a h)A@srddlmZmZmZmZddlZddlmZddlmZm Z m Z m Z m Z m Z mZmZmZmZmZmZmZmZddgZGdddeZd d ed e d ed ed e d e_eeeeeeeeeeeeeeeeeed ddZeeeeeeeeeeeeeeeeeed ddZe eddeeeeeeeeeeeeeeeeeeeedddZdS))AnycastOptionalUnionN)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 OptimizerParamsTAdadeltaadadeltac seZdZdddddeeeefeeeeeeeed fd d Z fd d Z e e e feeeeeeeeeed ddZedddZZS)r??ư>rNF) capturablemaximizedifferentiable) paramslrrhoeps weight_decayforeachrrrc st|tr|dkrtdd|ks4td|d|krHdksXntd|d|ksntd|d|kstd|t|||||||| d } t|| dS) NrzTensor lr must be 1-elementgzInvalid learning rate: rzInvalid rho value: zInvalid epsilon value: zInvalid weight_decay value: )rr r!r"rrr#r) isinstancerZnumel ValueErrordictsuper__init__) selfrrr r!r"r#rrrdefaults __class__B/var/www/auris/lib/python3.9/site-packages/torch/optim/adadelta.pyr(s*  zAdadelta.__init__cst||jD]}|dd|dd|dd|dd|dD]h}|j|g}t|dkrNt|dsNt |d}|drtj |t |j d ntj |t d |d<qNqdS) Nr#rFrrrrstepdtypedevicer1) r' __setstate__ param_groups setdefaultstategetlentorchZ is_tensorfloattensorrr2)r)r7grouppZp_stateZstep_valr+r-r.r4As          zAdadelta.__setstate__)r=params_with_gradgrads square_avgs acc_deltas state_stepsc Csd}|dD]}|jdurq |t|O}|||jjrDtd||j|j|} t| dkr|drtjdt |j dntjdt d| d <tj |tj d | d <tj |tj d | d <|| d || d || d q |S) NFrz*Adadelta does not support sparse gradientsrrr-r0r3r/)Z memory_format square_avg acc_delta) gradr: is_complexappendZ is_sparse RuntimeErrorr7r9zerosrr2Z zeros_likeZpreserve_format) r)r=r?r@rArBrC has_complexr>r7r-r-r. _init_groupTs2         zAdadelta._init_groupcCs|d}|durBt|}Wdn1s80Y|jD]}g}g}g}g}g}|d|d|d|d|d|d|d|d f\} } } } } }}}|||||||}t|||||| | | | | ||||d qH|S) zPerform a single optimization step. Args: closure (Callable, optional): A closure that reevaluates the model and returns the loss. Nrr r!r"r#rrr) rr r!r"r#rrrrK)Z _cuda_graph_capture_health_checkr:Z enable_gradr5rLr)r)closureZlossr=r?r@rArBrCrr r!r"r#rrrrKr-r-r.r/sb $    z Adadelta.step)rrrrN)N)__name__ __module__ __qualname__rrr;rrboolr(r4r&strrlistrLrr/ __classcell__r-r-r+r.rs<  $   +aImplements Adadelta algorithm. .. math:: \begin{aligned} &\rule{110mm}{0.4pt} \\ &\textbf{input} : \gamma \text{ (lr)}, \: \theta_0 \text{ (params)}, \: f(\theta) \text{ (objective)}, \: \rho \text{ (decay)}, \: \lambda \text{ (weight decay)} \\ &\textbf{initialize} : v_0 \leftarrow 0 \: \text{ (square avg)}, \: u_0 \leftarrow 0 \: \text{ (accumulate variables)} \\[-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 v_{t-1} \rho + g^2_t (1 - \rho) \\ &\hspace{5mm}\Delta x_t \leftarrow \frac{\sqrt{u_{t-1} + \epsilon }}{ \sqrt{v_t + \epsilon} }g_t \hspace{21mm} \\ &\hspace{5mm} u_t \leftarrow u_{t-1} \rho + \Delta x^2_t (1 - \rho) \\ &\hspace{5mm}\theta_t \leftarrow \theta_{t-1} - \gamma \Delta x_t \\ &\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 `ADADELTA: An Adaptive Learning Rate Method`_. z Args: ar lr (float, Tensor, optional): coefficient that scale delta before it is applied to the parameters (default: 1.0) rho (float, optional): coefficient used for computing a running average of squared gradients (default: 0.9). A higher value of `rho` will result in a slower average, which can be helpful for preventing oscillations in the learning process. eps (float, optional): term added to the denominator to improve numerical stability (default: 1e-6). weight_decay (float, optional): weight decay (L2 penalty) (default: 0) z zd .. _ADADELTA\: An Adaptive Learning Rate Method: https://arxiv.org/abs/1212.5701 ) rr@rArBrCrr r!r"rrrrKcsbtjsD| rDtddtfddt||DsDJddtjsVt|}t|||||D]\} }}}}|d7}| s|n| }|dkr|j | |d }t | rt |}t |}t |}| |j ||d|d | |}| |}| r |}|| || |j ||d|d t | rLt|}| j|| d qfdS) NFZ supports_xlac3s.|]&\}}|jj|jjko$|jjvVqdSNr2type.0r>r/Zcapturable_supported_devicesr-r. sz*_single_tensor_adadelta..IIf capturable=True, params and state_steps must be on supported devices: .rralphavalue)r:compiler is_compilingr allzipjit is_scriptingraddrGZ view_as_realZmul_Zaddcmul_Zsqrt_cloneZdiv_Zview_as_complexZadd_)rr@rArBrCrr r!r"rrrrKparamrFrDrEr/stddeltar-r[r._single_tensor_adadeltas@           rncsB| r JdtjsP| rPtddtfddt||DsPJddt|dkr`dSt|}t |||||g} | D]\\}}}}}}t t t |}t t t |}t t t |}t t t |}t t t |}| rt||||tjs$|djr$tj|tjd d d d d n t|d | r@t|}|dkrr| rbtj|||d ntj|||d }t||tj|||d |dt||}t|t||}t|t||t||t||tj|||d |d| r*t|tj r*t|| t||qtj||| d qdS)Nz#_foreach ops don't support autogradFrUc3s.|]&\}}|jj|jjko$|jjvVqdSrVrWrYr[r-r.r\Gsz)_multi_tensor_adadelta..r]r^rrcpu)r2r_rra)r:rcrdr rerfr9rrZ"_group_tensors_by_device_and_dtypevaluesrrSrrZis_cpuZ _foreach_add_r<Z _foreach_negZ _foreach_addZ _foreach_mul_Z_foreach_addcmul_Z_foreach_sqrt_Z _foreach_div_r$)rr@rArBrCrr r!r"rrrrKZgrouped_tensorsZdevice_params_Z device_grads_Zdevice_square_avgs_Zdevice_acc_deltas_Zdevice_state_steps__Z device_paramsZ device_gradsZdevice_square_avgsZdevice_acc_deltasZdevice_state_stepsrlZdeltasr-r[r._multi_tensor_adadelta0s|                    rr)Zsingle_tensor_fnF)rr@rArBrCrr#rrKrr r!r"rc Cstjs$tdd|Ds$td|dur>t||dd\}}|rTtjrTtd|rhtjsht}nt }||||||| | | | | |||d dS) zvFunctional API that performs Adadelta algorithm computation. See :class:`~torch.optim.Adadelta` for details. css|]}t|tjVqdSrV)r$r:r)rZtr-r-r.r\szadadelta..zPAPI has changed, `state_steps` argument must contain a list of singleton tensorsNF)Z use_fusedz6torch.jit.script not supported with foreach optimizers)rr r!r"rrrrK) r:rcrdrerIr rgrhrrrn)rr@rArBrCrr#rrKrr r!r"rrqfuncr-r-r.rs< )FNFF)typingrrrrr:rZ optimizerrr r r r r rrrrrrrr__all__r__doc__rSr;rQrnrrrr-r-r-r.s @&  6 < h