnUcSn[U[5(aU/OUnUc(URURUR UUUUS9nUR upxn URSUS5nURXs-US5nURU5n U RU5n XJ4$)N)rKrLrdr'rfrgrr) _execution_devicerzr{rrKrLrrepeatrrr) r^rdr'rrIrfrrrrtext_idss r# encode_prompt Flux2KleinPipeline.encode_prompts1411 >F'44&&   99!....$7%< :M"/!4!4 Q%,,Q0EqI %**:+MwXZ[ ))-8;;v&&&r%imager;cURS:wa[SURS35e[URR U5USS9nUR U5nURR RRSSSS5RURUR5n[R"URR RRSSSS5URRR -5RURUR5nX4- U- nU$)NrzExpected image dims 4, got rr@)r;r<rr)ndimr,rErJencoderbn running_meanrrr'rer~sqrt running_varrYbatch_norm_eps)r^rr;rlatents_bn_meanlatents_bn_stds r#_encode_vae_image$Flux2KleinPipeline._encode_vae_images ::?:5::,aHI I()?9bjk ..}= ((++22772q!DGG H\H\^k^q^qrDHHKK$;$;$@$@B1$MPTPXPXP_P_PnPn$norr  -"5"5 '8NJ r%c S[U5URS---nS[U5URS---nXS-US-US-4n [U[5(a*[ U5U:wa[ S[ U5SUS35eUc [ XXeS9nOURXeS9nURU5n U RU5n URU5nX4$)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.)r;r'rer) intrQrzrr4r,rrrr) r^rnum_latents_channelsrrrer'r;rArrs r#prepare_latents"Flux2KleinPipeline.prepare_latentssc&kd&;&;a&?@ASZD$9$9A$=>?A5v{EQJO i & &3y>Z+GA#i.AQR&<'gi  ?"5fZGjjj +22:q!D+..v6..r%c ^UbUTRS--S:wdUbBUTRS--S:wa,[RSTRS-SUSUS35 UbW[U4SjU55(d=[ STR S UVs/sHowTR ;dMUPM sn35eUbUb[ S US US 35eUcUc [ S 5eUbA[ U[5(d,[ U[5(d[ S[U535eUS:a6TRR(a[RSUS35 gggs snf)Nrrz-`height` and `width` have to be divisible by z but are z and z(. Dimensions will be resized accordinglyc3@># UHoTR;v M g7fN)_callback_tensor_inputs).0rr^s r# 2Flux2KleinPipeline.check_inputs..5sF 7Y!-- -7Ysz2`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 ?zGuidance scale z+ is ignored for step-wise distilled models.) rQloggerwarningallr,rrzr{rrrYrN)r^rdrrrI"callback_on_step_end_tensor_inputsguidance_scalers` r# check_inputsFlux2KleinPipeline.check_inputs"s  $//!349 ..23q8 NN?@U@UXY@Y?ZZcdjckkpqvpwx`a  . 9#F 7YF C C DTEaEaDbbn|^pH|^vwkokGkGbGpq|^pHoIJ   -";08N}o^00 ^ 5w  FC)@)@TZ\`IaIaQRVW]R^Q_`a a C DKK$<$< NN_^,<R,S5(a'UR>R,R@(aSnU RSn"[CU"US9n#[EUR>UUUU#S9un$n[G[U$5XPR>RH-- S5n%[U$5Ul%UR>RMS5 UROUS9n&[QU$5GHun'n(URR(aMU(UlU(RUU RS5RWU R25n)U RWUR*R25n*Un+U bQ[XRZ"U U /SS9RWUR*R25n*[XRZ"UU!/SS9n+UR*R]S5 UR+U*U)S- SU UU+UR^SS9Sn,SSS5 W,SS2SU RS524n,UR(aoUR*R]S5 UR+U*U)S- SU WU+URSS9Sn-SSS5 W-SS2SU RS524n-U-UU,U-- --n,U R2n.UR>RaU,U(U SS9Sn U R2U.:wa>[XRbRdRg5(aU RWU.5n UbJ0n/UHn0[i5U0U/U0'M U"UU'U(U/5n1U1RkSU 5n U1RkSU 5n U'[U$5S- :Xd)U'S-U%:a0U'S-UR>RH-S:XaU&Rm5 [n(dGM[pRr"5 GM SSS5 SUlS [uU5UR"S ---n2S [uU5UR"S ---n3URwU UU2S -U3S -5n UR6RxRzR}SSSS5RWU R~U R25n4[XR"UR6RxRR}SSSS5UR6R,R-5RWU R~U R25n5U U5-U4-n URU 5n U S:XaU nO6UR6RU SS9SnURRXS9nUR5 U(dU4$[US9$!,(df  GN=f!,(df  GN@=f!,(df  GN=f) ak Function invoked when calling the pipeline for generation. Args: image (`torch.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.Tensor]`, `List[PIL.Image.Image]`, or `List[np.ndarray]`): `Image`, numpy array or tensor representing an image batch to be used as the starting point. For both numpy array and pytorch tensor, the expected value range is between `[0, 1]` If it's a tensor or a list or tensors, the expected shape should be `(B, C, H, W)` or `(C, H, W)`. If it is a numpy array or a list of arrays, the expected shape should be `(B, H, W, C)` or `(H, W, C)` It can also accept image latents as `image`, but if passing latents directly it is not encoded again. 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. 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. For step-wise distilled models, `guidance_scale` is ignored. 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. 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.