:jm`"dZddlmZmZmZddlZddlmZddlmZm Z m Z m Z m Z m Z mZmZmZmZmZmZmZmZmZddgZGd deZd d ed e d ed ed e d ze_deedeedeedeedeededededededededededefdZdeedeedeedeedeededededededededededefdZe e  d#deedeedeedeedeeded!eedededededededededef d"Zy)$z'Implementation for the RAdam algorithm.)castOptionalUnionN)Tensor)_capturable_doc_default_to_fused_or_foreach_differentiable_doc_disable_dynamo_if_unsupported _foreach_doc!_get_capturable_supported_devices_get_scalar_dtype _get_value _maximize_doc _params_doc _to_scalar_use_grad_for_differentiable _view_as_real OptimizerParamsTRAdamradamceZdZ dddddddedeeefdeeefdeded ed e ed ed ed effdZ fdZ dZ e ddZxZS)rFN)foreachmaximize capturabledifferentiableparamslrbetaseps weight_decaydecoupled_weight_decayrrrrc t|tr|jdk7r tdd|kstd|d|kstd|d|dcxkrdksntd|dd|dcxkrdksntd |dd|kstd |t ||||||| || } t ||| y) NrzTensor lr must be 1-elementzInvalid learning rate: zInvalid epsilon value: r?z#Invalid beta parameter at index 0: z#Invalid beta parameter at index 1: zInvalid weight_decay value: ) rr r!r"rrrr#r) isinstancernumel ValueErrordictsuper__init__) selfrrr r!r"r#rrrrdefaults __class__s f/home/mcse/projects/srt_converter/srt-converter-venv/lib/python3.12/site-packages/torch/optim/radam.pyr,zRAdam.__init__ s b& !bhhjAo:; ;by6rd;< <cz6se<= =eAh$$B58*MN NeAh$$B58*MN Nl";L>JK K%!#9)   *cXt|||jD] }|jdd|jdd|jdd|jdd|jdd|dD]}|jj |g}t |dk7s.tj|d rGt|d }|dr*tj|t|j ntj|t |d < y) NrrFrr#rrrstepdtypedevicer5) r+ __setstate__ param_groups setdefaultstategetlentorch is_tensorfloattensorrr6)r-r;grouppp_statestep_valr/s r0r8zRAdam.__setstate__Hs  U#&& E   Y -   Z /   -u 5   5u =   \5 18_ **..B/w<1$U__WV_-M$WV_5H !. $,=,?#\\(:K:MN FO   r1cd}|dD]o}|j|tj|z}|j||jjr t d|j|j|j |} t| dk(r|dr*tjdt|jntjdt | d <tj|tj | d <tj|tj | d <|j| d |j| d |j| d r|S)NFrz'RAdam does not support sparse gradientsrrr4r%r7r3) memory_formatexp_avg exp_avg_sq)gradr> is_complexappend is_sparse RuntimeErrorr;r=zerosrr6rA zeros_likepreserve_format) r-rBparams_with_gradgradsexp_avgs exp_avg_sqs state_steps has_complexrCr;s r0 _init_groupzRAdam._init_group\sM x 2Avv!u//22  ''*66##&'PQQ QVV$ 1 u:?!. B.?.A!((S"\\#5F5HI&M (-'7'7)>)>(E)$+0*:*:)>)>+E,'i 01""5#67""5=17 2:r1c|jd}|$tj5|}ddd|jD]x}g}g}g}g}g}t t t t f|d\} } |j||||||} t|||||| | |d|d|d|d|d|d|d |d | z|S#1swYxYw) zPerform a single optimization step. Args: closure (Callable, optional): A closure that reevaluates the model and returns the loss. Nr rr"r!rrrrr#) beta1beta2rr"r!rrrrr#rX) _cuda_graph_capture_health_checkr> enable_gradr9rtupler@rYr) r-closurelossrBrSrTrUrVrWr[r\rXs r0r3z RAdam.steps --/  ""$ !y !