o pi0@sddlmZddlmZmZmZddlZddlZddlm Z ddl m Z m Z ddlmZmZmZmZmZmZejjGdd d ZeGd d d eZGd d d ee ZdS)) dataclass)OptionalTupleUnionN) ConfigMixinregister_to_config)CommonSchedulerStateFlaxKarrasDiffusionSchedulersFlaxSchedulerMixinFlaxSchedulerOutputadd_noise_commonget_velocity_commonc@sXeZdZUeed<ejed<ejed<dZee ed<e dedejdejfddZ dS)DDPMSchedulerStatecommoninit_noise_sigma timestepsNnum_inference_stepscCs||||dS)Nrrr)clsrrrrrg/home/ubuntu/SoloSpeech/.venv/lib/python3.10/site-packages/diffusers/schedulers/scheduling_ddpm_flax.pycreate,szDDPMSchedulerState.create) __name__ __module__ __qualname__r __annotations__jnpndarrayrrint classmethodrrrrrr#s    rc@seZdZUeed<dS)FlaxDDPMSchedulerOutputstateN)rrrrrrrrrr"1s r"c@seZdZUdZddeDZejed<e ddZ e ddd d d d d dej f de dedededeejdedededejfddZd7deedefddZ d7dedejdee dejfd d!Z "d8ded#e d$edefd%d&Zd9defd'd(Z d:ded)ejde dejd*eejd+edeeeffd,d-Zded.ejd/ejd0ejdejf d1d2Z dedejd/ejd0ejdejf d3d4Z!d5d6Z"d S);FlaxDDPMSchedulera Denoising diffusion probabilistic models (DDPMs) explores the connections between denoising score matching and Langevin dynamics sampling. [`~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/2006.11239 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 (`np.ndarray`, optional): option to pass an array of betas directly to the constructor to bypass `beta_start`, `beta_end` etc. variance_type (`str`): options to clip the variance used when adding noise to the denoised sample. Choose from `fixed_small`, `fixed_small_log`, `fixed_large`, `fixed_large_log`, `learned` or `learned_range`. clip_sample (`bool`, default `True`): option to clip predicted sample between -1 and 1 for numerical stability. 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. cCsg|]}|jqSr)name).0errr WszFlaxDDPMScheduler.dtypecCsdSNTrselfrrr has_state[szFlaxDDPMScheduler.has_stateig-C6?g{Gz?linearN fixed_smallTepsilonnum_train_timesteps beta_startbeta_end beta_schedule trained_betas variance_type clip_sampleprediction_typec Cs | |_dSNr)) r,r1r2r3r4r5r6r7r8r)rrr__init___s zFlaxDDPMScheduler.__init__rreturncCsP|dur t|}tjd|jd}td|jjddd}t j|||dS)N?r:rr) r rrarrayr)arangeconfigr1roundr)r,rrrrrr create_statens zFlaxDDPMScheduler.create_stater#sampletimestepcCs|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)r,r#rDrErrrscale_model_input}s z#FlaxDDPMScheduler.scale_model_inputrrshapecCs8|jj|}td||ddd}|j||dS)a Sets the discrete timesteps used for the diffusion chain. Supporting function to be run before inference. Args: state (`DDIMSchedulerState`): the `FlaxDDPMScheduler` state data class instance. num_inference_steps (`int`): the number of diffusion steps used when generating samples with a pre-trained model. rNr>)rr)rAr1rr@rBreplace)r,r#rrG step_ratiorrrr set_timestepss zFlaxDDPMScheduler.set_timestepsc Cs|jj|}t|dk|jj|dtjd|jd}d|d||jj|}|dur1|jj}|dkr>tj |dd}|S|dkrNt tj |dd}|S|d krZ|jj|}|S|d krit |jj|}|S|d kro|S|d kr|}|jj|} |dd } | | d| |}|S)Nrr r=r:r/g#B ;)a_minfixed_small_log fixed_largefixed_large_loglearned learned_ranger) ralphas_cumprodrwherer?r)betasrAr6cliplog) r,r#tpredicted_variancer6 alpha_prod_talpha_prod_t_prevvariancemin_logmax_logfracrrr _get_variances2 *     zFlaxDDPMScheduler._get_variance model_outputkey return_dictcs|dur tjdtjdkr3jd|jddkr3jjdvr3tj|jddd\ndj j }t dkj j dtj dj d}d|} d|} jjd krk|| d |d } n&jjd krt} njjd kr|d || d } n td jjdjjrt| dd} |d j j| } j jd | | } | | | |}fdd}t dk|tjjj d}||}|s|fSt|dS)a Predict the sample at the previous timestep by reversing the SDE. Core function to propagate the diffusion process from the learned model outputs (most often the predicted noise). Args: state (`DDPMSchedulerState`): the `FlaxDDPMScheduler` state data class instance. model_output (`jnp.ndarray`): direct output from learned diffusion model. timestep (`int`): current discrete timestep in the diffusion chain. sample (`jnp.ndarray`): current instance of sample being created by diffusion process. key (`jax.Array`): a PRNG key. return_dict (`bool`): option for returning tuple rather than FlaxDDPMSchedulerOutput class Returns: [`FlaxDDPMSchedulerOutput`] or `tuple`: [`FlaxDDPMSchedulerOutput`] if `return_dict` is True, otherwise a `tuple`. When returning a tuple, the first element is the sample tensor. Nrr r)rOrP)axisr=r:r0?rD v_predictionzprediction_type given as z? must be one of `epsilon`, `sample` for the FlaxDDPMScheduler.r>csBtjjddd}tjj|jjd}jdd|S)Nr )numr)rGr))rWrc)jaxrandomsplitnormalrGr)r^) split_keynoiser`r_rWr,r#rVrrrandom_variancesz/FlaxDDPMScheduler.step..random_variance) prev_sampler#)rfrgr`lenrGrAr6rrhrrQrRr?r)r8 ValueErrorr7rTrSalphaszerosr")r,r#r_rErDr`rarXrY beta_prod_tbeta_prod_t_prevpred_original_samplepred_original_sample_coeffcurrent_sample_coeffpred_prev_samplermrZrrlrsteps@   *   " zFlaxDDPMScheduler.steporiginal_samplesrkrcCt|j|||Sr9)rr)r,r#rzrkrrrr add_noisezFlaxDDPMScheduler.add_noisecCr{r9)rr)r,r#rDrkrrrr get_velocity%r}zFlaxDDPMScheduler.get_velocitycCs|jjSr9)rAr1r+rrr__len__.szFlaxDDPMScheduler.__len__r9)r)NNr*)#rrr__doc__r _compatiblesrr)rpropertyr-rfloat32r floatstrrrboolr;r rrCrFrrJr^rfArrayrr"ryr|r~rrrrrr$6s       '  Y   r$) dataclassesrtypingrrrflaxrf jax.numpynumpyrconfiguration_utilsrrscheduling_utils_flaxr r r r rrstructrr"r$rrrrs