o hSf&@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) isinstancernumel ValueErrordictsuper__init__) selfr r!r"r#r$r%r&rrrrdefaults __class__e/var/www/html/construction_image-detection-poc/venv/lib/python3.10/site-packages/torch/optim/nadam.pyr.s6 zNAdam.__init__cst||jD]|}|dd|dd|dd|dd|dd|dD]W}|j|g}t|dkrt|d sat |d }|drWtj |t |j d ntj |t d |d <t|d s|d }|drztj |t |j d ntj |t d |d <q-q dS) NrFrrrr&r rstepdtypedevicer7 mu_product) r- __setstate__ param_groups setdefaultstategetlentorch is_tensorfloattensorr r8)r/r>grouppp_statestep_val mu_prod_valr1r3r4r;Js:            zNAdam.__setstate__c Cs*d}|dD]} | jdur|t| O}|| | jjr!td|| j|j| } t| dkrv|dr@tjdt | j dntj dt d | d <|drXtj dt | j dntj d t d | d <tj | tjd | d<tj | tjd | d<|| d|| d|| d || d q|S)NFr z'NAdam does not support sparse gradientsrrr3r6r'r9r5r(r:) memory_formatexp_avg exp_avg_sq)gradrA is_complexappend is_sparse RuntimeErrorr>r@zerosr r8rDones zeros_likepreserve_format) r/rEparams_with_gradgradsexp_avgs exp_avg_sqs mu_products state_steps has_complexrFr>r3r3r4 _init_grouphs<        zNAdam._init_groupc Cs|d}|dur!t |}Wdn1swY|jD]N}g}g}g}g}g}g} ttttf|d\} } |||||||| } t|||||| | | |d|d|d|d|d|d|d |d |d | d q$|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&rrrr\) _cuda_graph_capture_health_checkrA enable_gradr<rtuplerCr]r) r/closurelossrErVrWrXrYrZr[r^r_r\r3r3r4r5sX   z NAdam.step)rrrrrFN)__name__ __module__ __qualname__rrrCrrbboolrr.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`_. 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 rWrXrYrZr[r^r_r!r$r%r#r&rrrr\c CsZt|D]%\}}| s||n|| }||}||}||}||}t|r=t|}t|}t|}t|}tjsg|rgt}|jj|jjkrW|jjkr_nn|jj|vsgJd|d|d7}|rp|}nt |}d||}| dkr| r| d|| n|j || d}|ddd|| }|ddd|d| }||9}| |d|| |j ||d|d ||}|s|r| | }||}|| d|d|}|| |d|}||||||qt ||}|| |j||| d|dt |d |j||| |d|d qdS) NzVIf capturable=True, params, mu_products and state_steps must be on supported devices: .rralphar(?Q?)value) enumeraterArN view_as_realcompiler is_compilingr r8typermul_addlerp_addcmul_divsqrtaddcdiv_add_)r rWrXrYrZr[r^r_r!r$r%r#r&rrrr\iparamrMrKrLr:step_tcapturable_supported_devicesr5bias_correction2mumu_nextdenommu_product_nextr3r3r4_single_tensor_nadamsd     $     rc (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: rkr(cpu)r8rlrrogcs g|] }dt|dqS)rrnrrr5)r_r3r4 sz'_multi_tensor_nadam..cs(g|]}dddt|qS)r(rnrorrr^r%r3r4rscs,g|]}dddt|dqS)r(rnrorrrrr3r4rs cs0g|]\}}td|dt|dqSr(r)rr:rr!r3r4r+scs0g|]\}}t|dt||dqSrr)rr:rrr3r4r1s) r@rArsrtr allzipr"_group_tensors_by_device_and_dtypevaluesrlistrr _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$r%r#r&rrrr\grouped_tensorsgrouped_params_grouped_grads_grouped_exp_avgs_grouped_exp_avg_sqs_grouped_mu_products_grouped_state_steps__grouped_params grouped_gradsgrouped_exp_avgsgrouped_exp_avg_sqsgrouped_mu_productsgrouped_state_stepsexp_avg_sq_sqrtexponentmusmu_nextsbias_correction_sqrtrstep_size_gradsstep_size_expavg numeratorr3)r^r_rr!r%r4_multi_tensor_nadamus                                            r)single_tensor_fnFrc Cstdd|Ds tdtdd|Dstd|dur't|| dd\}}|r2tjr2td |r.zPAPI has changed, `state_steps` argument must contain a list of singleton tensorscsrrerrr3r3r4rerzPAPI has changed, `mu_products` argument must contain a list of singleton tensorsNF) use_fusedz6torch.jit.script not supported with foreach optimizers) r^r_r!r$r%rr&r#rrr\)rrQrrAjit is_scriptingrr)r rWrXrYrZr[r&rrrr\rr^r_r!r$r%r#rfuncr3r3r4rDsH  )FNFFFF)__doc__typingrrrrAr optimizerrrr r r r r rrrrrrrr__all__rrrCrirrrr3r3r3r4s D8' C      ^      P