o wZhSf&@sdZddlmZmZmZddlZddlmZddlmZm Z m Z m Z m Z m Z mZmZmZmZmZmZmZmZmZddgZGd ddeZd d ed e d ed ed e d e_deedeedeedeedeedeedededededededededededef"d d!Zdeedeedeedeedeedeedededededededededededef"d"d#Ze ed$ %  % % % %d(deedeedeedeedeedeeded&eededededededededededef$d'dZdS))z'Implementation for the NAdam algorithm.)castOptionalUnionN)Tensor)_capturable_doc_default_to_fused_or_foreach_differentiable_doc_disable_dynamo_if_unsupported _foreach_doc!_get_capturable_supported_devices_get_scalar_dtype _get_value _maximize_doc _params_doc_stack_if_compiling_use_grad_for_differentiable _view_as_real OptimizerParamsTNAdamnadamcseZdZ      ddddddd ed eeefd eeefd ed ededede edededeffddZ fddZ ddZ e dddZZS)rMb`?g?g+?:0yE>rMbp?FN)foreachmaximize capturabledifferentiableparamslrbetaseps weight_decaymomentum_decaydecoupled_weight_decayrrrrc st|tr|dkrtdd|kstd|d|ks%td|d|dkr1dks;ntd|dd|dkrGdksQntd |dd|ks\td |d|ksgtd |t||||||| || | d } t|| dS) NrzTensor lr must be 1-elementzInvalid learning rate: zInvalid epsilon value: r?z#Invalid beta parameter at index 0: z#Invalid beta parameter at index 1: zInvalid weight_decay value: zInvalid momentum_decay value: ) r!r"r#r$r%r&rrrr) isinstancerZnumel ValueErrordictsuper__init__) selfr r!r"r#r$r%r&rrrrdefaults __class__@/var/www/auris/lib/python3.10/site-packages/torch/optim/nadam.pyr-s6 zNAdam.__init__cst||jD]|}|dd|dd|dd|dd|dd|dD]W}|j|g}t|dkrt|d sat |d }|drWtj |t |j d ntj |t d |d <t|d s|d }|drztj |t |j d ntj |t d |d <q-q dS) NrFrrrr&r rstepdtypedevicer6 mu_product) r, __setstate__ param_groups setdefaultstategetlentorchZ is_tensorfloattensorr r7)r.r=grouppZp_stateZstep_valZ mu_prod_valr0r2r3r:Js:            zNAdam.__setstate__c Cs*d}|dD]} | jdur|t| O}|| | jjr!td|| j|j| } t| dkrv|dr@tjdt | j dntj dt d | d <|drXtj dt | j dntj d t d | d <tj | tjd | d<tj | tjd | d<|| d|| d|| d || d q|S)NFr z'NAdam does not support sparse gradientsrrr2r5r'r8r4r(r9)Z memory_formatexp_avg exp_avg_sq)gradr@ is_complexappendZ is_sparse RuntimeErrorr=r?Zzerosr r7rBZonesZ zeros_likeZpreserve_format) r.rCparams_with_gradgradsexp_avgs exp_avg_sqs mu_products state_steps has_complexrDr=r2r2r3 _init_grouphs<        zNAdam._init_groupc Cs|d}|dur!t |}Wdn1swY|jD]N}g}g}g}g}g}g} ttttf|d\} } |||||||| } t|||||| | | |d|d|d|d|d|d|d |d |d | d q$|S) zPerform a single optimization step. Args: closure (Callable, optional): A closure that reevaluates the model and returns the loss. Nr"r!r$r%r#rr&rrr) beta1beta2r!r$r%r#rr&rrrrQ) Z _cuda_graph_capture_health_checkr@Z enable_gradr;rtuplerArRr) r.closureZlossrCrKrLrMrNrOrPrSrTrQr2r2r3r4sX   z NAdam.step)rrrrrFN)__name__ __module__ __qualname__rrrArrUboolrr-r:rRrr4 __classcell__r2r2r0r3rsN        +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`_. z Args: a lr (float, Tensor, 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 decouple the weight decay as in AdamW to obtain NAdamW. If True, the algorithm does not accumulate weight decay in the momentum nor variance. (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 rLrMrNrOrPrSrTr!r$r%r#r&rrrrQc CsZt|D]%\}}| s||n|| }||}||}||}||}t|r=t|}t|}t|}t|}tjsg|rgt}|jj|jjkrW|jjkr_nn|jj|vsgJd|d|d7}|rp|}nt |}d||}| dkr| r| d|| n|j || d}|ddd|| }|ddd|d| }||9}| |d|| |j ||d|d ||}|s|r| | }||}|| d|d|}|| |d|}||||||qt ||}|| |j||| d|dt |d |j||| |d|d qdS) NzVIf capturable=True, params, mu_products and state_steps must be on supported devices: .rralphar(?Q?)value) enumerater@rHZ view_as_realcompiler is_compilingr r7typerZmul_addZlerp_Zaddcmul_divsqrtZaddcdiv_Zadd_)r rLrMrNrOrPrSrTr!r$r%r#r&rrrrQiparamrGrErFr9Zstep_tcapturable_supported_devicesr4Zbias_correction2mumu_nextdenomZmu_product_nextr2r2r3_single_tensor_nadamsd     $     rpc (st|dkrdS|rJdtjs1|r1tddtfddt|||Ds1Jddt||||||g}| D]\\}}}}}}}t t t |}t t t |}t t t |}t t t |}t t t |}t t t |}|r~t ||||| rt|}tjs|djrtj|tjd d d d d nt|d | dkr| rt|d | n| rtj||| d ntj||| d }t||d t|t|||d t|}|rs $  z&_multi_tensor_nadam..zWIf capturable=True, params, mu_products, and state_steps must be on supported devices: r]r(cpu)r7r^rragcs g|] }dt|dqS)rr`rrqr4)rTr2r3 sz'_multi_tensor_nadam..cs(g|]}dddt|qS)r(r`rarurvrSr%r2r3rwscs,g|]}dddt|dqS)r(r`rarrurvrxr2r3rws cs0g|]\}}td|dt|dqSr(ru)rqr9rmr!r2r3rw+scs0g|]\}}t|dt||dqSryru)rqr9rnr{r2r3rw1s) r?r@rdrer allziprZ"_group_tensors_by_device_and_dtypevaluesrlistrrZ _foreach_negZis_cpuZ _foreach_add_rBZ _foreach_mul_Z _foreach_addZ_foreach_lerp_Z_foreach_addcmul_Z _foreach_sqrtZ _foreach_mulZ _foreach_powZ _foreach_sub_Z _foreach_neg_Z_foreach_sqrt_Z _foreach_div_Z _foreach_subZ_foreach_addcdiv_r)(r rLrMrNrOrPrSrTr!r$r%r#r&rrrrQZgrouped_tensorsZgrouped_params_Zgrouped_grads_Zgrouped_exp_avgs_Zgrouped_exp_avg_sqs_Zgrouped_mu_products_Zgrouped_state_steps__Zgrouped_paramsZ grouped_gradsZgrouped_exp_avgsZgrouped_exp_avg_sqsZgrouped_mu_productsZgrouped_state_stepsZexp_avg_sq_sqrtexponentZmusZmu_nextsZbias_correction_sqrtroZstep_size_gradsZstep_size_expavg numeratorr2)rSrTrlr!r%r3_multi_tensor_nadamus                                            r)Zsingle_tensor_fnFrc Cstdd|Ds tdtdd|Dstd|dur't|| dd\}}|r2tjr2td |r.zPAPI has changed, `state_steps` argument must contain a list of singleton tensorscsrrWrrr2r2r3rserzPAPI has changed, `mu_products` argument must contain a list of singleton tensorsNF)Z use_fusedz6torch.jit.script not supported with foreach optimizers) rSrTr!r$r%rr&r#rrrQ)r|rJrr@ZjitZ is_scriptingrrp)r rLrMrNrOrPr&rrrrQrrSrTr!r$r%r#rfuncr2r2r3rDsH  )FNFFFF)__doc__typingrrrr@rZ optimizerrrr r r r r rrrrrrrr__all__rrrAr[rprrr2r2r2r3s D8' C      ^      P