o Giu@shddlmZmZmZmZmZddlZddlZ ddl Z ddl m Z m Z ddlmZmZddlmZddlmZmZmZddlmZdd lmZmZmZmZdd lmZdd l m!Z!d d l"m#Z#ddl$m%Z%ernddl&m'Z'nGdddZ'erddl(m)m*Z+dZ,ndZ,e-e.Z/de j0de1fddZ2 d#de j0de j3dBde4fddZ5dZ6d Z7Gd!d"d"e#Z8dS)$)CallableDictListOptionalUnionN) AutoTokenizer"Qwen2_5_VLForConditionalGeneration)MultiPipelineCallbacksPipelineCallback)PipelineImageInput)AutoencoderKLWanCosmosControlNetModelCosmosTransformer3DModel)UniPCMultistepScheduler)is_cosmos_guardrail_availableis_torch_xla_availableloggingreplace_example_docstring) randn_tensor)VideoProcessor)DiffusionPipeline)CosmosPipelineOutput)CosmosSafetyCheckerc@seZdZddZdS)rcOstd)Nz|`cosmos_guardrail` is not installed. Please install it to use the safety checker for Cosmos: `pip install cosmos_guardrail`.) ImportError)selfargskwargsr j/home/ubuntu/.local/lib/python3.10/site-packages/diffusers/pipelines/cosmos/pipeline_cosmos2_5_transfer.py__init__&szCosmosSafetyChecker.__init__N)__name__ __module__ __qualname__r"r r r r!r%s rTFvideo num_framesc Cs||jd}|dkr0|ddddddddddf}tj||dd|ddfdd}|S||jdkrJ|ddddd|ddddf}|S)Nrrrdim)shapetorchcatrepeat)r&r' n_pad_frames last_framer r r!_maybe_pad_or_trim_video6s& &r1sampleencoder_output generator sample_modecCsRt|dr|dkr|j|St|dr|dkr|jSt|dr%|jStd)N latent_distr2argmaxlatentsz3Could not access latents of provided encoder_output)hasattrr6r2moder8AttributeError)r3r4r5r r r!retrieve_latentsAs   r<aThe video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.a Examples: ```python >>> import cv2 >>> import numpy as np >>> import torch >>> from diffusers import Cosmos2_5_TransferPipeline, AutoModel >>> from diffusers.utils import export_to_video, load_video >>> model_id = "nvidia/Cosmos-Transfer2.5-2B" >>> # Load a Transfer2.5 controlnet variant (edge, depth, seg, or blur) >>> controlnet = AutoModel.from_pretrained(model_id, revision="diffusers/controlnet/general/edge") >>> pipe = Cosmos2_5_TransferPipeline.from_pretrained( ... model_id, controlnet=controlnet, revision="diffusers/general", torch_dtype=torch.bfloat16 ... ) >>> pipe = pipe.to("cuda") >>> # Video2World with edge control: Generate video guided by edge maps extracted from input video. >>> prompt = ( ... "The video is a demonstration of robotic manipulation, likely in a laboratory or testing environment. It" ... "features two robotic arms interacting with a piece of blue fabric. The setting is a room with a beige" ... "couch in the background, providing a neutral backdrop for the robotic activity. The robotic arms are" ... "positioned on either side of the fabric, which is placed on a yellow cushion. The left robotic arm is" ... "white with a black gripper, while the right arm is black with a more complex, articulated gripper. At the" ... "beginning, the fabric is laid out on the cushion. The left robotic arm approaches the fabric, its gripper" ... "opening and closing as it positions itself. The right arm remains stationary initially, poised to assist." ... "As the video progresses, the left arm grips the fabric, lifting it slightly off the cushion. The right arm" ... "then moves in, its gripper adjusting to grasp the opposite side of the fabric. Both arms work in" ... "coordination, lifting and holding the fabric between them. The fabric is manipulated with precision," ... "showcasing the dexterity and control of the robotic arms. The camera remains static throughout, focusing" ... "on the interaction between the robotic arms and the fabric, allowing viewers to observe the detailed" ... "movements and coordination involved in the task." ... ) >>> negative_prompt = ( ... "The video captures a series of frames showing ugly scenes, static with no motion, motion