&& E-/ "$E%'H(*K(*KeUl 3U7^DLE5**'+{K  ;">2%Lz*i( .$%56',-E'F'!  > E ! !s CC )gMbP?)g?g+?g:0yE>rFN)__name__ __module__ __qualname__rrr@rr_boolrr,r8rYrr3 __classcell__)r/s@r0rrs$(%1',&+#' $&+&+ %- &+UE\" &+  &+  &+!%&+$&+&+&+&+P(!F"-"-r1aImplements RAdam algorithm. .. math:: \begin{aligned} &\rule{110mm}{0.4pt} \\ &\textbf{input} : \gamma \text{ (lr)}, \: \beta_1, \beta_2 \text{ (betas)}, \: \theta_0 \text{ (params)}, \:f(\theta) \text{ (objective)}, \: \lambda \text{ (weightdecay)}, \:\textit{maximize} \\ &\hspace{13mm} \epsilon \text{ (epsilon)}, \textit{decoupled\_weight\_decay} \\ &\textbf{initialize} : m_0 \leftarrow 0 \text{ ( first moment)}, v_0 \leftarrow 0 \text{ ( second moment)}, \\ &\hspace{18mm} \rho_{\infty} \leftarrow 2/(1-\beta_2) -1 \\[-1.ex] &\rule{110mm}{0.4pt} \\ &\textbf{for} \: t=1 \: \textbf{to} \: \ldots \: \textbf{do} \\ &\hspace{6mm}\textbf{if} \: \textit{maximize}: \\ &\hspace{12mm}g_t \leftarrow -\nabla_{\theta} f_t (\theta_{t-1}) \\ &\hspace{6mm}\textbf{else} \\ &\hspace{12mm}g_t \leftarrow \nabla_{\theta} f_t (\theta_{t-1}) \\ &\hspace{6mm} \theta_t \leftarrow \theta_{t-1} \\ &\hspace{6mm} \textbf{if} \: \lambda \neq 0 \\ &\hspace{12mm}\textbf{if} \: \textit{decoupled\_weight\_decay} \\ &\hspace{18mm} \theta_t \leftarrow \theta_{t} - \gamma \lambda \theta_{t} \\ &\hspace{12mm}\textbf{else} \\ &\hspace{18mm} g_t \leftarrow g_t + \lambda \theta_{t} \\ &\hspace{6mm}m_t \leftarrow \beta_1 m_{t-1} + (1 - \beta_1) g_t \\ &\hspace{6mm}v_t \leftarrow \beta_2 v_{t-1} + (1-\beta_2) g^2_t \\ &\hspace{6mm}\widehat{m_t} \leftarrow m_t/\big(1-\beta_1^t \big) \\ &\hspace{6mm}\rho_t \leftarrow \rho_{\infty} - 2 t \beta^t_2 /\big(1-\beta_2^t \big) \\[0.1.ex] &\hspace{6mm}\textbf{if} \: \rho_t > 5 \\ &\hspace{12mm} l_t \leftarrow \frac{\sqrt{ (1-\beta^t_2) }}{ \sqrt{v_t} +\epsilon } \\ &\hspace{12mm} r_t \leftarrow \sqrt{\frac{(\rho_t-4)(\rho_t-2)\rho_{\infty}}{(\rho_{\infty}-4)(\rho_{\infty}-2) \rho_t}} \\ &\hspace{12mm}\theta_t \leftarrow \theta_t - \gamma \widehat{m_t} r_t l_t \\ &\hspace{6mm}\textbf{else} \\ &\hspace{12mm}\theta_t \leftarrow \theta_t - \gamma \widehat{m_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 `On the variance of the adaptive learning rate and beyond`_. This implementation provides an option to use either the original weight_decay implementation as in Adam (where the weight_decay is applied to the gradient) or the one from AdamW (where weight_decay is applied to the weight) through