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. negative_prompt_embeds (`torch.Tensor`, *optional*): Pre-generated negative text embeddings. Note that "" is used as the negative prompt in this pipeline. If not provided, will be generated from "". 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.qwenimage.QwenImagePipelineOutput`] instead of a plain tuple. 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`. text_encoder_out_layers (`tuple[int]`): Layer indices to use in the `text_encoder` to derive the final prompt embeddings. Examples: Returns: [`~pipelines.flux2.Flux2PipelineOutput`] or `tuple`: [`~pipelines.flux2.Flux2PipelineOutput`] if `return_dict` is True, otherwise a `tuple`. When returning a tuple, the first element is a list with the generated images. )rdrrrIr r NFrr)rdrIr'rrfrrircrop)rr resize_moder)rrrrrer'r;rA)rrr;r'reruse_flow_sigmas)rr)r)r)totalrvcondi)rtimestepguidanceencoder_hidden_statestxt_idsimg_idsjoint_attention_kwargsr*uncond)r*rArIrr)r))r)Hr rrrr#rzr{rr4rrrrr[check_image_inputsize_resize_to_target_arearQ preprocessr}r]rMrY in_channelsrrerrJnplinspacerBr5r/r$r9rorderrset_begin_index progress_bar enumerater$rrr~r cache_contextrstepbackendsmps is_availablelocalspopupdate XLA_AVAILABLExm mark_steprrrrrr'rrrrdecode postprocessmaybe_free_model_hooksr)6r^rrdrrr&r)r rr;rArIr(r)r*rr+r rfrrr'rnegative_promptnegative_text_idscondition_imagesimg image_width image_height multiple_ofrrrrrrr(num_warmup_stepsrBrrr2latent_model_inputlatent_image_ids noise_predneg_noise_pred latents_dtypecallback_kwargsrcallback_outputs latent_height latent_widthrrs6 r#__call__Flux2KleinPipeline.__call__csJ '/Q)   .!1!%  *VS"9"9J  Jvt$<$<VJ&,,Q/J''#'"4"4'"7 3$; #5#  x  + + O!j&>&>#2"3c&k"A8<8J8J&4&;$7(? 9K9 5 "$5  Zt%<%<GE  $$66s; " ,/HH) \- ;..EEc;WC03-K"33a7 *k9[H , ;{J **55c,Vaou5v '',/<,K433d6K6KKI11D4I4II $//66BBaG"22!$99!5%%3     '.2.H.H'%(==#hhnn /I/ +M+TZSaS!&9"9;NOgm 4>>((*; < <AVAVAfAfF a( ! I\ ]); NN   * & &s9~0CnnFZFZ0ZZ\]^!)n &&q)   %8  9\!),1>>)*&88GMM!$4588G%,ZZ0@0@0F0F%G"#-  ,).G]3KQR)S)V)VW[WgWgWmWm)n&',yy*>N1OUV'W$%%33F;!%!1!1&8!)D!%.; ( 0/3/D/D$)"2 " "J<(+>W\\!_+>(>? 33))77A)-)9)9*<%-_%)2H$5$4373I3I(-*: * *B&4A7Ja7J4J%KN!/.JQ_D_2`!`J!( ..--j!WRW-XYZ[==M1~~))6688")**]";'3&(O?-3Xa[*@';D!Q'X$.229gFG$4$8$8-$XMI**A9I/IqSTuX\XfXfXlXlNlpqNq '') =LLNG-:L"&S[T-B-BQ-FGH CJ4+@+@1+DEF //]VWEWYeijYjk((++22772q!DGGX_XeXefDHHKK$;$;$@$@B1$MPTPXPXP_P_PnPn$norr NNGMM N*_<**73 ( "EHHOOGO?BE((44U4TE ##%8O"%00_<;BA?: 9sE-C:b''a*Ab'a<Ebb* a9 4b< b b b) rrrr#rr]r[r\rQ)F)NNrRr`r) )NN)NrNrRr`)NNN)8__name__ __module__ __qualname____firstlineno____doc__model_cpu_offload_seqrr r rrr boolrU staticmethodr{rr~rer'rrTensorrrrrrrrtupler Generatorrrrr propertyr rrrr r$no_gradrEXAMPLE_DOC_STRINGPILImagerdictrrrc__static_attributes__ __classcell__)r3s@r#rGrGs>(=(/:#'2' '' ' & ' - '''8 %)&*#&*56&6%6d3i6{{T! 6  t# 6 ! 6#3i66p(,$ <<$$$$$>/ ELL)/ / / b]a* <<* % *69Dj*PSVZPZ* ell **>'+%&-1#&.9 'd3i ' t# ' # ' ||d* ' ! '"'s 'F u||  2(,#??#$#D /U\\" /?? /N+/'jR$$II&&##&& ]]_12AE"&! #%%) #%&DH'+-19=  26HL9B #&.9)x1CIIOO$syy6=x1d3ix1d x1 Tz x1 ! x1U d"x1x1 #x1??T%//%::TAx1$x1||d*x1!$d3i$ 6x1x1x1 sCx.4/!x1"'S$'7'=>E#x1$-1I%x1&!'x1("'s)x13x1r%rG)NNNN)Nr?)2r.typingrrnumpyr>rtr~ transformersrrloadersrmodelsr r schedulersr utilsr r rutils.torch_utilsrpipeline_utilsrr[rpipeline_outputrtorch_xla.core.xla_modelcore xla_modelrMrL get_loggerrfrrsrrr$r{r'rr9rnrprErGr+r%r#rs(  =+A9OO-.00))MM   H %(*'+(,"&!% 8*t8* %,,  %8*Cy4 8* K$  8*z`h TLL T-2__t-C TY\ TB 1*,@B 1r%