blur, " ... "over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, " ... "underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky " ... "movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, " ... "fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. " ... "Overall, the video is of poor quality." ... ) >>> input_video = load_video( ... "https://github.com/nvidia-cosmos/cosmos-transfer2.5/raw/refs/heads/main/assets/robot_example/robot_input.mp4" ... ) >>> num_frames = 93 >>> # Extract edge maps from the input video using Canny edge detection >>> edge_maps = [ ... cv2.Canny(cv2.cvtColor(np.array(frame.convert("RGB")), cv2.COLOR_RGB2BGR), 100, 200) ... for frame in input_video[:num_frames] ... ] >>> edge_maps = np.stack(edge_maps)[None] # (T, H, W) -> (1, T, H, W) >>> controls = torch.from_numpy(edge_maps).expand(3, -1, -1, -1) # (1, T, H, W) -> (3, T, H, W) >>> controls = [Image.fromarray(x.numpy()) for x in controls.permute(1, 2, 3, 0)] >>> export_to_video(controls, "edge_controlled_video_edge.mp4", fps=30) >>> # Transfer inference with controls. >>> video = pipe( ... controls=controls, ... controls_conditioning_scale=1.0, ... prompt=prompt, ... negative_prompt=negative_prompt, ... num_frames=num_frames, ... ).frames[0] >>> export_to_video(video, "edge_controlled_video.mp4", fps=30) ``` c2seZdZdZdZgdZdgZdgZ dPdede de d e d e d e deeffd d Z    dQdeeeefdedejdBdejdBfddZ        dRdeeeefdeeeeefdededejdBdejdBdedejdBdejdBfddZ   ! !      "dSd#eejd$ed%ed&ed'ed(ed)ededeejdeejd*eeejeejfd+eejd,ed-ejfd.d/Z       0dTd1d2Zed3d4Z ed5d6Z!ed7d8Z"ed9d:Z#ed;d<Z$e%e&e'd=de(dddd!d>d?dddddd@ddd+gdd0ddfdAe)ee)BdBee*ee*fdeeeefdBdeeeefd&ed'eedCeedDedEedFe*ded*eeejeejfd+eejdeejdeejdGeedHedIeee+eee,gdfe-e.fdJeededKe*dLeedMeef.dNdOZ/Z0S)UCosmos2_5_TransferPipelinea Pipeline for Cosmos Transfer2.5, supporting auto-regressive inference. This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods implemented for all pipelines (downloading, saving, running on a particular device, etc.). Args: text_encoder ([`Qwen2_5_VLForConditionalGeneration`]): Frozen text-encoder. Cosmos Transfer2.5 uses the [Qwen2.5 VL](https://huggingface.co/Qwen/Qwen2.5-VL-7B-Instruct) encoder. tokenizer (`AutoTokenizer`): Tokenizer associated with the Qwen2.5 VL encoder. transformer ([`CosmosTransformer3DModel`]): Conditional Transformer to denoise the encoded image latents. scheduler ([`UniPCMultistepScheduler`]): A scheduler to be used in combination with `transformer` to denoise the encoded image latents. vae ([`AutoencoderKLWan`]): Variational Auto-Encoder (VAE) Model to encode and decode videos to and from latent representations. controlnet ([`CosmosControlNetModel`]): ControlNet used to condition generation on control inputs. z*text_encoder->transformer->controlnet->vae)r8 prompt_embedsnegative_prompt_embedssafety_checkerN text_encoder tokenizer transformervae scheduler controlnetc s.t|dur t}|j|||||||dt|ddr&dt|jjnd|_t|ddr7dt |jjnd|_ t |j d|_ t|jj dddur_t|jj jd|jj jdddnd}t|jj d ddurt|jj jd|jj jdddnd} ||_| |_|jdus|jdurtd dS) N)rDrArBrCrFrEr@rDr)vae_scale_factor latents_meanr latents_stdzDVAE configuration must define both `latents_mean` and `latents_std`.)superr"rregister_modulesgetattrsumrDtemperal_downsamplevae_scale_factor_temporallenvae_scale_factor_spatialrvideo_processorconfigr,tensorrJviewz_dimfloatrK ValueError) rrArBrCrDrErFr@rJrK __class__r r!r"s8 " "((z#Cosmos2_5_TransferPipeline.