o iZh(@sdZddlmZddlZddlmZddlmZmZmZm Z m Z m Z m Z m Z mZmZmZmZmZmZmZmZddgZGd ddeZd d ed e d ed ed ed e_deedeedeedeedeedeedededededededededededed df$d!d"Zdeedeedeedeedeedeedededededededededededed df$d#d$Ze ed% &  & & & &d)deedeedeedeedeedeeded'edBdedededededededededed df&d(dZdS)*z'Implementation for the NAdam algorithm.)castN)Tensor)_capturable_doc_default_to_fused_or_foreach_differentiable_doc_disable_dynamo_if_unsupported _foreach_doc!_get_capturable_supported_devices_get_scalar_dtype _get_value _maximize_doc _params_doc_stack_if_compiling _to_scalar_use_grad_for_differentiable _view_as_real OptimizerParamsTNAdamnadamcseZdZ      ddddddd ed eeBd eeefd ed ededededBdedededdffddZfddZ ddZ e dddZ Z S)rMb`?g?g+?:0yE>rMbp?FN)foreachmaximize capturabledifferentiableparamslrbetaseps weight_decaymomentum_decaydecoupled_weight_decayrrrrreturnc st|tr|dkrtdd|kstd|d|ks%td|d|dkr1dks;ntd|dd|dkrGdksQntd |dd|ks\td |d|ksgtd |||||||| || | d } t|| dS) NrzTensor lr must be 1-elementzInvalid learning rate: zInvalid epsilon value: r?z#Invalid beta parameter at index 0: z#Invalid beta parameter at index 1: zInvalid weight_decay value: zInvalid momentum_decay value: ) r r!r"r#r$r%rrrr) isinstancernumel ValueErrorsuper__init__) selfrr r!r"r#r$r%rrrrdefaults __class__S/sda-disk/www/egybert/egybert_env/lib/python3.10/site-packages/torch/optim/nadam.pyr-!s4 zNAdam.__init__cst||jD]|}|dd|dd|dd|dd|dd|dD]W}|j|g}t|dkrt|d sat |d }|drWtj |t |j d ntj |t d |d <t|d s|d }|drztj |t |j d ntj |t d |d <q-q dS) NrFrrrr%rrstepdtypedevicer6 mu_product) r, __setstate__ param_groups setdefaultstategetlentorch is_tensorfloattensorr r7)r.r=grouppp_statestep_val mu_prod_valr0r2r3r:Ls:            zNAdam.__setstate__c Cs*d}|dD]} | jdur|t| O}|| | jjr!td|| j|j| } t| dkrv|dr@tjdt | j dntj dt d | d <|drXtj dt | j dntj d t d | d <tj | tjd | d<tj | tjd | d<|| d|| d|| d || d q|S)NFrz'NAdam does not support sparse gradientsrrr2r5r'r8r4r(r9) memory_formatexp_avg exp_avg_sq)gradr@ is_complexappend is_sparse RuntimeErrorr=r?zerosr r7rCones zeros_likepreserve_format) r.rDparams_with_gradgradsexp_avgs exp_avg_sqs mu_products state_steps has_complexrEr=r2r2r3 _init_groupjs<        zNAdam._init_groupc Cs|d}|dur!t |}Wdn1swY|jD]N}g}g}g}g}g}g} ttttf|d\} } |||||||| } t|||||| | | |d|d|d|d|d|d|d |d |d | d q$|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%rrrr[) _cuda_graph_capture_health_checkr@ enable_gradr;rtuplerBr\r) r.closurelossrDrUrVrWrXrYrZr]r^r[r2r2r3r4sX   z NAdam.step)rrrrrFN)__name__ __module__ __qualname__rrBrraboolr-r:r\rr4 __classcell__r2r2r0r3r sR       +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 rrVrWrXrYrZr]r^r r#r$r"r%rrrr[r&c  