wiB3ddlmZddlmZmZmZddlZddlmZ ddl m Z m Z ddl mZmZmZmZmZmZejjGdd ZeGd d eZGd d ee ZdS)) dataclass)OptionalTupleUnionN) ConfigMixinregister_to_config)CommonSchedulerStateFlaxKarrasDiffusionSchedulersFlaxSchedulerMixinFlaxSchedulerOutputadd_noise_commonget_velocity_commonc eZdZUeed<ejed<ejed<ejed<dZee ed<e dedejdejdejfdZ dS)DDIMSchedulerStatecommonfinal_alpha_cumprodinit_noise_sigma timestepsNnum_inference_stepsc |||||S)Nrrrr)clsrrrrs y/root/.openclaw/workspace/chatterbox_venv_py311/lib/python3.11/site-packages/diffusers/schedulers/scheduling_ddim_flax.pycreatezDDIMSchedulerState.create-s)s 3-     ) __name__ __module__ __qualname__r __annotations__jnpndarrayrrint classmethodrrrrrr#s    $$$k!!!{)-#---  $  ![  +  ;    [    rrceZdZUeed<dS)FlaxDDIMSchedulerOutputstateN)rr r!rr"rrrr(r(=s rr(c:eZdZUdZdeDZejed<e dZ e ddddd d d d d d ej f de dedededeejdededede dedejfdZd0deedefdZ d0dedejdee dejfdZ d1ded!e d"edefd#Zdefd$Z d2ded&ejde dejd'ed(edeeeffd)Zded*ejd+ejd,ejdejf d-Zdedejd+ejd,ejdejf d.Zd/Z d S)3FlaxDDIMSchedulerav Denoising diffusion implicit models is a scheduler that extends the denoising procedure introduced in denoising diffusion probabilistic models (DDPMs) with non-Markovian guidance. [`~ConfigMixin`] takes care of storing all config attributes that are passed in the scheduler's `__init__` function, such as `num_train_timesteps`. They can be accessed via `scheduler.config.num_train_timesteps`. [`SchedulerMixin`] provides general loading and saving functionality via the [`SchedulerMixin.save_pretrained`] and [`~SchedulerMixin.from_pretrained`] functions. For more details, see the original paper: https://arxiv.org/abs/2010.02502 Args: num_train_timesteps (`int`): number of diffusion steps used to train the model. beta_start (`float`): the starting `beta` value of inference. beta_end (`float`): the final `beta` value. beta_schedule (`str`): the beta schedule, a mapping from a beta range to a sequence of betas for stepping the model. Choose from `linear`, `scaled_linear`, or `squaredcos_cap_v2`. trained_betas (`jnp.ndarray`, optional): option to pass an array of betas directly to the constructor to bypass `beta_start`, `beta_end` etc. clip_sample (`bool`, default `True`): option to clip predicted sample between for numerical stability. The clip range is determined by `clip_sample_range`. clip_sample_range (`float`, default `1.0`): the maximum magnitude for sample clipping. Valid only when `clip_sample=True`. set_alpha_to_one (`bool`, default `True`): each diffusion step uses the value of alphas product at that step and at the previous one. For the final step there is no previous alpha. When this option is `True` the previous alpha product is fixed to `1`, otherwise it uses the value of alpha at step 0. steps_offset (`int`, default `0`): An offset added to the inference steps, as required by some model families. prediction_type (`str`, default `epsilon`): indicates whether