__init__promptmax_sequence_lengthdevicedtypec CsN|p|j}|p |jj}t|tr|gn|}g}tt|D]<}ddddgddd||dgdg}|jj|ddd|dd d }t|t sMd |vrM|d n|}t |}| |qt j |d d }|j||dd} | j} g} tdt| D]} | | | | jddd| | jdddd} | | qwt j| dd }|j||d}|S)NsystemtextzKYou are a helpful assistant who will provide prompts to an image generator.)typerc)rolecontentuserTF max_length)tokenizeadd_generation_prompt add_vision_idrh truncationpadding input_idsrr))output_hidden_statesrr()r*keepdimg:0yE>rar`)_execution_devicerAra isinstancestrrangerRrBapply_chat_templatelistr, LongTensorappendstackto hidden_statesmeanstdr-)rr^r_r`rainput_ids_batch sample_idx conversationsrnoutputsr|normalized_hidden_states layer_idxnormalized_stater>r r r!_get_prompt_embedss\       z-Cosmos2_5_TransferPipeline._get_prompt_embedsTrnegative_promptdo_classifier_free_guidancenum_videos_per_promptr>r?c CsZ|p|j}t|tr |gn|}|durt|} n|jd} |dur@|j|||| d}|j\} } } |d|d}|| || d}|r|dur|pId}t|trT| |gn|}|durqt|t|urqt dt|dt|d | t|krt d |d t|d |d | d |j|||| d}|j\} } } |d|d}|| || d}||fS)a" Encodes the prompt into text encoder hidden states. Args: prompt (`str` or `List[str]`, *optional*): prompt to be encoded 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 `guidance_scale` is less than `1`). do_classifier_free_guidance (`bool`, *optional*, defaults to `True`): Whether to use classifier free guidance or not. num_videos_per_prompt (`int`, *optional*, defaults to 1): Number of videos that should be generated per prompt. 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. negative_prompt_embeds (`torch.Tensor`, *optional*): Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input argument. device: (`torch.device`, *optional*): torch device dtype: (`torch.dtype`, *optional*): torch dtype Nr)r^r_r`rarr(z?`negative_prompt` should be the same type to `prompt`, but got z != .z`negative_prompt`: z has batch size z, but `prompt`: zT. Please make sure that passed `negative_prompt` matches the batch size of `prompt`.) rrrsrtrRr+rr.rWrd TypeErrorrZ) rr^rrrr>r?r_r`ra batch_size_seq_lenr r r! encode_prompt%sH &       z(Cosmos2_5_TransferPipeline.encode_prompt]rr&rnum_channels_latentsheightwidth num_frames_innum_frames_outr4r8num_cond_latent_framesreturncs<ttrt|krtdtd|d|}|}|djd}|j}|j}|||||f}| dur[| jdd|ddkrStd| jd|d| j| | d} nt|| | d } |d krt j |d|||f| j | j d }t j |d|ddf| j | j d }t | }| |||fSdurtd j| jj dttrfd dt|D}n fddD}t j|d d| }jj| | d}jj| | d}|||}|d|||f}| |}| |}| |d| ddd}d|ddddd | ddddf<||d||}| |||fS)Nz/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.rz Unexpected `latents` shape, got z , expected rr`ra)r4r`rarrqz@`video` must be provided when `num_frames_in` is greater than 0.cs.g|]}tj|d|dqSr)r4r<rDencode unsqueeze.0ir4rr&r r!  z>Cosmos2_5_TransferPipeline.prepare_latents..cs$g|]}tj|dqS)rrrvidr4rr r!rs$r)r?)rsrwrRrZrQrSr+r{rr,zerosrar` zeros_likerDrur-rJrKnew_ones new_zerossize)rr&rrrrrrrrar`r4r8rBCTHWr+ cond_maskcond_indicator cond_latentsrJrK padding_shape ones_padding zeros_paddingr rr!prepare_latentszsb        &z*Cosmos2_5_TransferPipeline.prepare_latents皙?c s\|dks|dks|ddks|ddkrtd|d|d| dur/| dkr/td| d| dks7| dkr?td| d|dur`tfd d |Ds`td jd fd d|D|durs|durstd|d|d|dur|durtd|durt|tst|tstdt||dur|durtd|dur|durtd|dur|dkrtd|dur|dkrtd|durtd|djd}jd} || krt d|d| d| }d} t j j dddurtdd tj j jj j jD} || krtd|d| d||kr,td|d |d!|S)"NrrzE`height` and `width` have to be divisible by 16 (& positive) but are z and rz?`num_frames` has to be a positive integer when provided but is rzM`conditional_frame_timestep` has to be a float in the [0, 1] interval but is c3s|]}|jvVqdSN_callback_tensor_inputsrkrr r! s  z:Cosmos2_5_TransferPipeline.check_inputs..z2`callback_on_step_end_tensor_inputs` has to be in z , but found csg|] }|jvr|qSr rrrr r!rsz;Cosmos2_5_TransferPipeline.check_inputs..