oMju SSKrSSKrSSKJrJr SSKrSSKJrJrJ r J r J r J r SSK Jr SSKJr SSKJrJr SSKJrJr SS KJrJrJrJr SS KJr S S KJr S SK J!r! \"5(a SSK"J#s J$r% Sr&OSr&\RN"\(5r)Sr*SS\+S\+S\,S\,4Sjjr-SS\+S-S\.\R^-S-S\0\+S-S\0\,S-4Sjjr1"SS\\5r2g) N)AnyCallable)CLIPTextModelWithProjection CLIPTokenizerLlamaForCausalLMPreTrainedTokenizerFastT5EncoderModel T5Tokenizer)VaeImageProcessor)HiDreamImageLoraLoaderMixin) AutoencoderKLHiDreamImageTransformer2DModel)FlowMatchEulerDiscreteSchedulerUniPCMultistepScheduler) deprecateis_torch_xla_availableloggingreplace_example_docstring) randn_tensor)DiffusionPipeline)HiDreamImagePipelineOutputTFa Examples: ```py >>> import torch >>> from transformers import AutoTokenizer, LlamaForCausalLM >>> from diffusers import HiDreamImagePipeline >>> tokenizer_4 = AutoTokenizer.from_pretrained("meta-llama/Meta-Llama-3.1-8B-Instruct") >>> text_encoder_4 = LlamaForCausalLM.from_pretrained( ... "meta-llama/Meta-Llama-3.1-8B-Instruct", ... output_hidden_states=True, ... output_attentions=True, ... torch_dtype=torch.bfloat16, ... ) >>> pipe = HiDreamImagePipeline.from_pretrained( ... "HiDream-ai/HiDream-I1-Full", ... tokenizer_4=tokenizer_4, ... text_encoder_4=text_encoder_4, ... torch_dtype=torch.bfloat16, ... ) >>> pipe.enable_model_cpu_offload() >>> image = pipe( ... 'A cat holding a sign that says "Hi-Dreams.ai".', ... height=1024, ... width=1024, ... guidance_scale=5.0, ... num_inference_steps=50, ... generator=torch.Generator("cuda").manual_seed(0), ... ).images[0] >>> image.save("output.png") ``` base_seq_len max_seq_len base_shift max_shiftc4XC- X!- - nX5U-- nX-U-nU$N) image_seq_lenrrrrmbmus C:\Users\johnn\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.13_qbz5n2kfra8p0\LocalCache\local-packages\Python313\site-packages\diffusers/pipelines/hidream_image/pipeline_hidream_image.pycalculate_shiftr'Vs3  K$>?A%%A  Q B Inum_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*r!) ValueErrorsetinspect signature set_timesteps parameterskeys __class__r+len) schedulerr)r*r+r,kwargsaccepts_timesteps accept_sigmass r&retrieve_timestepsr;ds}>!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(c;^\rSrSrSr/SQrS\S\S\S\ S\S \ S \ S \ S \ S \ S\4U4SjjrSDS\\\-S\S\R(S-S\R*S-4SjjrSES\\\-S\S\R(S-S\R*S-4SjjrSDS\\\-S\S\R(S-S\R*S-4SjjrSFS\\\-S-S\\\-S-S\\\-S-S\\\-S-S\R(S-S\R*S-S\S\S \\\-S-S!\\\-S-S"\\\-S-S#\\\-S-S$\\R4S-S%\\R4S-S&\\R4S-S'\\R4S-S(\R4S-S)\R4S-S\S*\S-4(S+jjrSGS,jrSHS-jr\S.5r \S/5r!\S05r"\S15r#\S25r$\RJ"5\&"\'5SSSSSSS3SS4SSSSSSSSSSSSSS5SSSS6/S4S\\\-S\\\-S-S\\\-S-S\\\-S-S7\S-S8\S-S9\S:\\S-S;\S \\\-S-S!\\\-S-S"\\\-S-S#\\\-S-S\S-S<\RP\\RP-S-S6\R4S-S$\R4S-S%\R4S-S&\R4S-S'\R4S-S(\R4S-S)\R4S-S=\S-S>\S?\)\\*4S-S@\+\\/S4S-SA\\S\48SBjj55r,SCr-U=r.