a dO;@sJddlZddlmZddlmZmZmZmZmZmZm Z m Z ddl m Z m Z ddlmZddgZGd ddeZd d je ed e_de ee ee ee ee ee ee eeeeeeeedddZe ee ee ee ee ee eeeeeeeed ddZe ee ee ee ee ee eeeeeeeed ddZdS)N)Tensor) Optimizer_use_grad_for_differentiable _get_value_dispatch_sqrt_stack_if_compiling_differentiable_doc _foreach_doc_default_to_fused_or_foreach)ListOptional)"_group_tensors_by_device_and_dtypeNAdamnadamcsTeZdZddddeeedfd d Zfd d Zd dZedddZ Z S)rMb`?g?g+?:0yE>rMbp?NFforeachdifferentiablec 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: {}z Invalid momentum_decay value: {})lrbetaseps weight_decaymomentum_decayrr) ValueErrorformatdictsuper__init__) selfparamsrrrrrrrdefaults __class__Z/var/www/html/stable-diffusion-webui/venv/lib/python3.9/site-packages/torch/optim/nadam.pyr# s"zNAdam.__init__cst||jD]}|dd|ddqt|j}t|dkoZt |dd}|s|D]}t t |d|d<qdt|dkot |dd}|s|D]}t |d|d<qdS)NrrFrstep mu_product) r" __setstate__ param_groups setdefaultliststatevalueslentorchZ is_tensortensorfloat)r$r1groupZ state_valuesZstep_is_tensorsZmu_product_is_tensorr'r)r*r-s   zNAdam.__setstate__c Cs|dD]}|jdur|||jjr0td||j|j|} t| dkrtd| d<td| d<tj|tj d| d <tj|tj d| d <|| d || d || d|| dqdS) Nr%z'NAdam does not support sparse gradientsrrr+rr,)Z memory_formatexp_avg exp_avg_sq) gradappendZ is_sparse RuntimeErrorr1r3r4r5Z zeros_likeZpreserve_format) r$r7params_with_gradgradsexp_avgs exp_avg_sqs mu_products state_stepspr1r)r)r* _init_group.s       zNAdam._init_groupc Csd}|dur:t|}Wdn1s00Y|jD]z}g}g}g}g}g}g} |d\} } |||||||| t|||||| | | |d|d|d|d|d|dd q@|S) zPerforms a single optimization step. Args: closure (Callable, optional): A closure that reevaluates the model and returns the loss. Nrrrrrrr)beta1beta2rrrrrr)r4Z enable_gradr.rEr) r$closureZlossr7r>r?r@rArBrCrFrGr)r)r*r+Es: $  z NAdam.step)rrrrr)N) __name__ __module__ __qualname__r boolr#r-rErr+ __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)} \\ &\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}if \: \lambda \neq 0 \\ &\hspace{10mm} 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-1} - \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) {foreach} {differentiable} .. _Incorporating Nesterov Momentum into Adam: https://openreview.net/forum?id=OM0jvwB8jIp57ZJjtNEZ rF)r%r?r@rArBrCrrrFrGrrrrcCstdd|Dstdtdd|Ds4td|durNt||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* znadam..zPAPI has changed, `state_steps` argument must contain a list of singleton tensorscss|]}t|tjVqdSrNrOrQr)r)r*rTrUzPAPI has changed, `mu_products` argument must contain a list of singleton tensorsNF)Z use_fusedz6torch.jit.script not supported with foreach optimizers)rFrGrrrrr)allr=r r4ZjitZ is_scripting_multi_tensor_nadam_single_tensor_nadam)r%r?r@rArBrCrrrFrGrrrr_funcr)r)r*rs2) r%r?r@rArBrCrFrGrrrrrcCst|D]\} }|| }|| }|| }|| }|| }|d7}t|}d||}| dkrl|j|| d}|ddd|| }|ddd|d| }||9}||j|d|d||j||d|d||}| rN|| }||}|| d|d|}|| d|d|}||||||qt||}|| |j||| d|dt|d|j||| |d|dqdS)Nrralphar?Q?)value) enumerateraddZmul_Zadd_Zaddcmul_divsqrtZaddcdiv_)r%r?r@rArBrCrFrGrrrrriparamr;r9r:r,Zstep_tr+bias_correction2mumu_nextdenomZmu_product_nextr)r)r*rXs8     &rXcst|dkrdS| rJdt||||||g} | D]B\}}}}}}t|dfdd|D}fdd|D}fdd|D}t||| dkrtj||| d}t|tj||ddt|t|||dt|}d d|D}t ||t|| }t fd dt ||D}t fd dt ||D}t ||||t ||||q8dS) Nrz#_foreach ops don't support autogradrcsg|]}dt|qS)rrrRr+)rGr)r* .rUz'_multi_tensor_nadam..cs(g|] }dddt|qS)rr]r^rjrkrFrr)r*rl/rUcs,g|]$}dddt|dqS)rr]r^rrjrkrmr)r*rl0sr[cSsg|] }t|qSr))r)rRZbcr)r)r*rlArUcs,g|]$\}}d|dt|dqSrrj)rRr,rgrr)r*rlEscs,g|]$\}}|dt||dqSrnrj)rRr,rhrpr)r*rlGs) r3rr2r4Z _foreach_add_Z _foreach_mul_Z _foreach_addZ_foreach_addcmul_Z _foreach_sqrtZ _foreach_div_rzipZ_foreach_addcdiv_)r%r?r@rArBrCrFrGrrrrrZgrouped_tensorsZgrouped_paramsZ grouped_gradsZgrouped_exp_avgsZgrouped_exp_avg_sqsZgrouped_mu_productsZgrouped_state_stepsrfZmusZmu_nextsZexp_avg_sq_sqrtZbias_correction_sqrtriZstep_size_gradsZstep_size_expavgr))rFrGrrr*rWsD             rW)NF)r4rZ optimizerrrrrrr r r typingr r Ztorch.utils._foreach_utilsr__all__rr __doc__rLr6rrXrWr)r)r)r*sp ( d: 7 ;