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 z`Provide only one of `num_ar_conditional_frames` or `num_ar_latent_conditional_frames`, not both.zQProvide either `num_ar_conditional_frames` or `num_ar_latent_conditional_frames`.z0`num_ar_latent_conditional_frames` must be >= 0.z)`num_ar_conditional_frames` must be >= 0.znum_frames_per_chunk=z# must be larger than min_chunk_len=z, setting to min_chunk_lenmax_sizecss|] \}}||VqdSrr )rrpatchr r r!rs  z3 is too large for RoPE setting (max_frames_by_rope=z(). Please reduce `num_frames_per_chunk`.znum_ar_conditional_frames=z+ must be smaller than num_frames_per_chunk=z for chunked generation.)rZallrrsrtrwrdmaxrQloggerwarningrNrCrUzipr patch_size) rr^rrr>"callback_on_step_end_tensor_inputsnum_ar_conditional_frames num_ar_latent_conditional_framesnum_frames_per_chunkr'conditional_frame_timestep min_chunk_lenmax_frames_by_roper rr! check_inputssx(    z'Cosmos2_5_TransferPipeline.check_inputscC|jSr_guidance_scalerr r r!guidance_scale"z)Cosmos2_5_TransferPipeline.guidance_scalecCs |jdkS)Nrrrr r r!r&s z6Cosmos2_5_TransferPipeline.do_classifier_free_guidancecCrr)_num_timestepsrr r r! num_timesteps*rz(Cosmos2_5_TransferPipeline.num_timestepscCrr)_current_timesteprr r r!current_timestep.rz+Cosmos2_5_TransferPipeline.current_timestepcCrr) _interruptrr r r! interrupt2rz$Cosmos2_5_TransferPipeline.interruptr$g@pilcontrolscontrols_conditioning_scaler'rnum_inference_stepsr output_type return_dictcallback_on_step_endrrrrcJs jdurtd jdt|ttfr|j}|duryt|tr%|dn|}t|tr0|d}t|tj t j frL|j dkrC|d}n |j dkrL|d}t|t jjr`t|d|j|j}n|j d kritd t|d|jd |jd } |||||||| |dur|tdd  jd n tdd  jd | _d _d _ j jdurވ j|durt|tr|gn|}|D]} j|std|dq|durt|trd }n|durt|trt|}n|jd} j|| j| |||d\}} j j!} j"j!}t# j"j$ddrDtj%| j"j$j& j"j$j'|j(|d}| d kr;|j)| dd}||f}||f} n|}|} j*+|||}!|!jd|krx|!jdd krj|!,|d d d d }!ntd|d|!jdd|!jd |durt-|t.|!}!d  jd }"}#fddt/d|#D}$g}% j0j|d j1j|d fdd}&| }'d}(g})t|$}*| |*}+ j2|+d},t3|$D] \}-\}.}/|-dkrtj%|d ||f|d}0 j*+|0||}0n7|%d4}0dkr1|0dddd df|0dddddf<d|0dddddf<n|05d|0j|d}1t.|1}1 j6|1||  j"j$j7d |||1jd  jtj8|-dkrb|(n|'d \} }2}3}4|3|}3t9|4|}5| j:d d |||d}6|!dddd|.|/d fj j j!dt.ttr fd!dt/jdD}7n  fd"dD}7tj;|7dd|}7|7}7 j}8t|8 _?| |2|3}9t3|8D]\}:}; j@rq|;AB _tC j jF|7|=|>||3||6d d$}?|?d}@ j"|=|>||@|3|6d d%d}A|9|Ad |3}A jr jF|7|=|>| |3||6d d$}?|?d}@ j"|=|>| |@|3|6d d%d}B|9|Bd |3}B|A jG|A|B}A j 1`. num_videos_per_prompt (`int`, *optional*, defaults to 1): The number of images to generate per prompt. generator (`torch.Generator` or `List[torch.Generator]`, *optional*): A [`torch.Generator`](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. Can be used to tweak the same generation with different prompts. If not provided, a latents tensor is 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.FloatTensor`, *optional*): Pre-generated negative text embeddings. For PixArt-Sigma this negative prompt should be "". If not provided, negative_prompt_embeds will be generated from `negative_prompt` input argument. output_type (`str`, *optional*, defaults to `"pil"`): The output format of the generated image. Choose between `PIL.Image` or `np.array`. return_dict (`bool`, *optional*, defaults to `True`): Whether or not to return a [`CosmosPipelineOutput`] instead of a plain tuple. callback_on_step_end (`Callable`, `PipelineCallback`, `MultiPipelineCallbacks`, *optional*): A function or a subclass of `PipelineCallback` or `MultiPipelineCallbacks` that is called at the end of each denoising step during the inference. 