bik ddlZddlmZmZmZmZmZmZddlZ ddl Z ddl m Z m Z ddlmZmZddlmZddlmZmZddlmZdd lmZmZmZdd lmZdd lmZd d l m!Z!ddl"m#Z#er ddl$m%cm&Z'dZ(ndZ(ejRe*Z+dZ,ddZ- ddee.deee/e j`fdeee.deee1fdZ2Gdde!eZ3y)N)AnyCallableDictListOptionalUnion)T5EncoderModelT5TokenizerFast)MultiPipelineCallbacksPipelineCallback)Mochi1LoraLoaderMixin)AutoencoderKLMochiMochiTransformer3DModel)FlowMatchEulerDiscreteScheduler)is_torch_xla_availableloggingreplace_example_docstring) randn_tensor)VideoProcessor)DiffusionPipeline)MochiPipelineOutputTFa_ Examples: ```py >>> import torch >>> from diffusers import MochiPipeline >>> from diffusers.utils import export_to_video >>> pipe = MochiPipeline.from_pretrained("genmo/mochi-1-preview", torch_dtype=torch.bfloat16) >>> pipe.enable_model_cpu_offload() >>> pipe.enable_vae_tiling() >>> prompt = "Close-up of a chameleon's eye, with its scaly skin changing color. Ultra high resolution 4k." >>> frames = pipe(prompt, num_inference_steps=28, guidance_scale=3.5).frames[0] >>> export_to_video(frames, "mochi.mp4") ``` cN||dz}t|Dcgc] }||z|z  }}|||zz }||z }|||dzzz }||z d|z|dzz z }||dzz} t||Dcgc]}||dzz||zz| z} }|| z} | D cgc]} d| z  } } | Scc}wcc}wcc} w)Nr?)range) num_stepsthreshold_noise linear_stepsilinear_sigma_schedulethreshold_noise_step_diffquadratic_stepsquadratic_coef linear_coefconstquadratic_sigma_schedulesigma_schedulexs c/home/cdr/jupyterlab/.venv/lib/python3.12/site-packages/diffusers/pipelines/mochi/pipeline_mochi.pylinear_quadratic_scheduler,As A~ IN|I\]AQ0<?]] ,/J J,.O.,RSAS2STN!L017P3PTcefTf3ggK lAo .ECHW`Ca >?!Q$+/1E9  +-EEN'56!cAg6N6 ^  7sB%B B"num_inference_stepsdevice timestepssigmasc | | td|dttj|jj j v}|std|jd|jd ||d||j}t|}||fS|dttj|jj j v}|std|jd|jd ||d||j}t|}||fS|j|fd |i||j}||fS) a Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`. Args: scheduler (`SchedulerMixin`): The scheduler to get timesteps from. num_inference_steps (`int`): The number of diffusion steps used when generating samples with a pre-trained model. If used, `timesteps` must be `None`. device (`str` or `torch.device`, *optional*): The device to which the timesteps should be moved to. If `None`, the timesteps are not moved. timesteps (`List[int]`, *optional*): Custom timesteps used to override the timestep spacing strategy of the scheduler. If `timesteps` is passed, `num_inference_steps` and `sigmas` must be `None`. sigmas (`List[float]`, *optional*): Custom sigmas used to override the timestep spacing strategy of the scheduler. If `sigmas` is passed, `num_inference_steps` and `timesteps` must be `None`. Returns: `Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the second element is the number of inference steps. zYOnly one of `timesteps` or `sigmas` can be passed. Please choose one to set custom valuesr/zThe current scheduler class zx's `set_timesteps` does not support custom timestep schedules. Please check whether you are using the correct scheduler.)r/r.r0zv's `set_timesteps` does not support custom sigmas schedules. Please check whether you are using the correct scheduler.)r0r.r.) ValueErrorsetinspect signature set_timesteps parameterskeys __class__r/len) schedulerr-r.r/r0kwargsaccepts_timesteps accept_sigmass r+retrieve_timestepsr@Ss>!3tuu'3w/@/@AXAX/Y/d/d/i/i/k+ll .y/B/B.CDab  M)FMfM'' !)n ) ))   C(9(9):Q:Q(R(](](b(b(d$ee .y/B/B.CD_`  GvfGG'' !)n ) ))  3MFMfM'' ) ))c-TeZdZdZdZgZgdZ d:dedede de de d e f fd Z d;deeeefdededeej(deej*f dZ dd!Zed"Z ed#Z!ed$Z"ed%Z#ed&Z$ed'Z%ejLe'e(d d d d d(d)d d*d d d d d d d d+dd d d,gd fdeeeefdeeeeefd-eed.eed/ed0ed1eed2e)deed3eeejTeejTfd,eej.deej.deej.deej.deej.d4eed5e d6ee+ee,fd7ee-eee+gd fd8eedef*d9Z.xZ/S)? MochiPipelinea The mochi pipeline for text-to-video generation. Reference: https://github.com/genmoai/models Args: transformer ([`MochiTransformer3DModel`]): Conditional Transformer architecture to denoise the encoded video latents. scheduler ([`FlowMatchEulerDiscreteScheduler`]): A scheduler to be used in combination with `transformer` to denoise the encoded image latents. vae ([`AutoencoderKLMochi`]): Variational Auto-Encoder (VAE) Model to encode and decode videos to and from latent representations. text_encoder ([`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 (`T5TokenizerFast`): Second Tokenizer of class [T5TokenizerFast](https://huggingface.co/docs/transformers/en/model_doc/t5#transformers.T5TokenizerFast). ztext_encoder->transformer->vae)latents prompt_embedsnegative_prompt_embedsr<vae text_encoder tokenizer transformerforce_zeros_for_empty_promptcXt||j|||||d|_d|_d|_t |j|_t|dr"|j|jjnd|_ d|_ d |_ |j| y) N)rGrHrIrJr<r)vae_scale_factorrIiiP)rK)super__init__register_modulesvae_spatial_scale_factorvae_temporal_scale_factor patch_sizervideo_processorhasattrrImodel_max_lengthtokenizer_max_lengthdefault_height default_widthregister_to_config)selfr<rGrHrIrJrKr:s r+rRzMochiPipeline.__init__s  %#  )*%)*&-t?\?\]/6t[/IdnnNhDNN + +nq !"  =YZrANrrPpromptnum_videos_per_promptmax_sequence_lengthr.dtypecN|xs |j}|xs|jj}t|tr|gn|}t |}|j |d|ddd}|j}|j} | jj|} |jjrJ|dk(s|ddk(r=tj||}tj| tj|} |j |d d j} | jd|jdk\rXtj || sB|j j#| dd|d z df} t$j'd |d | |j|j|| d} | j||} | j\} }} | j)d |d } | j+||z|d} | j+|d} | j)|d } | | fS)N max_lengthTpt)paddingrd truncationadd_special_tokensreturn_tensors)r.)rbr.longest)rfrirzXThe following part of your input was truncated because `max_sequence_length` is set to z tokens: )attention_maskr)_execution_devicerHrb isinstancestrr;rI input_idsrmbooltoconfigrKtorch zeros_likeshapeequal batch_decodeloggerwarningrepeatview)r^r_r`rar.rb batch_size text_inputstext_input_idsprompt_attention_maskuntruncated_ids removed_textrE_seq_lens r+_get_t5_prompt_embedsz#MochiPipeline._get_t5_prompt_embedss:14110**00'4&&[ nn  *# % %.. + : : 5 : : < ? ? G ;; 3 32PRWYIY"--nVLN$)$4$45JRWR\R\ek$l !..SW.Xbb   $(<(>66qJ]`aJadfJfGf7ghL NN'( ,A  )).*;*;F*CTi)jklm %((uV(D &++ 7A%,,Q0EqI %**:8M+MwXZ[ 5 : ::r J 5 < <=RTU V333rATnegative_promptdo_classifier_free_guidancerErFrnegative_prompt_attention_maskc | xs |j} t|tr|gn|}| t|} n|jd} ||j ||| | | \}}|r||xsd}t|tr| |gzn|}|:t |t |ur$tdt |dt |d| t|k7r!td|dt|d |d| d |j ||| | | \}}||||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 r)r_r`rar.rbrjz?