Csttjs t|}t|D])\}}| s||n|| }||}||}||}||}t|rFt|}t|}t|}t|}tjsp|rpt }|j j |j j kr`|j j krhnn|j j |vspt d|d|d7}|ry|}nt |}d||}| dkr| r|d|| n|j|| d}|ddd|| }|ddd|d| }||9}||d|||j||d|d ||}|s|r|| }||}|| d|d|}|| |d|}||||||q t ||}|| |j||| d|dt |d |j||tt| |d|d q dS) NzVIf capturable=True, params, mu_products and state_steps must be on supported devices: .rralphar(?Q?)value)r@jit is_scriptingr enumeraterM view_as_realcompiler is_compilingr r7typeAssertionErrorr mul_addlerp_addcmul_divsqrtaddcdiv_add_rrB)rrVrWrXrYrZr]r^r r#r$r"r%rrrr[iparamrLrJrKr9step_tcapturable_supported_devicesr4bias_correction2mumu_nextdenommu_product_nextr2r2r3_single_tensor_nadamsn      $       rc (st|dkrdS|rtdtjs3|r3tddtfddt|||ddDs3td d tt ||||||g}| D]\\}}}}}}}t t t|}t t t|}t t t|}t t t|}t t t|}t t t|}|rt||||| rt|}tjs|djrtj|tjd d d d dnt|d| dkr| rt|d| n| rtj||| dntj||| d}t||dt|t|||dt|}|rBt|} 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 ||'|qFt!fddt||!ddD}%t!fddt||"ddD}&t ||||%t ||||&qFdS)Nrz#_foreach ops don't support autogradF) supports_xlac3sF|]\}}}|jj|jjko|jjkno|jjvVqdSrd)r7rv).0rEmpr4)rr2r3 s $  z&_multi_tensor_nadam..T)strictzWIf capturable=True, params, mu_products, and state_steps must be on supported devices: rjr(cpu)r7rkrrngcs g|] }dt|dqS)rrmr rr4)r^r2r3 sz'_multi_tensor_nadam..cs(g|]}dddt|qS)r(rmrnrrr]r$r2r3rscs,g|]}dddt|dqS)r(rmrnrrrrr2r3r s cs0g|]\}}td|dt|dqSr(r)rr9rr r2r3r;scs0g|]\}}t|dt||dqSrr)rr9rrr2r3rAs)"r?rwr@rtrur allziprr"_group_tensors_by_device_and_dtypevaluesrlistrr _foreach_negis_cpu _foreach_add_rC _foreach_mul_ _foreach_add_foreach_lerp__foreach_addcmul_ _foreach_sqrt _foreach_mul _foreach_pow _foreach_sub_ _foreach_neg__foreach_sqrt_ _foreach_div_ _foreach_sub_foreach_addcdiv_r)(rrVrWrXrYrZr]r^r r#r$r"r%rrrr[grouped_tensorsgrouped_params_grouped_grads_grouped_exp_avgs_grouped_exp_avg_sqs_grouped_mu_products_grouped_state_steps__grouped_params grouped_gradsgrouped_exp_avgsgrouped_exp_avg_sqsgrouped_mu_productsgrouped_state_stepsexp_avg_sq_sqrtexponentmusmu_nextsbias_correction_sqrtrstep_size_gradsstep_size_expavg numeratorr2)r]r^rr r$r3_multi_tensor_nadam}s                                          r)single_tensor_fnFrc Cstdd|Ds tdtdd|Dstd|dur't|| dd\}}|r2tjr2td |r.zPAPI has changed, `state_steps` argument must contain a list of singleton tensorscsrrdrrr2r2r3r}rzPAPI has changed, `mu_products` argument must contain a list of singleton tensorsNF) use_fusedz6torch.jit.script not supported with foreach optimizers) r]r^r r#r$rr%r"rrr[)rrPrr@rprqrr)rrVrWrXrYrZr%rrrr[rr]r^r r#r$r"rfuncr2r2r3r\sH  )FNFFFF)__doc__typingrr@r optimizerrrrrr r r r r rrrrrrr__all__rrrBrhrrrr2r2r2r3s  H8' C      d      `