$)IHiDreamImagePipelinezNtext_encoder->text_encoder_2->text_encoder_3->text_encoder_4->transformer->vae)latentsprompt_embeds_t5prompt_embeds_llama3pooled_prompt_embedsr7vae text_encoder tokenizertext_encoder_2 tokenizer_2text_encoder_3 tokenizer_3text_encoder_4 tokenizer_4 transformerc >[T U]5 URUUUUU UUUU UU S9 [US5(a<URb/S[ URR R5S- -OSUl[URS-S9Ul SUl [USS5b&URRURlgg) N) rCrDrFrHrJrErGrIrKr7rLrCrr)vae_scale_factorrK)super__init__register_moduleshasattrrCr6configblock_out_channelsrOr image_processordefault_sample_sizegetattrrK eos_token pad_token) selfr7rCrDrErFrGrHrIrJrKrLr5s r&rRHiDreamImagePipeline.__init__s  %)))####  CJ$PUBVBV[_[c[c[oA#dhhoo889A= >uv  1$BWBWZ[B[\#& 4 - 9)-)9)9)C)CD   & :r(NrPpromptmax_sequence_lengthr*dtypec XU=(d URnU=(d URRn[U[5(aU/OUnUR US[ X R R5SSSS9nURnURnUR USSS9RnURSURS:a[R"Xh5(d|UR RUSS2[ X R R5S- S245n [RS [ X R R5S U 35 URUR!U5UR!U5S 9S n U R!XCS 9n U $)N max_lengthTptpaddingrb truncationadd_special_tokensreturn_tensorslongestrerhrXThe following part of your input was truncated because `max_sequence_length` is set to tokens: )attention_maskrr`r*)_execution_devicerHr` isinstancestrrIminmodel_max_length input_idsrnshapetorchequal batch_decodeloggerwarningto) r\r^r_r*r` text_inputstext_input_idsrnuntruncated_ids removed_text prompt_embedss r&_get_t5_prompt_embeds*HiDreamImagePipeline._get_t5_prompt_embedss14112,,22'44&&&&  .0@0@0Q0QR# ' %..$33**69UY*Zdd   $(<(UYZ &a( %((u(D r(c U=(d URnU=(d URRn[U[5(aU/OUnUR US[ X R R5SSSS9nURnURnUR USSS9RnURSURS:a[R"Xh5(d|UR RUSS2[ X R R5S- S245n [RS [ X R R5S U 35 URUR!U5UR!U5SSS 9n U R"SSn [R$"U S S 9n U $)NrbTrcrdrirjrkrrlrm)rnroutput_attentionsrdim)rprJr`rqrrrKrsrtrurnrvrwrxryrzr{r| hidden_statesstack) r\r^r_r*r`r}r~rnrroutputsrs r&_get_llama3_prompt_embeds.HiDreamImagePipeline._get_llama3_prompt_embedss14112,,22'44&&&&  .0@0@0Q0QR# ' %..$33**69UY*Zdd   $(<(%>zHH8}q S]j%@ #CJ%>zHH8}q S]j%@ #CJ,E,L,LZYZ\],^)  ')6H%/#%>%>zHH8}q S]j%@ #CJST388F[9[]_` 055'1 q=Z!^/66z1aH  \h*4OPXzZ[ [+2216KQO+00>S1SU\^`a%9$>$>!8Wc q=Z!^#7#>#>q*aQR#S \h*4ST\S]^_ _3::1aAVXYZ388ZJ_=_ahjmn & = C C Hg1}a0M0T0TU_ab0c-A(j"8 #`ai`j!kll,I,P,PQRTi,j ),I,N,Nz\qOqsu,v )$=#B#B Hgq1}a,E,L,LZYZ\],^)A(j"8 #\]e\f!ghh(A(H(HLacd(e %(A(F(FzTiGikrtv(w %)F(K(K %Ax#1}a0M0T0TUVXbdegh0i-A(j"8 #`ai`j!kll,I,P,PQRTUWlno,p ),I,N,NJ!66- )  % ) )   r(c t^UbX[U4SjU55(d>[STRSUVs/sHnUTR;dMUPM sn35eUbU b[SUSU S35eUbU b[SUSU S35eUbU b[SUS U S35eUbU b[S US U S35eUcU c [S 5eUcU c [S 5eUcU c [S5eUbA[U[5(d,[U[ 5(d[S[ U535eUbA[U[5(d,[U[ 5(d[S[ U535eUbA[U[5(d,[U[ 5(d[S[ U535eUbA[U[5(d,[U[ 5(d[S[ U535eUbUb[SUSUS35eUbUb[SUSUS35eUbU b[SUSU S35eUbU b[SUSU S35eU bCUb@U RUR:wa&[SU RSURS35eU bCU b@U RU R:wa&[SU RSU RS35eU bEU bAU RU R:wa&[SU RS U RS35egggs snf)!Nc3@># UHoTR;v M g7fr )_callback_tensor_inputs).0kr\s r& 4HiDreamImagePipeline.check_inputs..sF 7Y!