˲jBtddlmZmZmZmZmZmZddlZddlmZddl m Z m Z m Z m Z mZmZmZmZmZmZmZmZmZddgZGddeZd d ed ed e d ed e d ze_deedeedeedeedeededededededededefdZdeedeedeedeedeededededededededefdZe e d!deedeedeedeedeededeedededededededefd ZdS)")AnycastDictListOptionalUnionN)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_use_grad_for_differentiable _view_as_real OptimizerParamsTAdadeltaadadeltaceZdZ dddddded eeefd ed ed ed eedededeffdZ fdZ de e e fdeedeedeedeedeef dZeddZxZS)r??ư>rNF) capturablemaximizedifferentiableparamslrrhoeps weight_decayforeachrrr c t|tr'|dkrtdd|kstd|d|cxkrdksntd|d|kstd|d|kstd|t |||||||| } t || dS) Nr zTensor lr must be 1-elementgzInvalid learning rate: rzInvalid rho value: zInvalid epsilon value: zInvalid weight_decay value: )r"r#r$r%rrr&r ) isinstancer numel ValueErrordictsuper__init__) selfr!r"r#r$r%r&rrr defaults __class__s EC:\Projects\Historical\venv\Lib\site-packages\torch/optim/adadelta.pyr-zAdadelta.__init__s b& ! ! zerosrr7 zeros_likepreserve_format) r.rCrGrHrIrJrK has_complexrDr<s r1 _init_groupzAdadelta._init_groupTs x . .Av~ 5+A.. .K  # #A & & &v Q"#OPPP LL JqME5zzQ\*DEK*;*=*=ahOOOOR/@/B/BCCCf ',&6U%:'''l#&+%5U%:&&&k"   u\2 3 3 3   eK0 1 1 1   uV} - - - -r2c|d}|5tj5|}dddn #1swxYwY|jD]}g}g}g}g}g}|d|d|d|d|d|d|d|d f\} } } } } }}}|||||||}t |||||| | | | | |||| |S) zPerform a single optimization step. Args: closure (Callable, optional): A closure that reevaluates the model and returns the loss. Nr"r#r$r%r&rr r) r"r#r$r%r&rr rrY) _cuda_graph_capture_health_checkr? enable_gradr:rZr)r.closurelossrCrGrHrIrJrKr"r#r$r%r&rr rrYs r1r4z Adadelta.steps --///  "$$ ! !wyy ! ! ! ! ! ! ! ! ! ! ! ! ! ! !&- - E-/ "$E(*K')J(*Kd e e n%i j!&'l#  **' ZK  )!-%'     " s AA A)rrrrNN)__name__ __module__ __qualname__rrrAr rboolr-r9rstrrrrZrr4 __classcell__)r0s@r1rrss$'"&"+!$"+"+"+"+ %- "+ "+  "+  "+$"+"+"+"+"+"+"+"+"+H&)CH~)v,)F| ) &\ ) L )&\))))V"==="!=====r2aImplements 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 r!rHrIrJrKr"r#r$r%rr rrYcTtjsI| rGtdt fdt ||DsJddt |||||D]\} }}}}|dz }| s|n| }|dkr|| |}tj| rz*_single_tensor_adadelta.. ]  4 HMT[- - > !==      r2IIf capturable=True, params and state_steps must be on supported devices: .r ralphavalue)r?compiler is_compilingrallzipaddrR view_as_realmul_addcmul_sqrt_clonediv_view_as_complexadd_)r!rHrIrJrKr"r#r$r%rr rrYparamrQrOrPr4stddeltaros @r1_single_tensor_adadeltarsf" > & & ( (wZw'H( ( ( $    v{33      w w wWs v v v  w w 58{J 55%%0tZD  #.tt$ 1  88E866D  E " " ,+J77J*955I%d++D%%dDC%@@@nnS!!'')) c""((**  "KKMME 3T"""s$$UES$AAA  E " " 1)%00E 5 $$$$1%%r2c| r JdtjsI| rGtdt fdt ||DsJddt |dkrdStj|||||g} | D]\\}}}}}}ttt|}ttt|}ttt|}ttt|}ttt|}| rt||||tjs9|djr,tj|tjdd d ntj|d | rtj|}|dkr1| rtj||| ntj||| }tj||tj|||d |z tj||}tj|tj||}tj|tj||tj||tj||tj|||d |z | rGt/|tj r-tj|| tj||tj||| dS)Nz#_foreach ops don't support autogradFrhc3nK|]/\}}|jj|jjko |jjvV0dSr`rkrms r1rpz)_multi_tensor_adadelta..Brqr2rrrsrrcpu)r7rtr rv)r?rxryrrzr{r>r"_group_tensors_by_device_and_dtypevaluesrrr ris_cpu _foreach_add_rB _foreach_neg _foreach_add _foreach_mul__foreach_addcmul__foreach_sqrt_ _foreach_div_r()r!rHrIrJrKr"r#r$r%rr rrYgrouped_tensorsdevice_params_ device_grads_device_square_avgs_device_acc_deltas_device_state_steps__ device_params device_gradsdevice_square_avgsdevice_acc_deltasdevice_state_stepsrdeltasros @r1_multi_tensor_adadeltar+s DDDDD  > & & ( (wZw'H( ( ( $    v{33      w w wWs v v v  w w  6{{aB  Z=O  " " $ $ >B>B  T&\>:: DL-88 !$v,0CDD f/ABB!$v,0CDD   |-?AR   ~**,, 71CA1F1M 7  "ELU$C$C$C3        2A 6 6 6  < -l;;L 1   #L-|TTTTT$1 -|     .444   l!c'     !3S99 S!!!#$5s;; V$$$ FC((( FL111 -s333  166SQQQQ  B*R66 B   , , ,   v 6 6 6 6   vbS A A A A A}>B>Br2)single_tensor_fnFr&c tjs(td|Dst d|t ||d\}}|r-tjrt d|r&tjst}nt}||||||| | | | | ||| dS)zvFunctional API that performs Adadelta algorithm computation. See :class:`~torch.optim.Adadelta` for details. c3JK|]}t|tjVdSr`)r(r?r )rnts r1rpzadadelta..s?55() 1el##555555r2zPAPI has changed, `state_steps` argument must contain a list of singleton tensorsNF) use_fusedz6torch.jit.script not supported with foreach optimizers)r"r#r$r%rr rrY) r?rxryrzrUr jit is_scriptingrr)r!rHrIrJrKrr&r rYr"r#r$r%rrfuncs r1rrs'6 > & & ( ( 55-855522  ^   1 Ne   7U59))++USTTT'uy--//'%&D   !%r2)FNFF)typingrrrrrrr?r optimizerr r r rrrrrrrrrr__all__r__doc__rArdrrrrMr2r1rs:999999999999999 " z "aaaaayaaaJ8     90 j3% L3% <3%f3%V 3% f 3% 3% 3% 3%3%3%3%3%3%3%3%3%laB LaB <aBfaBV aB f aB aB aB aBaBaBaBaBaBaBaBaBH 1HIII" == L= <=f=V = f ==d^=== = = = !="#===JI===r2