o ,ic&@sddlmZmZmZmZmZddlZddlmZddlm Z m Z m Z m Z m Z mZmZmZmZmZmZmZmZmZmZddgZGdddeZd d ed ed e d e d e_d eedeedeedeedeedeedededededededededededef"ddZd eedeedeedeedeedeedededededededededededef"d d!Ze ed" #  # # # #d&d eedeedeedeedeedeeded$eededededededededededef$d%dZdS)')castListOptionalTupleUnionN)Tensor)_capturable_doc_default_to_fused_or_foreach_differentiable_doc_disable_dynamo_if_unsupported_dispatch_sqrt _foreach_doc!_get_capturable_supported_devices_get_scalar_dtype _get_value _maximize_doc_stack_if_compiling_use_grad_for_differentiable _view_as_real OptimizerParamsTNAdamnadamcseZdZ      ddddddd ed ed eeefd ed edededeedededeffddZfddZ ddZ e dddZ Z S)rMb`?g?g+?:0yE>rMbp?FN)foreachmaximize capturabledifferentiableparamslrbetaseps weight_decaymomentum_decaydecoupled_weight_decayrrr r!c sd|ks td|d|kstd|d|dkr"dks,ntd|dd|dkr8dksBntd|dd|ksMtd |d|ksXtd |t||||||| || | d } t|| dS) NzInvalid learning rate: zInvalid epsilon value: r?z#Invalid beta parameter at index 0: rz#Invalid beta parameter at index 1: zInvalid weight_decay value: zInvalid momentum_decay value: ) r#r$r%r&r'r(rrr r!) ValueErrordictsuper__init__) selfr"r#r$r%r&r'r(rrr r!defaults __class__O/home/ubuntu/SoloSpeech/.venv/lib/python3.10/site-packages/torch/optim/nadam.pyr.s2 zNAdam.__init__cst||jD]|}|dd|dd|dd|dd|dd|dD]W}|j|g}t|dkrt|d sat |d }|drWtj |t |j d ntj |t d |d <t|d s|d }|drztj |t |j d ntj |t d |d <q-q dS) NrFrr r!r(r"rstepdtypedevicer7 mu_product) r- __setstate__ param_groups setdefaultstategetlentorch is_tensorfloattensorrr8)r/r>grouppp_statestep_val mu_prod_valr1r3r4r;Es:            zNAdam.__setstate__c Cs*d}|dD]} | jdur|t| O}|| | jjr!td|| j|j| } t| dkrv|dr@tjdt | j dntj dt d | d <|drXtj dt | j dntj d t d | d <tj | tjd | d<tj | tjd | d<|| d|| d|| d || d q|S)NFr"z'NAdam does not support sparse gradientsrr r3r6r)r9r5r*r:) memory_formatexp_avg exp_avg_sq)gradrA is_complexappend is_sparse RuntimeErrorr>r@zerosrr8rDones zeros_likepreserve_format) r/rEparams_with_gradgradsexp_avgs exp_avg_sqs mu_products state_steps has_complexrFr>r3r3r4 _init_groupcs<        zNAdam._init_groupc Cs|d}|dur!t |}Wdn1swY|jD]N}g}g}g}g}g}g} ttttf|d\} } |||||||| } t|||||| | | |d|d|d|d|d|d|d |d |d | d q$|S) zPerforms a single optimization step. Args: closure (Callable, optional): A closure that reevaluates the model and returns the loss. Nr$r#r&r'r%rr(rr r!) beta1beta2r#r&r'r%rr(rr r!r\) _cuda_graph_capture_health_checkrA enable_gradr<rrrCr]r) r/closurelossrErVrWrXrYrZr[r^r_r\r3r3r4r5sX   z NAdam.step)rrrrrFN)__name__ __module__ __qualname__rrCrboolrr.r;r]rr5 __classcell__r3r3r1r4rsN       )2a Implements NAdam algorithm. .. math:: \begin{aligned} &\rule{110mm}{0.4pt} \\ &\textbf{input} : \gamma_t \text{ (lr)}, \: \beta_1,\beta_2 \text{ (betas)}, \: \theta_0 \text{ (params)}, \: f(\theta) \text{ (objective)} \\ &\hspace{13mm} \: \lambda \text{ (weight decay)}, \:\psi \text{ (momentum decay)} \\ &\hspace{13mm} \: \textit{decoupled\_weight\_decay}, \:\textit{maximize} \\ &\textbf{initialize} : m_0 \leftarrow 0 \text{ ( first moment)}, v_0 \leftarrow 0 \text{ ( second moment)} \\[-1.ex] &\rule{110mm}{0.4pt} \\ &\textbf{for} \: t=1 \: \textbf{to} \: \ldots \: \textbf{do} \\ &\hspace{5mm}\textbf{if} \: \textit{maximize}: \\ &\hspace{10mm}g_t \leftarrow -\nabla_{\theta} f_t (\theta_{t-1}) \\ &\hspace{5mm}\textbf{else} \\ &\hspace{10mm}g_t \leftarrow \nabla_{\theta} f_t (\theta_{t-1}) \\ &\hspace{5mm} \theta_t \leftarrow \theta_{t-1} \\ &\hspace{5mm} \textbf{if} \: \lambda \neq 0 \\ &\hspace{10mm}\textbf{if} \: \textit{decoupled\_weight\_decay} \\ &\hspace{15mm} \theta_t \leftarrow \theta_{t-1} - \gamma \lambda \theta_{t-1} \\ &\hspace{10mm}\textbf{else} \\ &\hspace{15mm} g_t \leftarrow g_t + \lambda \theta_{t-1} \\ &\hspace{5mm} \mu_t \leftarrow \beta_1 \big(1 - \frac{1}{2} 0.96^{t \psi} \big) \\ &\hspace{5mm} \mu_{t+1} \leftarrow \beta_1 \big(1 - \frac{1}{2} 0.96^{(t+1)\psi}\big)\\ &\hspace{5mm}m_t \leftarrow \beta_1 m_{t-1} + (1 - \beta_1) g_t \\ &\hspace{5mm}v_t \leftarrow \beta_2 v_{t-1} + (1-\beta_2) g^2_t \\ &\hspace{5mm}\widehat{m_t} \leftarrow \mu_{t+1} m_t/(1-\prod_{i=1}^{t+1}\mu_i)\\[-1.ex] & \hspace{11mm} + (1-\mu_t) g_t /(1-\prod_{i=1}^{t} \mu_{i}) \\ &\hspace{5mm}\widehat{v_t} \leftarrow v_t/\big(1-\beta_2^t \big) \\ &\hspace{5mm}\theta_t \leftarrow \theta_t - \gamma \widehat{m_t}/ \big(\sqrt{\widehat{v_t}} + \epsilon \big) \\ &\rule{110mm}{0.4pt} \\[-1.ex] &\bf{return} \: \theta_t \\[-1.ex] &\rule{110mm}{0.4pt} \\[-1.ex] \end{aligned} For further details regarding the algorithm we refer to `Incorporating Nesterov Momentum into Adam`_. a Args: params (iterable): iterable of parameters to optimize or dicts defining parameter groups lr (float, optional): learning rate (default: 2e-3) betas (Tuple[float, float], optional): coefficients used for computing running averages of gradient and its square (default: (0.9, 0.999)) eps (float, optional): term added to the denominator to improve numerical stability (default: 1e-8) weight_decay (float, optional): weight decay (L2 penalty) (default: 0) momentum_decay (float, optional): momentum momentum_decay (default: 4e-3) decoupled_weight_decay (bool, optional): whether to use decoupled weight decay as in AdamW to obtain NAdamW (default: False) z z .. _Incorporating Nesterov Momentum into Adam: https://openreview.net/forum?id=OM0jvwB8jIp57ZJjtNEZ .. _Decoupled Weight Decay Regularization: https://arxiv.org/abs/1711.05101 r"rWrXrYrZr[r^r_r#r&r'r%r(rr r!r\c CsZt|D]%\}}| s||n|| }||}||}||}||}t|r=t|}t|}t|}t|}tjsg|rgt}|jj|jjkrW|jjkr_nn|jj|vsgJd|d|d7}|rp|}nt |}d||}| dkr| r| d|| n|j || d}|ddd|| }|ddd|d| }||9}| |d|| |j ||d|d ||}|s|r| | }||}|| d|d|}|| |d|}||||||qt ||}|| |j||| d|dt |d |j||| |d|d qdS) NzVIf capturable=True, params, mu_products and state_steps must be on supported devices: .rralphar*?Q?)value) enumeraterArN view_as_real_utils is_compilingrr8typermul_addlerp_addcmul_divsqrtaddcdiv_add_)r"rWrXrYrZr[r^r_r#r&r'r%r(rr r!r\iparamrMrKrLr:step_tcapturable_supported_devicesr5bias_correction2mumu_nextdenommu_product_nextr3r3r4_single_tensor_nadamsd     $     rc "s~t|dkrdS|rJdtjs1|r1tddtfddt|||Ds1Jddt||||||g}| D]{\\}}}}}}}|rTt ||||| r[t |}|dj rntj |tjd d d d d nt |d | dkr| rt|d | n| rtj ||| d ntj||| d }t||d t|t|||d t|}|r t|}td|}t|dt |d t|t |td|}t|dt |d t|~t|}t|d t|t|nfdd|D}fdd|D}fdd|D}t||t||t || ~|rt|d t|t|d }t|t|||}~t||}t|t|d t|||} ~t||}!t|!| |t||!|q@tfddt||D}tfddt||D} t||||t|||| q@dS)Nrz#_foreach ops don't support autogradF) supports_xlac3sF|]\}}}|jj|jjko|jjkno|jjvVqdSrd)r8rt).0rFmpr5)rr3r4 s $  z&_multi_tensor_nadam..zWIf capturable=True, params, mu_products, and state_steps must be on supported devices: rjr*cpu)r8rkrrngcs g|] }tdt|qS)r)r rrr5)r_r3r4 sz'_multi_tensor_nadam..cs(g|]}dddt|qS)r*rmrnrrr^r'r3r4rscs,g|]}dddt|dqS)r*rmrnrrrrr3r4rs cs0g|]\}}td|dt|dqSr*r)rr:rr#r3r4r scs0g|]\}}t|dt||dqSrr)rr:rrr3r4r&s)r@rArrrsrallzipr"_group_tensors_by_device_and_dtypevaluesr _foreach_negis_cpu _foreach_add_rD _foreach_mul_ _foreach_add_foreach_lerp__foreach_addcmul_ _foreach_sqrt _foreach_mul _foreach_pow _foreach_sub_ _foreach_neg__foreach_sqrt_ _foreach_div_ _foreach_sub_foreach_addcdiv_r)"r"rWrXrYrZr[r^r_r#r&r'r%r(rr r!r\grouped_tensorsgrouped_params grouped_gradsgrouped_exp_avgsgrouped_exp_avg_sqsgrouped_mu_productsgrouped_state_steps_exp_avg_sq_sqrtexponentmusmu_nextsbias_correction_sqrtrstep_size_gradsstep_size_expavg numeratorr3)r^r_rr#r'r4_multi_tensor_nadamps                                             r)single_tensor_fnFrc Cstdd|Ds tdtdd|Dstd|dur't|| dd\}}|r2tjr2td |r.zPAPI has changed, `state_steps` argument must contain a list of singleton tensorscsrrdrrr3r3r4r[rzPAPI has changed, `mu_products` argument must contain a list of singleton tensorsNF) use_fusedz6torch.jit.script not supported with foreach optimizers) r^r_r#r&r'rr(r%r r!r\)rrQr rAjit is_scriptingrr)r"rWrXrYrZr[r(rr r!r\rr^r_r#r&r'r%rfuncr3r3r4r9sH  )FNFFFF) typingrrrrrrAr optimizerr r r r r rrrrrrrrrr__all__r__doc__rCrhrrrr3r3r3r4s  D6' C      ^      J