o igP&@sdZddlmZddlZddlmZddlmZmZmZm Z m Z m Z m Z m Z mZmZmZmZmZmZddgZGd ddeZd d ed ed e d e d ed e_deedeedeedeedeedeedededededededededededdf"d d!Zdeedeedeedeedeedeedededededededededededdf"d"d#Ze ed$  % % % %d(deedeedeedeedeedeed&edBdededededededededededdf$d'dZdS))z)Implementation for the RMSprop algorithm.)castN)Tensor)_capturable_doc_default_to_fused_or_foreach_differentiable_doc_disable_dynamo_if_unsupported _foreach_doc!_get_capturable_supported_devices_get_scalar_dtype _maximize_doc _params_doc _to_scalar_use_grad_for_differentiable _view_as_real OptimizerParamsTRMSproprmspropcseZdZ          ddedeeBd ed ed ed ed edededBdededdffdd ZfddZddZ e dddZ Z S)r{Gz?Gz?:0yE>rFNparamslralphaeps weight_decaymomentumcentered capturableforeachmaximizedifferentiablereturnc st|tr|dkrtdd|kstd|d|ks%td|d|ks0td|d|ks;td|d|ksFtd||||||||| | | d } t|| dS) NrzTensor lr must be 1-elementgzInvalid learning rate: zInvalid epsilon value: zInvalid momentum value: zInvalid weight_decay value: zInvalid alpha value: ) rrrrrrrr r!r") isinstancernumel ValueErrorsuper__init__) selfrrrrrrrrr r!r"defaults __class__U/sda-disk/www/egybert/egybert_env/lib/python3.10/site-packages/torch/optim/rmsprop.pyr(s0 zRMSprop.__init__cst||jD]_}|dd|dd|dd|dd|dd|dd|d D]4}|j|g}t|dkrgt|d sgt |d }|dr]tj |t |j d ntj |t d |d <q3q dS) NrrrFr r!r"rrstepdtypedevicer1) r' __setstate__ param_groups setdefaultstategetlentorch is_tensorfloattensorr r2)r)r7grouppp_statestep_valr+r-r.r4Hs*           zRMSprop.__setstate__c Cs8d}|dD]} | jdurq|t| O}|| | jjr"td|| j|j| } t| dkrs|drAtjdt | j dntjdt d| d <tj | tj d | d <|d dkretj | tj d | d <|drstj | tj d | d<|| d || d |d dkr|| d |dr|| dq|S)NFrz)RMSprop does not support sparse gradientsrrr-r0r3r/) memory_format square_avgrmomentum_bufferrgrad_avg) gradr: is_complexappend is_sparse RuntimeErrorr7r9zerosr r2 zeros_likepreserve_format) r)r>params_with_gradgrads square_avgsmomentum_buffer_list grad_avgs state_steps has_complexr?r7r-r-r. _init_group]sD          zRMSprop._init_groupc Cs|d}|dur!t |}Wdn1swY|jD]B}g}g}g}g}g}g} |||||||| } t|||||| |d|d|d|d|d|d|d|d |d |d | d q$|S) zPerform a single optimization step. Args: closure (Callable, optional): A closure that reevaluates the model and returns the loss. Nrrrrrrr r!r"r) rrrrrrr r!r"rrT) _cuda_graph_capture_health_checkr: enable_gradr5rUr) r)closurelossr>rNrOrPrRrQrSrTr-r-r.r/sT   z RMSprop.step) rrrrrFFNFFN) __name__ __module__ __qualname__rr<rboolr(r4rUrr/ __classcell__r-r-r+r.rsP      )3aj Implements RMSprop algorithm. .. math:: \begin{aligned} &\rule{110mm}{0.4pt} \\ &\textbf{input} : \alpha \text{ (alpha)}, \: \gamma \text{ (lr)}, \: \theta_0 \text{ (params)}, \: f(\theta) \text{ (objective)} \\ &\hspace{13mm} \lambda \text{ (weight decay)},\: \mu \text{ (momentum)}, \: centered, \: \epsilon \text{ (epsilon)} \\ &\textbf{initialize} : v_0 \leftarrow 0 \text{ (square average)}, \: \textbf{b}_0 \leftarrow 0 \text{ (buffer)}, \: g^{ave}_0 \leftarrow 0 \\[-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}v_t \leftarrow \alpha v_{t-1} + (1 - \alpha) g^2_t \hspace{8mm} \\ &\hspace{5mm} \tilde{v_t} \leftarrow v_t \\ &\hspace{5mm}if \: centered \\ &\hspace{10mm} g^{ave}_t \leftarrow g^{ave}_{t-1} \alpha + (1-\alpha) g_t \\ &\hspace{10mm} \tilde{v_t} \leftarrow \tilde{v_t} - \big(g^{ave}_{t} \big)^2 \\ &\hspace{5mm}if \: \mu > 0 \\ &\hspace{10mm} \textbf{b}_t\leftarrow \mu \textbf{b}_{t-1} + g_t/ \big(\sqrt{\tilde{v_t}} + \epsilon \big) \\ &\hspace{10mm} \theta_t \leftarrow \theta_{t-1} - \gamma \textbf{b}_t \\ &\hspace{5mm} else \\ &\hspace{10mm}\theta_t \leftarrow \theta_{t-1} - \gamma g_t/ \big(\sqrt{\tilde{v_t}} + \epsilon \big) \hspace{3mm} \\ &\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 `lecture notes `_ by G. Hinton. and centered version `Generating Sequences With Recurrent Neural Networks `_. The implementation here takes the square root of the gradient average before adding epsilon (note that TensorFlow interchanges these two operations). The effective learning rate is thus :math:`\gamma/(\sqrt{v} + \epsilon)` where :math:`\gamma` is the scheduled learning rate and :math:`v` is the weighted moving average of the squared gradient. z Args: a0 lr (float, Tensor, optional): learning rate (default: 1e-2) alpha (float, optional): smoothing constant (default: 0.99) 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 (float, optional): momentum factor (default: 0) centered (bool, optional) : if ``True``, compute the centered RMSProp, the gradient is normalized by an estimation of its variance z z rrOrPrRrQrSrrrrrrr!r"rrTr#c Cstjs t|}t|D]\}}||}tjs5|r5t}|jj |jj kr-|jj |vs5t d|d||}| s=|n| }||}|d7}| dkrS|j || d}t |}|rit |}t |}t |}||j||d|d| r||}|rt |}||d||j||dd}n|}| r| |}n||}| dkr||}|rt |}|| |||j|| dq |j||| dq dS)NIIf capturable=True, params and state_steps must be on supported devices: .rrrvalue)r:jit is_scriptingr enumeratecompiler is_compilingr r2typeAssertionErroraddrG view_as_realmul_addcmul_lerp_addcmulsqrt_sqrtadd_addcdiv_)rrOrPrRrQrSrrrrrrr!r"rrTiparamr/capturable_supported_devicesrFrCis_complex_paramrEavgbufr-r-r._single_tensor_rmsprop sR           r}c !st|dkrdS| rtdtjs0|r0ttfddt||ddDs0tddt|}t ||||||g}| D]0\\}}}}}}}t t t|}t t t|}t t t|}t t t|}|r||g}| dkrt t t|}||| rt t t|}||t|g|R| rt|}tjs|djrtj|tjd d d d d nt|d | dkr| rtj||| d ntj||| d }t||tj|||d |d| r t t t|}t||d |tj|||dd}t|t||n t|}t||| dkrQt t t|}t|| t||||rGt|tjrGt|| } t|| qCtj||| d qC|rjt|tjrjt|| t|||qCtj|||| dqCdS)Nrz#_foreach ops don't support autogradc3s0|]\}}|jj|jjko|jjvVqdSrZ)r2rk).0r?r/ryr-r. rs   z(_multi_tensor_rmsprop..T)strictr`rag?cpu)r2rbrrcre) r9rlr:rirjr allziprr"_group_tensors_by_device_and_dtypevaluesrlistrrHr _foreach_negis_cpu _foreach_add_r= _foreach_add _foreach_mul__foreach_addcmul__foreach_lerp__foreach_addcmul_foreach_sqrt_ _foreach_sqrt_foreach_addcdiv_r$ _foreach_mul _foreach_div_)!rrOrPrRrQrSrrrrrrr!r"rrTgrouped_tensorsgrouped_params_grouped_grads_grouped_square_avgs_grouped_grad_avgs_grouped_momentum_buffer_list_grouped_state_steps__grouped_params grouped_gradsgrouped_square_avgsgrouped_state_stepsstate_and_gradsgrouped_momentum_buffer_listgrouped_grad_avgsr{ momentum_lrr-rr._multi_tensor_rmspropVs                r)single_tensor_fnFr c Cstjstdd|Dstd|durt||dd\}}|r*tjr*td|r4tjs4t}nt }|||||||| | | ||||| || ddS) ztFunctional API that performs rmsprop algorithm computation. See :class:`~torch.optim.RMSProp` for details. css|] }t|tjVqdSrZ)r$r:r)r~tr-r-r.rs  zrmsprop..zPAPI has changed, `state_steps` argument must contain a list of singleton tensorsNF) use_fusedz6torch.jit.script not supported with foreach optimizers) rrrrrrr!rr"rT) r:rirjrrJrrfrgrr})rrOrPrRrQrSr r!r"rrTrrrrrrrfuncr-r-r.rsB  )NFFFF)__doc__typingrr:r optimizerrrrrr r r r r rrrrr__all__rrr<r^r}rrr-r-r-r.s  @,,   A      M