-- -7Ysz2`callback_on_step_end_tensor_inputs` has to be in z , but found zCannot forward both `prompt`: z and `pooled_prompt_embeds`: z2. Please make sure to only forward one of the two.z Cannot forward both `prompt_2`: z Cannot forward both `prompt_3`: z and `prompt_embeds_t5`: z Cannot forward both `prompt_4`: z and `prompt_embeds_llama3`: zsProvide either `prompt` or `pooled_prompt_embeds`. Cannot leave both `prompt` and `pooled_prompt_embeds` undefined.zkProvide either `prompt` or `prompt_embeds_t5`. Cannot leave both `prompt` and `prompt_embeds_t5` undefined.zsProvide either `prompt` or `prompt_embeds_llama3`. Cannot leave both `prompt` and `prompt_embeds_llama3` 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 z4`prompt_3` has to be of type `str` or `list` but is z4`prompt_4` has to be of type `str` or `list` but is z'Cannot forward both `negative_prompt`: z& and `negative_pooled_prompt_embeds`: z)Cannot forward both `negative_prompt_2`: z)Cannot forward both `negative_prompt_3`: z" and `negative_prompt_embeds_t5`: z)Cannot forward both `negative_prompt_4`: z& and `negative_prompt_embeds_llama3`: z`pooled_prompt_embeds` and `negative_pooled_prompt_embeds` must have the same shape when passed directly, but got: `pooled_prompt_embeds` z$ != `negative_pooled_prompt_embeds` .z~`prompt_embeds_t5` and `negative_prompt_embeds_t5` must have the same shape when passed directly, but got: `prompt_embeds_t5` z != `negative_prompt_embeds_t5` z`prompt_embeds_llama3` and `negative_prompt_embeds_llama3` must have the same shape when passed directly, but got: `prompt_embeds_llama3` z$ != `negative_prompt_embeds_llama3` )allr.rrqrrlisttyperv)r\r^rrrrrrrr@rArrrBr"callback_on_step_end_tensor_inputsrs` r& check_inputs!HiDreamImagePipeline.check_inputss$ . 9#F 7YF C C DTEaEaDbbn|^pH|^vwbckokGkGbGpq|^pHoIJ   "6"B08UVjUkl00  !&:&F28*?6A ?"5fZG }}% #A'--P[\a[b!cddjj(Gr(cUR$r _guidance_scaler\s r&guidance_scale#HiDreamImagePipeline.guidance_scales###r(c URS:$)Nrrrs r&r0HiDreamImagePipeline.do_classifier_free_guidances##a''r(cUR$r )_attention_kwargsrs r&attention_kwargs%HiDreamImagePipeline.attention_kwargss%%%r(cUR$r )_num_timestepsrs r& num_timesteps"HiDreamImagePipeline.num_timestepss"""r(cUR$r ) _interruptrs r& interruptHiDreamImagePipeline.interrupts r(2g@pilr?rrr)r,rr output_type return_dictrcallback_on_step_endrc URSS5nURSS5nUbSn [SSU 5 USnUSnUbSn [SSU 5 USnUSnU=(d URUR-nU=(d URUR-nURS -n!URUR-S -n"U"Xe-- n#[R "U#5n#[ UU#-U!-U!-5[ UU#-U!-U!