the decoupled_weight_decay option. When decoupled_weight_decay is set to False (default), it uses the original Adam style weight decay, otherwise, it uses the AdamW style which corresponds more closely to the `author's implementation`_ in the RAdam paper. Further information about decoupled weight decay can be found in `Decoupled Weight Decay Regularization`_. z Args: a lr (float, Tensor, optional): learning rate (default: 1e-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) decoupled_weight_decay (bool, optional): whether to decouple the weight decay as in AdamW to obtain RAdamW. If True, the algorithm does not accumulate weight decay in the momentum nor variance. (default: False) z a .. _On the variance of the adaptive learning rate and beyond: https://arxiv.org/abs/1908.03265 .. _author's implementation: https://github.com/LiyuanLucasLiu/RAdam .. _Decoupled Weight Decay Regularization: https://arxiv.org/abs/1711.05101 rrTrUrVrWr[r\rr"r!r#rrrrXc  tjjs t|}t |D]L\}}| s||n|| }||}||||}tj j s\| rZt}|jj|jjk(r|jj|vs Jd|dtj|rTtj|}tj|}tj|}tj|dz }| r|n t|}|dk7r-| r|jd||zz n|j||}|j|d|z j|j!||d|z d||zz }d||zz ||z }dd|z z dz d|z||zzz z fd} fd }| rBtj"d kD||zd }|j%||z|zd  d kDr(|j%||z|z|zd 7|j%||zd Oy) NIIf capturable=True, params and state_steps must be on supported devices: .rralpha)valuecDdz dz zzdz dz zzz dzS)Nrn?rG)rho_infrho_tsr0 _compute_rectz+_single_tensor_radam.._compute_rectDsI19aKGaK058:  r1c~j}r|j}n|j}dz|z S)Nrq)sqrtaddadd_)exp_avg_sq_sqrtbias_correction2rr!rJs r0_compute_adaptive_lrz2_single_tensor_radam.._compute_adaptive_lrLsB(oo/O"1"5"5c":"1"6"6s";$c)_< jit is_scriptingr enumeratecompiler is_compilingr r6typerL view_as_realrmul_rwlerp_addcmul_whererx)rrTrUrVrWr[r\rr"r!r#rrrrXiparamrKrIstep_tcapturable_supported_devicesr3bias_correction1bias_corrected_exp_avgrtr{updaterzrJrrrss ` ` @@@@r0_single_tensor_radamrs$ 99 ! ! # ^f%PD5'uQxeAhY1+ ^ Q~~**,+L+N ( !!V]]%7%77LL%%)EE \\x[yyz{  F   E "&&u-E%%d+D((1G++J7J ! #vF); 1 % 1rL001xx\x:  dAI&''d!e)'Dud{?ud{?")+;!;q5y/A%!d(eTk25EEE  = [[ ]_/C/EEsF JJ-2V;4J Hs{ **,-$o&  1B6d CaPDr1c \*t|dk(ry| rJdtjjs7| r5t d*t *fdt ||Ds Jd*dt|}tj|||||g}|jD]Y\\}}}}}}ttt|}ttt|}ttt|}ttt|}ttt|}tjjs=|djr.tj|tj dd d ntj|d |rt#||||| rtj$|}d d |z z d z }| rtj&||}tj(|tj|d tj&||}tj*||tj*|d tj,||tj(|tj|||}n?|Dcgc]4}|d t/|z|t/|zzd |t/|zz z z 6}}|dk7rR| rtj*|d ||zz n3| rtj||| ntj0||| }tj2||d |z tj*||tj4|||d |z ~| r7tj6|d} tj6|d }!tj*| |!~!tj*| ||dz |d z z}tj8||}"tj,| |"~"tj:| t | |D#$cgc]\}#}$tj<|$dkD|#d!