the model predicts the noise (epsilon), or the samples. One of `epsilon`, `sample`. `v-prediction` is not supported for this scheduler. dtype (`jnp.dtype`, *optional*, defaults to `jnp.float32`): the `dtype` used for params and computation. cg|] }|j Sr)name).0es r zFlaxDDIMScheduler.isBBBqAFBBBrdtypecdS)NTrselfs r has_statezFlaxDDIMScheduler.has_statemstrig-C6?g{Gz?linearNT?repsilonnum_train_timesteps beta_startbeta_end beta_schedule trained_betas clip_sampleclip_sample_rangeset_alpha_to_one steps_offsetprediction_typec | |_dSNr1) r4r9r:r;r<r=r>r?r@rArBr1s r__init__zFlaxDDIMScheduler.__init__qs rrreturnc||tj|}|jjrt jd|jn |jd}t jd|j}t jd|jj  ddd}t||||S)Nr7rErr) r rconfigr@r#arrayr1alphas_cumprodaranger9roundr)r4rrrrs r create_statezFlaxDDIMScheduler.create_states >)066F15 0L jCIc , , , ,RXRghiRj  9S ;;;Jq$+"ABBHHJJ44R4P !(( 3- )   rr)sampletimestepc|S)a Args: state (`PNDMSchedulerState`): the `FlaxPNDMScheduler` state data class instance. sample (`jnp.ndarray`): input sample timestep (`int`, optional): current timestep Returns: `jnp.ndarray`: scaled input sample r)r4r)rPrQs rscale_model_inputz#FlaxDDIMScheduler.scale_model_inputs  rrrshapec|jj|z}tjd||zddd|jjz}|||S)a Sets the discrete timesteps used for the diffusion chain. Supporting function to be run before inference. Args: state (`DDIMSchedulerState`): the `FlaxDDIMScheduler` state data class instance. num_inference_steps (`int`): the number of diffusion steps used when generating samples with a pre-trained model. rNrI)rr)rJr9r#rMrNrAreplace)r4r)rrT step_ratiors r set_timestepszFlaxDDIMScheduler.set_timestepssr[48KK Z#677*DKKMMddPRdSVZVaVnn }} 3   rc|jj|}tj|dk|jj||j}d|z }d|z }||z d||z z z}|S)Nrr )rrLr#wherer) r4r)rQ prev_timestep alpha_prod_talpha_prod_t_prev beta_prod_tbeta_prod_t_prevvariances r _get_variancezFlaxDDIMScheduler._get_variancess|28< I Q  ;M JELe  ,& 00${2qclipr?rar()r4r)rcrQrPrdrer[rLrr\r]r^pred_original_sample pred_epsilonr` std_dev_tpred_sample_directionrhs rstepzFlaxDDIMScheduler.steps44  $ ,s  !4;#BeF_#__ 4#7&h/ Imq&8.:WYlmm,&  ; &) 3 3$*[S-AL-P$PT`ehTi#i 'LL [ (H 4 4#/ "\c%:=Q%QQU`ehUiiLL [ (N : :$0#$5#?;PSCSWcBc#c (#-=cAQU[@[L"DK,G"""  ; " #7#<#<.. 0M$$ %%eX}EE(s++ "#%6!6A!E3 OR^ ^(C03GGJ__  (' '&;eLLLLroriginal_samplesnoiserc0t|j|||SrD)rr)r4r)rprqrs r add_noisezFlaxDDIMScheduler.add_noise's  .>yQQQrc0t|j|||SrD)rr)r4r)rPrqrs r get_velocityzFlaxDDIMScheduler.get_velocity0s#5< JJJrc|jjSrD)rJr9r3s r__len__zFlaxDDIMScheduler.__len__9s {..rrD)r)rbT)!rr r!__doc__r _compatiblesr#r1r"propertyr5r float32r%floatstrrr$boolrFr rrOrSrrXrarr(rorsrurwrrrr+r+Bs8$$LCB$ABBBL 9 X$("%/3 #&!%(;       , !y   8,@#A M_    2Y]  ' 14 HPQT       SU  ' >A JO     , #5    $ [M[M![Mk[M [M  [M  [M[M &- .[M[M[M[MzR!R+R{ R ; R  RRRRK!K K{ K ; K  KKKK/////rr+) dataclassesrtypingrrrflax jax.numpynumpyr#configuration_utilsrr scheduling_utils_flaxr r r rrrstructrr(r+rrrrsl$"!!!!!)))))))))) AAAAAAAA        2 1 x/x/x/x/x/*Kx/x/x/x/x/r