-5pVURUUUUU U U U UUUUUUUS 9 XlUUl S Ul Ub[U[5(aSn$O6Ub![U[5(a [U5n$OUbURSn$UR n%UR"bUR"RS S5OSn&UR%UUUUU U U U UR&UUUUUUU%UUU&S 9unnnnnnUR&(aE[(R*"UU/SS9n[(R*"UU/SS9n[(R*"UU/SS9nUR,R.R0n'UR3W$U-U'UUUR4U%UU5n[7UR,R85n(SU(0n)[:(aSn*OU%n*[UR<[>5(a1UR<RAUU*S9 UR<RBn+O[EUR<UU*4SU0U)D6un+n[G[U+5XpR<RH-- S5n,[U+5Ul%URMUS9n-[OU+5GHun.n/URP(aMUR&(a[(R*"U/S -5OUn0U/RSU0RS5n1UR-U0U1UUUS S9Sn2U2*n2UR&(a)U2RUS 5un3n4U3URVU4U3- --n2UR4n5UR<RYU2U/US S9SnUR4U5:wa>[(RZR\R_5(aURaU55nUbn0n6UHn7[c5U7U6U7'M U"UU.U/U65n8U8ReSU5nU8ReSU5nU8ReSU5nU8ReSU5nU.[U+5S- :Xd)U.S-U,:a0U.S-UR<RH-S:XaU-Rg5 [:(dGM[hRj"5 GM SSS5 US:XaUn9O{UURlR.Rn- URlR.Rp-nURlRsUS S9Sn9URtRwU9US9n9URy5 U(dU94$[{U9S9$!,(df  N=f)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. prompt_3 (`str` or `list[str]`, *optional*): The prompt or prompts to be sent to `tokenizer_3` and `text_encoder_3`. If not defined, `prompt` is will be used instead. prompt_4 (`str` or `list[str]`, *optional*): The prompt or prompts to be sent to `tokenizer_4` and `text_encoder_4`. If not defined, `prompt` is will be used instead. 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 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. negative_prompt (`str` or `list[str]`, *optional*): The prompt or prompts not to guide the image generation. If not defined, one has to pass `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `true_cfg_scale` is not greater than `1`). negative_prompt_2 (`str` or `list[str]`, *optional*): The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and `text_encoder_2`. If not defined, `negative_prompt` is used in all the text-encoders. negative_prompt_3 (`str` or `list[str]`, *optional*): The prompt or prompts not to guide the image generation to be sent to `tokenizer_3` and `text_encoder_3`. If not defined, `negative_prompt` is used in all the text-encoders. negative_prompt_4 (`str` or `list[str]`, *optional*): The prompt or prompts not to guide the image generation to be sent to `tokenizer_4` and `text_encoder_4`. If not defined, `negative_prompt` is used in all the text-encoders. 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 be generated by sampling using the supplied random `generator`. prompt_embeds_t5 (`torch.FloatTensor`, *optional*): Pre-generated T5 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_embeds_llama3 (`torch.FloatTensor`, *optional*): Pre-generated LLaMA3 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_t5 (`torch.FloatTensor`, *optional*): Pre-generated negative T5 text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not provided, embeddings will be generated from `negative_prompt` input argument. negative_prompt_embeds_llama3 (`torch.FloatTensor`, *optional*): Pre-generated negative LLaMA3 text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not provided, embeddings will be generated from `negative_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. negative_pooled_prompt_embeds (`torch.FloatTensor`, *optional*): Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_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. 