U ʹjR@sXddlZddlmZddlmZmZmZmZmZmZm Z m Z m Z ddl m Z mZddgZGdddeZd d e d ed e d e_de ee ee ee ee ee eeeeeeeeeeeedddZe ee ee ee ee ee eeeeeeeeeedddZe ee ee ee ee ee eeeeeeeeeedddZdS)N)Tensor) Optimizer_use_grad_for_differentiable _get_value_dispatch_sqrt_stack_if_compiling_capturable_doc_differentiable_doc _foreach_doc_default_to_fused_or_foreach)ListOptionalNAdamnadamcsZeZdZdddddeeeeed fd d Zfd d ZddZedddZ Z S)rMb`?g?g+?:0yE>rMbp?FN)foreach capturabledifferentiable)decoupled_weight_decayrrrc sd|kstd|d|ks,td|d|dkrDdksXntd|dd|dkrpdksntd|dd|kstd |d|kstd |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: ) lrbetaseps weight_decaymomentum_decayrrrr) ValueErrordictsuper__init__) selfparamsrrrrrrrrrdefaults __class__P/var/www/html/bwave/python/venv/lib/python3.8/site-packages/torch/optim/nadam.pyr# s*zNAdam.__init__cst||jD]4}|dd|dd|dd|ddqt|j}t|dkort |dd}|s|D]}t t |d|d<q|t|dkot |dd}|s|D]}t |d|d<qdS) NrrFrrrstep mu_product) r" __setstate__ param_groups setdefaultliststatevalueslentorchZ is_tensortensorfloat)r$r1groupZ state_valuesZstep_is_tensorsZmu_product_is_tensorr'r)r*r- s     zNAdam.__setstate__c Cs|dD]}|jdk r|||jjr2td||j|j|} t| dkr|drptjdtj|j dnt d| d<|drtj dtj|j dnt d | d <tj |tj d | d <tj |tj d | d <|| d || d || d || dqdS)Nr%z'NAdam does not support sparse gradientsrrr))Zdtypedevicerr+rr,)Z memory_formatexp_avg exp_avg_sq)gradappendZ is_sparse RuntimeErrorr1r3r4Zzerosr6r9r5ZonesZ zeros_likeZpreserve_format) r$r7params_with_gradgradsexp_avgs exp_avg_sqs mu_products state_stepspr1r)r)r* _init_group1s,     zNAdam._init_groupc Cs|d}|dk r.t |}W5QRX|jD]}g}g}g}g}g}g} |d\} } |||||||| t|||||| | | |d|d|d|d|d|d|d |d d q4|S) zPerforms a single optimization step. Args: closure (Callable, optional): A closure that reevaluates the model and returns the loss. Nrrrrrrrrr) beta1beta2rrrrrrrr)Z _cuda_graph_capture_health_checkr4Z enable_gradr.rFr) r$closureZlossr7r?r@rArBrCrDrGrHr)r)r*r+Qs@   z NAdam.step)rrrrrF)N) __name__ __module__ __qualname__boolrr#r-rFrr+ __classcell__r)r)r'r*r s$  a 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} \\ &\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}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 F)r%r@rArBrCrDrrrrrGrHrrrrc Cstdd|Dstdtdd|Ds4td|dkrNt|| dd\}}|rdtjrdtd |rxtjsxt}nt}|||||||| | | | ||||| d dS) zpFunctional API that performs NAdam algorithm computation. See :class:`~torch.optim.NAdam` for details. css|]}t|tjVqdSN isinstancer4r.0tr)r)r* sznadam..zPAPI has changed, `state_steps` argument must contain a list of singleton tensorscss|]}t|tjVqdSrOrPrRr)r)r*rUszPAPI has changed, `mu_products` argument must contain a list of singleton tensorsNF)Z use_fusedz6torch.jit.script not supported with foreach optimizers) rGrHrrrrrrr)allr>r r4ZjitZ is_scripting_multi_tensor_nadam_single_tensor_nadam)r%r@rArBrCrDrrrrrGrHrrrr_funcr)r)r*rs6)r%r@rArBrCrDrGrHrrrrrrrc Cst|D]\}}||}||}||}||}||}tjst| rt|jrZ|jrZ|jst|jrl|jrl|jsttd|d7}| r|}nt|}d||}| dkr| r|d|| n|j || d}|ddd|| }|ddd|d| }||9}| |d|||j ||d|d| | }|sH| r| | }||}|| d|d|}|| |d|}||||||qt||}|| |j||| d|dt|d|j||| |d|dqdS) NzUIf capturable=True, params, mu_products, and state_steps must be CUDA or XLA tensors.rralphar?Q?)value) enumerater4_utils is_compilingis_cudaZis_xlaAssertionErrorrZmul_addZlerp_Zaddcmul_divsqrtZaddcdiv_Zadd_)r%r@rArBrCrDrGrHrrrrrrriparamr<r:r;r,Zstep_tr+Zbias_correction2mumu_nextdenomZmu_product_nextr)r)r*rXs^      &rXc s t|dkrdS|rtdtjsL| rLtddt|||DsLtdt||||||g}| D]\\}}}}}}}t |d| dkr| rt |d| ntj ||| d}t ||dt |t|||dt|}| rt|}td|}t |d t |d t |t |td|}t |d t |d t |~t|}t|d t|t|n:fd d |D}fd d |D}fdd |D}t ||t||t || ~| rt|d t |t|d }t|t|||}~t||}t |t|d t|||}~t||}t|||t|||qjtfdd t||D}tfdd t||D}t||||t||||qjdS)Nrz#_foreach ops don't support autogradcss&|]\}}}|jo|jo|jVqdSrO)rc)rSrEmpr+r)r)r*rUYsz&_multi_tensor_nadam..zNIf capturable=True, params, mu_products, and state_steps must be CUDA tensors.rr[r^grcs g|]}tdt|qS)r)rrrSr+)rHr)r* sz'_multi_tensor_nadam..cs(g|] }dddt|qS)rr]r^rrnrGrr)r*roscs,g|]$}dddt|dqS)rr]r^rrprnrqr)r*roscs,g|]$\}}d|dt|dqSrrp)rSr,rjrr)r*roscs,g|]$\}}|dt||dqSrrrp)rSr,rkrtr)r*ros)r3rdr4rarbrVziprZ"_group_tensors_by_device_and_dtyper2Z _foreach_add_Z _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%r@rArBrCrDrGrHrrrrrrrZgrouped_tensorsZgrouped_paramsZ grouped_gradsZgrouped_exp_avgsZgrouped_exp_avg_sqsZgrouped_mu_productsZgrouped_state_stepsrYZexp_avg_sq_sqrtexponentZmusZmu_nextsZbias_correction_sqrtrlZstep_size_gradsZstep_size_expavg numeratorr))rGrHrrr*rWAs                                    rW)FNFF)r4rZ optimizerrrrrrr r r r typingr r__all__r__doc__rMr6rrXrWr)r)r)r*s ,u# D ; L