`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`.) rnrorpr;rwrtype TypeErrorr3) r^r_rrr`rErFrrrar.rbr~s r+ encode_promptzMochiPipeline.encode_promptsP1411'4&&  VJ&,,Q/J  373M3M&;$7 4N4 0M0 '+A+I-3O@J?\_@`jO+<z-MochiPipeline.check_inputs..cs#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.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 zEMust provide `prompt_attention_mask` when specifying `prompt_embeds`.zWMust provide `negative_prompt_attention_mask` when specifying `negative_prompt_embeds`.zu`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but got: `prompt_embeds` z != `negative_prompt_embeds` z`prompt_attention_mask` and `negative_prompt_attention_mask` must have the same shape when passed directly, but got: `prompt_attention_mask` z% != `negative_prompt_attention_mask` )r3allrrorplistrrw) r^r_heightwidth"callback_on_step_end_tensor_inputsrErFrrrs ` r+ check_inputszMochiPipeline.check_inputsUs7 A:?eai1nVW]V^^cdicjjklm m - 9#F 7YF C DTEaEaDbbn|^pHvwbckokGkGbGpqpHoIJ   -";08N}o^00 ^ 5w  FC)@TZ\`IaQRVW]R^Q_`a a  $)>)Fde e ! -2P2Xvw w  $)?)K""&<&B&BB --:-@-@,AB.445Q8 %**.L.R.RR 55J5P5P4QR6<<=Q@S*L $+pHs E3%E3c8|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)rGenable_slicingr^s r+enable_vae_slicingz MochiPipeline.enable_vae_slicing !rAc8|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)rGdisable_slicingrs r+disable_vae_slicingz!MochiPipeline.disable_vae_slicings   "rAc8|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)rG enable_tilingrs r+enable_vae_tilingzMochiPipeline.enable_vae_tilings  rAc8|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)rGdisable_tilingrs r+disable_vae_tilingz MochiPipeline.disable_vae_tilingrrAc r||jz}||jz}|dz |jzdz}|||||f} | | j||St|tr)t ||k7rt dt |d|dt| ||tj} | j|} | S)Nr)r.rbz/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.) generatorr.rb) rTrUrsrorr;r3rrufloat32) r^r~num_channels_latentsrr num_framesrbr.rrDrws r+prepare_latentszMochiPipeline.prepare_latentss4888666 1n)G)GG!K 1:vuM  ::V5:9 9 i &3y>Z+GA#i.AQR&<'gi  u &PUP]P]^**U#rAc|jSr_guidance_scalers r+guidance_scalezMochiPipeline.guidance_scales###rAc |jdkDS)Nrrrs r+rz)MochiPipeline.do_classifier_free_guidances##c))rAc|jSr)_num_timestepsrs r+ num_timestepszMochiPipeline.num_timestepss"""rAc|jSr)_attention_kwargsrs r+attention_kwargszMochiPipeline.attention_kwargs%%%rAc|jSr)_current_timesteprs r+current_timestepzMochiPipeline.current_timesteprrAc|jSr) _interruptrs r+ interruptzMochiPipeline.interrupts rA@g@pilrDrrrr-r/rr output_type return_dictrcallback_on_step_endrc d t|ttfr |j}|xs |j}|xs |j }|j ||||| || |||_||_d|_ d|_ |t|trd}n-|t|tr t|}n| jd}|j}|j!|||j"| | || ||| \} } }}|j$j&j(}|j+|| z||||| j,|| | } |j"r2t/j0|| gd} t/j0|| gd} d}t3||}t5j6|}t9|j:||||\}}t=t|||j:j>zz d}t||_ |jC| 5}tE|D]_\}}|jFrd |z |_ |j"rt/j0| gd zn| }|jI|jdjK| j,} |j$jMd 5|j%|| | | |d d}!ddd!jKt.jN}!|j"r)|!jQd \}"}#|"|jR|#|"z zz}!| j,}$|j:jU|!