}%}#}$~ ~|%D%cgc]}%tj<|%dkDdd}&}%tj*|&|tj&||}tj(|tj|d tj,|&|tj(|&tj&||}tj(|tj|d tj:|tj*||tj*|%~%tj(|tj,||~n|D$cgc])}$|$dkDr |$dz |$d z z|z|dz |d z z|$zz dznd+}%}$|%D%cgc] }%|%dkDrdnd }'}%|Dcgc]}d |t/|zz }}t |'|D%(cgc]\}%}(||%z|(z dz}&}%}(t |%|D%(cgc]&\}}%}(d |t/|zz dz||%z|(z zdz(}}%}}(tj>|})tj|)| tj,|)|tj@|)tj|)|&tj4|||)\ycc}wcc}$}#wcc}%wcc}$wcc}%wcc}wcc}(}%wcc}(}%}w)Nrz#_foreach ops don't support autogradF) supports_xlac3K|]N\}}|jj|jjk(xr|jjvPywrb)r6r).0rCr3rs r0 z&_multi_tensor_radam..sQ 4 HHMMT[[-- - > !== > sAArirjr&cpu)r6rkrrnrpr|r%rq)!r=r>rrr allziprr"_group_tensors_by_device_and_dtypevaluesrlistris_cpu _foreach_add_rAr _foreach_neg _foreach_pow _foreach_neg_ _foreach_mul_ _foreach_div_r _foreach_add_foreach_lerp__foreach_addcmul_ _foreach_sub _foreach_mul_foreach_sqrt_r _foreach_sqrt_foreach_reciprocal_)+rrTrUrVrWr[r\rr"r!r#rrrrXgrouped_tensorsgrouped_params_grouped_grads_grouped_exp_avgs_grouped_exp_avg_sqs_grouped_state_steps__grouped_params grouped_gradsgrouped_exp_avgsgrouped_exp_avg_sqsgrouped_state_stepsrrrrz rho_t_listr3numsub2denomnrsrectunrect_step_size unrectifiedbcbufferrs+ @r0_multi_tensor_radamrhsJ$ 6{aDDD  >> & & (Z'H( $ v{3   XXtWuuv w    BBBB +{;O  " " $ [J  d6lO<T&\>: V .?@"4<1EF"4<1EF ~~**,1DQ1G1N1N   #U\\#e%DC     3Q 7   /?AT  !..}=Mq5y/A% $11%9LM     0 1    0! 4$11%9LM     02E F    0! 4    02B C    0 1    0' :)J0   T"#Jt,,.u 4 00022J 1 %##NA\8I4IJ''%~\%*$6$6%~\%M -}a%iH /7   q5y  $$Z3C%%j!4D   T *   W -{w{3G&&z7;E   U +   %BES*AU5=Q ECKC0DLPQD D1Hc3 ?Q Q    0" 5$11%9LM     0 1    0! 4    02B C    0 1$11%9LM     0 1    0! 4  !1 2    0" 5    0$ 7    0 1    02B C ( 19 QYqy" ! ! 4u<>   D ?CCdq1c1CKC;N 26EZ---   7:+GW6X *2$dR2%    '**=tEU&V  "D$ez$///C7BINKbP    $$%89 FC( F$45 ""6* F$45 0@&Iw[Jd^ R* D   s09\*$\!\ 0.\$\:\#\! +\' )single_tensor_fnrc Btd|Ds td|t||d\}}|r)tjj r td|r%tjj st }nt}||||||| | | ||| |||| y)zpFunctional API that performs RAdam algorithm computation. See :class:`~torch.optim.RAdam` for details. c3PK|]}t|tj ywrb)r'r>r)rts r0rzradam..Is@qz!U\\*@s$&zPAPI has changed, `state_steps` argument must contain a list of singleton tensorsNF) use_fusedz6torch.jit.script not supported with foreach optimizers) r[r\rr"r!rr#rrrX)rrOr r>r}r~rr)rrTrUrVrWr#rrrrXrr[r\rr"r!rfuncs r0rr/s4 @K@ @ ^  1 Ne 7599))+STTuyy--/"#  ! 5%r1)FNFFFF)__doc__typingrrrr>r optimizerrr r r r r rrrrrrrrr__all__rrr@rfrrrrGr1r0rs.(( & G NINd2f        gK `eD LeD <eD6leDf eD f eD eD eD eDeD eD!eDeDeDeD !eDPDJ LDJ <DJ6lDJf DJ f DJ DJ DJ DJDJ DJ!DJDJDJDJ !DJN 1EF$)" ; L; <;6l;f ; f ;!;d^;;;;; ; !;" #;$%;& ';G;r1