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 128): Maximum sequence length to use with the `prompt`. Examples: Returns: [`~pipelines.hidream_image.HiDreamImagePipelineOutput`] or `tuple`: [`~pipelines.hidream_image.HiDreamImagePipelineOutput`] if `return_dict` is True, otherwise a `tuple`. When returning a tuple, the first element is a list with the generated. images. rNnegative_prompt_embedszmThe `prompt_embeds` argument is deprecated. Please use `prompt_embeds_t5` and `prompt_embeds_llama3` instead.z0.35.0rrzThe `negative_prompt_embeds` argument is deprecated. Please use `negative_prompt_embeds_t5` and `negative_prompt_embeds_llama3` instead.r) rrrrr@rArrrBrrFscale)r^rrrrrrrrr@rArrrBrr*rr_rrr%cpu)r*r,)total)rr+encoder_hidden_states_t5encoder_hidden_states_llama3 pooled_embedsr)rr?r@rArBlatent)r)images)>getrrXrOmathsqrtrrrrrrqrrrr6rvrprrrrwrrLrU in_channelsrr`r'max_seq XLA_AVAILABLEr7rr2r+r;maxorderr progress_bar enumeraterexpandchunkrstepbackendsmps is_availabler|localspopupdatexm mark_steprCscaling_factor shift_factordecoderW postprocessmaybe_free_model_hooksr):r\r^rrrrrr)r,rrrrrrrr?r@rArrrBrrrrrrr_r8rrdeprecation_messagedivisionS_maxrrr*rrr%scheduler_kwargstimestep_devicer+num_warmup_stepsritlatent_model_inputtimestep noise_prednoise_pred_uncondnoise_pred_text latents_dtypecallback_kwargsrcallback_outputsimages: r&__call__HiDreamImagePipeline.__call__sP ?D9 !',Dd!K  $#R  ox1D E,Q/ #0#3 ! -#m  .:M N(>q(A %,B1,E )K433d6K6KKI11D4I4II((1,))D,A,AAaG( % EEMX5@A3vPU~YaGadlGlCmv     +///-!5&?*G!5*G/Q  $ .!1  *VS"9"9J  Jvt$<$<VJ ! --33A6J''BFAVAVAbT**..w=hl    +///(,(H(H-!5&?*G!5*G"7 3!'    % ) ).  + +$yy*CEU)V\]^ #(99.KMa-bhi#j #(99.KMa-bhi#j  $//66BB&& . .   & &     T--55 6 ": =#O$O dnn&= > > NN ( ()<_ ( U00I-?#. . # . *I*s9~0CnnFZFZ0ZZ\]^!)n  %8  9\!),1>>BFAaAaUYYy1}%=gn"88$6$<$'E#E $$//$6%E &(-'8'84'?'E (!)E *DL+E Z!"&&*!&*+/!kl4$$((&&## ]]_12#'+/+/+/! #%%) #26484848,-DH,0599=>BBF9=BF"' 26BF9B #&;I8d3iI8S /D(I8S /D( I8 S /D( I8 d I8TzI8!I8U d"I8I8tCy4/I8c?T1I8c?T1I8c?T1I8 #TzI8 ??T%//%::TA!I8"""T)#I8$ ++d2%I8&$//$6'I8($)#4#4t#;)I8*(-'8'84'?+I8,$//$6-I8.(-'8'84'?/I804Z1I823I84sCx.4/5I86'Sz4'784?7I88-1I9I8:!;I83I8r(r=)ig?gffffff?)NNNN)3r0rtypingrrrw transformersrrrrr r rWr loadersr modelsrr schedulersrrutilsrrrrutils.torch_utilsrpipeline_utilsrpipeline_outputrtorch_xla.core.xla_modelcore xla_modelrr get_loggerrrzrrrr'rrr*rr;r=r!r(r&r2s 12CRZZ-.7))MM   H %!N       '+(,"&!% 8*t8* %,,  %8*Cy4 8* K$  8*vN 8,.IN 8r(