|| jKt.jNdd} | jK|$} | j,|$k7r9t.jVjXj[r| jK|$} |Hi}%|D]}&t]|&|%|&<|||||%}'|'j_d| } |'j_d| } |t|dz k(s'|dz|kDr/|dz|j:j>zdk(r|jatbsLtejfb dddd|_ |dk(r| }(nti|jjj&dxr"|jjj&jldu})ti|jjj&dxr"|jjj&jndu}*|)r|*rt/jp|jjj&jljsdddddjK| jt| j,}+t/jp|jjj&jnjsdddddjK| jt| j,},| |,z|jjj&jvz |+z} n#| |jjj&jvz } |jjjy| dd}(|jzj}|(|}(|j|s|(fSt|(S#1swYxYw#1swY3xYw)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. height (`int`, *optional*, defaults to `self.default_height`): The height in pixels of the generated image. This is set to 480 by default for the best results. width (`int`, *optional*, defaults to `self.default_width`): The width in pixels of the generated image. This is set to 848 by default for the best results. num_frames (`int`, defaults to `19`): The number of video frames to generate 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. 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`, defaults to `4.5`): 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_videos_per_prompt (`int`, *optional*, defaults to 1): The number of videos 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 ge 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. prompt_attention_mask (`torch.Tensor`, *optional*): Pre-generated attention mask for text embeddings. 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. negative_prompt_attention_mask (`torch.FloatTensor`, *optional*): Pre-generated attention mask for negative text embeddings. 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.mochi.MochiPipelineOutput`] 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 `256`): Maximum sequence length to use with the `prompt`. Examples: Returns: [`~pipelines.mochi.MochiPipelineOutput`] or `tuple`: If `return_dict` is `True`, [`~pipelines.mochi.MochiPipelineOutput`] is returned, otherwise a `tuple` is returned where the first element is a list with the generated images. )r_rrrrErFrrNFrr) r_rrr`rErFrrrar.)dimg?)totalir cond_uncond) hidden_statesencoder_hidden_statestimestepencoder_attention_maskrr)rrDrElatent latents_mean latents_std )r)frames)Aror r tensor_inputsr[r\rrrrrrprr;rwrnrrrJrt in_channelsrrbrucatr,nparrayr@r<maxorderr progress_bar enumeraterexpandrs cache_contextrchunkrstepbackendsmps is_availablelocalspopupdate XLA_AVAILABLExm mark_steprXrGrrtensorr}r.scaling_factordecoderWpostprocess_videomaybe_free_model_hooksr)-r^r_rrrrr-r/rr`rrDrErrFrrrrrrrar~r.rrr0num_warmup_stepsrr!tlatent_model_inputr noise_prednoise_pred_uncondnoise_pred_text latents_dtypecallback_kwargsrcallback_outputsvideohas_latents_meanhas_latents_stdrrs- r+__call__zMochiPipeline.__call__sUH *-=?U,V W1E1S1S ..4..+++ /Q'#9"7+I   .!1!%  *VS"9J  Jvt$<VJ&,,Q/J''   +(,(H(H"7'#9"7+I 3    ! " * $//66BB&& . .             + +!II'=}&MSTUM$)II/MOd.ekl$m ! *+>P&!); NN     * & &s9~0CdnnFZFZ0ZZ\]^!)n  %8  94 #\!),3 #1>>*.&AEAaAaUYYy1}%=gn"88$6$<$CCAr1aQRSVVW^WeWegngtgtuLL!Ql!Ql&Ql $Qlp,0#"'+3j"#!""<$$**##&&&&U]]_12)-;? $##%# #/0MQ*.048<9=AE%* 59KO9B #&-J1c49n%J1"%T#Y"78J1 J1 } J1  J1!J19J1J1 (}J1E%//43H"HIJJ1%,,'J1  -J1 ( 5J1!) 6J1 )1(>!J1"c]#J1$%J1&#4S>2'J1('xc40@$0F'GH)J1*-1I+J1,!-J13J1rArCr)NNNN)4r5typingrrrrrrnumpyrru transformersr r callbacksr r loadersrmodelsrr schedulersrutilsrrrutils.torch_utilsrrWrpipeline_utilsrpipeline_outputrtorch_xla.core.xla_modelcore xla_modelrr get_loggerrrzrr,rrpr.rr@rCr2rAr+rs== 8A,A9 .-.0))MM   H %$(*.15%)$( 8*!#8* U3 ,- .8*S " 8* T%[ ! 8*v[ 1%'<[ 1rA