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`): The maximum number of tokens in the prompt. If the prompt exceeds this length, it will be truncated. If the prompt is shorter than this length, it will be padded. num_ar_conditional_frames (`int`, *optional*, defaults to `1`): Number of frames to condition on subsequent inference loops in auto-regressive inference, i.e. for the second chunk and onwards. Only used if `num_ar_latent_conditional_frames` is `None`. This is only used when auto-regressive inference is performed, i.e. when the number of frames in controls is > num_frames_per_chunk num_ar_latent_conditional_frames (`int`, *optional*): Number of latent frames to condition on subsequent inference loops in auto-regressive inference, i.e. for the second chunk and onwards. Only used if `num_ar_conditional_frames` is `None`. This is only used when auto-regressive inference is performed, i.e. when the number of frames in controls is > num_frames_per_chunk Examples: Returns: [`~CosmosPipelineOutput`] or `tuple`: If `return_dict` is `True`, [`CosmosPipelineOutput`] is returned, otherwise a `tuple` is returned where the first element is a list with the generated images and the second element is a list of `bool`s indicating whether the corresponding generated image contains "not-safe-for-work" (nsfw) content. Nz)You have disabled the safety checker for z. This is in violation of the [NVIDIA Open Model License Agreement](https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license). Please ensure that you are compliant with the license agreement.r)rrrGrr z0`controls` must contain 3D frames in CHW format.rrFz5Cosmos Guardrail detected unsafe text in the prompt: zR. Please ensure that the prompt abides by the NVIDIA Open Model License Agreement.)r^rrrr>r?r`r_img_context_dim_inrr)zExpected controls batch size z% to match prompt batch size, but got rcsg|] }|t|fqSr )min)r start_idx)rrr r!r'sz7Cosmos2_5_TransferPipeline.__call__..rqcs2|}jj|jjjdddd}|S)NrqFrr)rDdecoder{ra)r8r&)r`rJrKrr r!decode_latents0s "z;Cosmos2_5_TransferPipeline.__call__..decode_latents)total)rar() r&rrrrrrrrar`r4rr8.cs.g|]}tj|d|dqSrrr)chunk_control_videor4rr r!rcrcs&g|]}tj|ddqSrrrrr r!rhs)r`)controls_latentsr8timestepencoder_hidden_statescondition_maskconditioning_scale padding_maskr)r|rrblock_controlnet_hidden_statesrrrrr8r>r?latentc8g|]\}}|dkr|ddddddfn|qSrN.r r chunk_idxchunk)rr r!r&np)rgo@crrr r)rr r!rr)frames)\r@rZr\rsr r tensor_inputsrwr,TensorrndarrayndimPILImageintrrr+rrrQrrrrrr{rtcheck_text_safetyrRrrrDrarCrNrUrimg_context_num_tokensrr`repeat_interleaverTpreprocess_videor.rr1rurJrK progress_bar enumerateclonefill_r in_channelsfloat32 ones_likerr-rE set_timesteps timestepsrrcpuitemrVsigmasrrFrsteplocalspoporderupdate XLA_AVAILABLExm mark_steprydetachpostprocess_videoastypeuint8check_video_safetyrrz from_numpypermutemaybe_free_model_hooksr)Jrrrr^rrrr'rrrrr4r8r>r?rrrrr_rrrframe prompt_listpr vae_dtypetransformer_dtype img_contextrneg_encoder_hidden_states control_videonum_latent_frames_per_chunk chunk_stride chunk_idxs video_chunksr latents_arginitial_num_cond_latent_frames latent_chunks num_chunks total_stepsr rrend_idx prev_output chunk_video cond_latentrr cond_timesteprrr gt_velocityrtsigma_t in_latents in_timestepcontrol_outputcontrol_blocks noise_prednoise_pred_negcallback_kwargsrcallback_outputsr& video_batchrlatent_Tr ) rr`r4rJrKrrrrrr!__call__6s t                         2                     $       z#Cosmos2_5_TransferPipeline.__call__r)Nr]NN)NTrNNr]NN) rrrrrTNNNNr)NNNNNNr)1r#r$r%__doc__model_cpu_offload_seqr_optional_components_exclude_from_cpu_offloadrrrr rrrrr"rrtrrr,r`rarboolrr Generatorrrpropertyrrrrrno_gradrEXAMPLE_DOC_STRINGDEFAULT_NEGATIVE_PROMPTr rYrrr r rK __classcell__r r r[r!r=s / I  Y      W V           r=)Nr2)9typingrrrrrnumpyr PIL.Imagerr, transformersrr callbacksr r image_processorr modelsr rr schedulersrutilsrrrrutils.torch_utilsrrTrpipeline_utilsrpipeline_outputrcosmos_guardrailrtorch_xla.core.xla_modelcore xla_modelrr get_loggerr#rrrr1rQrtr<rUrTr=r r r r!sD          E