9i,SSKJrJrJr SSKrSSKJr SSKJrJrJ r J r J r J r J r JrJrJrJrJrJrJrJrJrJrJrJrJr SS/r"SS\5rS S \S \ S \S \S \ S \ S 3 -\lS\\S\\S\\S\\S\\S\\S\\S\\S\S\S\\ \4S\\ \4S\\ \4S\ S\ S\S\S\S \4&S!jr!S\\S\\S\\S\\S\\S\\S\\S\\S\S\S\\ \4S\\ \4S\\ \4S\ S\ S\S\S\S \4&S"jr"S\\S\\S\\S\\S\\S\\S\\S\\S\S\S\ S\ S\\ \4S\ S\ S\S\S\S \S#S4(S$jr#\ "\!S%9S)S\\S\\S\\S\\S\\S\\S&\\S\S\S'\\S\\S\\S\S \S\S\ S\ S\\ \4S\ S\ S\4*S(jj5r$g)*)castOptionalUnionN)Tensor)_capturable_doc_default_to_fused_or_foreach_device_dtype_check_for_fused_differentiable_doc_disable_dynamo_if_unsupported _foreach_doc _fused_doc!_get_capturable_supported_devices_get_scalar_dtype _get_value _maximize_doc _params_doc_stack_if_compiling _to_scalar_use_grad_for_differentiable _view_as_real DeviceDictDeviceDtypeDict OptimizerParamsTAdamadamc^\rSrSrSSSSSSSS.S\S\\\4S\\\\4\\\44S\S \S \ S \ \ S \ S \ S\ S\ \ S\ 4U4Sjjjjr U4Sjr Sr \SSj5rSrU=r$)r"FN)foreachmaximize capturabledifferentiablefuseddecoupled_weight_decayparamslrbetaseps weight_decayamsgradr r!r"r#r$r%c f>[U[5(a8U(aU (d [S5eUR5S:wa [S5eSU::d[SU35eSU::d[SU35eSUSs=::aS:dO [S US35eSUSs=::aS:dO [S US35eSU::d[S U35e[US[5(a[US[5(d;[US[5(a[US[5(d [S 5e[US[5(a;U (dU(a [S 5eUSR5S:wa [S5e[US[5(a;U (dU(a [S5eUSR5S:wa [S5eUUUUUUUU U U U S. n [ TU]X5 U (a,U (a [S5eSUlU(a [S5egg)NElr as a Tensor is not supported for capturable=False and foreach=TruerTensor 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: z0betas must be either both floats or both TensorszKbetas[0] as a Tensor is not supported for capturable=False and foreach=Truez!Tensor betas[0] must be 1-elementzKbetas[1] as a Tensor is not supported for capturable=False and foreach=Truez!Tensor betas[1] must be 1-element) r'r(r)r*r+r!r r"r#r$r%z)`fused` does not support `differentiable`Tz0`fused` and `foreach` cannot be `True` together.) isinstancer ValueErrornumelfloatsuper__init__ RuntimeError_step_supports_amp_scaling)selfr&r'r(r)r*r+r r!r"r#r$r%defaults __class__s P/var/www/html/land-doc-ocr/venv/lib/python3.13/site-packages/torch/optim/adam.pyr6 Adam.__init__#s+ b& ! !z [xxzQ !>??by6rd;< <cz6se<= =eAh$$B58*MN NeAh$$B58*MN Nl";L>JK K a% ( (Za%-H-H58V,,E!Hf1M1MOP P eAh ' '' aQx~~1$ !DEE eAh ' '' aQx~~1$ !DEE( $,&<   * "#NOO.2D + "#UVV c >[TU]U5 URGHDnURSS5 URSS5 URSS5 URSS5 URSS5 URSS5 URSS5nUS HnURR U/5n[ U5S :wdM0[R"US 5(aMP[US 5nUS(d US(a'[R"U[US 9URS 9O[R"U[5S9US 'M GMG g)Nr+Fr!r r"r#r%r$r&rstepis_fuseddtypedevicerD) r5 __setstate__ param_groups setdefaultstategetlentorch is_tensorr4tensorrrE)r9rJgroupr$pp_statestep_valr;s r<rGAdam.__setstate__rs/ U#&&E   Y .   Z /   Y -   \5 1   -u 5   5u =$$Wd3E8_**..B/w<1$U__WV_-M-M$WV_5H!.%.  $"3U"C#$88 #\\(:K:MNFO %'r>cSnUSGHKn U RcMU[R"U 5-nURU 5 U RR(a [ S5eURU R5 UR U n [U 5S:XaUS(a [U 5 US(d US(a*[R"S[USS9U RS 9O[R"S [5S 9U S '[R"U [RS 9U S'[R"U [RS 9U S'US(a&[R"U [RS 9U S'URU S5 URU S5 US(aURU S5 US(aU S R(a [ S5eUS(a3[R "US5(aUS(d [ S5eURU S 5 GMN U$)NFr&zJAdam does not support sparse gradients, please consider SparseAdam insteadrr$r"rArCr/rFr@) memory_formatexp_avg exp_avg_sqr+max_exp_avg_sqr#zB`requires_grad` is not supported for `step` in differentiable moder r'r-)gradrM is_complexappend is_sparser7rJrLr zerosrrErO zeros_likepreserve_format requires_gradrN) r9rPparams_with_gradgradsexp_avgs exp_avg_sqsmax_exp_avg_sqs state_steps has_complexrQrJs r< _init_groupAdam._init_groups xAvv!u//22  ''*66##&d QVV$ 1 u:?W~5a8!.%.  "3U7^"L#$88 #\\#5F5HI&M(-'7'7)>)>(E)$+0*:*:)>)>+E,'Y'272B2BU-B-B3./i 01""5#67##**51A+BC)*uV}/J/J&\ )$d 44!,/&_""5=1{!|r>cUR5 SnUb%[R"5 U"5nSSS5 URHn/n/n/n/n/n/n USupUR UUUUUUU 5n [ UUUUUU 4USU U U USUSUSUSUSUS US US [ US S5[ US S5USS.6 M U$!,(df  N=f)zPerform a single optimization step. Args: closure (Callable, optional): A closure that reevaluates the model and returns the loss. Nr(r+r'r*r)r!r r"r#r$ grad_scale found_infr%)r+ribeta1beta2r'r*r)r!r r"r#r$rmrnr%) _cuda_graph_capture_health_checkrM enable_gradrHrjrgetattr) r9closurelossrPrcrdrerfrgrhrorpris r<r@ Adam.steps9 --/  ""$y%&&E-/ "$E%'H(*K,.O(*K >LE** K    i(';">2%Lz*i( .$%56Gn"4t<!$ T:',-E'F+ ''V ]%$s C  C)r8)gMbP?)g?g+?g:0yE>rFN)__name__ __module__ __qualname____firstlineno__rrr4rtupleboolrr6rGrjrr@__static_attributes__ __classcell__)r;s@r<rr"s$(COMW#' $ $',MWMW %- MWU5&=)5+??@ MW  MW  MWMW$MWMWMWMW~MW!%MWMW^0IV"9"9r>af Implements Adam 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)} \\ &\hspace{13mm} \lambda \text{ (weight decay)}, \: \textit{amsgrad}, \:\textit{maximize}, \: \epsilon \text{ (epsilon)} \\ &\textbf{initialize} : m_0 \leftarrow 0 \text{ ( first moment)}, v_0\leftarrow 0 \text{ (second moment)},\: v_0^{max}\leftarrow 0 \\[-1.ex] &\rule{110mm}{0.4pt} \\ &\textbf{for} \: t=1 \: \textbf{to} \: \ldots \: \textbf{do} \\ &\hspace{5mm}\textbf{if} \: \textit{maximize}: \\ &\hspace{10mm}g_t \leftarrow -\nabla_{\theta} f_t (\theta_{t-1}) \\ &\hspace{5mm}\textbf{else} \\ &\hspace{10mm}g_t \leftarrow \nabla_{\theta} f_t (\theta_{t-1}) \\ &\hspace{5mm}\textbf{if} \: \lambda \neq 0 \\ &\hspace{10mm} g_t \leftarrow g_t + \lambda \theta_{t-1} \\ &\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 m_t/\big(1-\beta_1^t \big) \\ &\hspace{5mm}\textbf{if} \: amsgrad \\ &\hspace{10mm} v_t^{max} \leftarrow \mathrm{max}(v_{t-1}^{max},v_t) \\ &\hspace{10mm}\widehat{v_t} \leftarrow v_t^{max}/\big(1-\beta_2^t \big) \\ &\hspace{5mm}\textbf{else} \\ &\hspace{10mm}\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 `Adam: A Method for Stochastic Optimization`_. z Args: a lr (float, Tensor, optional): learning rate (default: 1e-3). A tensor LR is not yet supported for all our implementations. Please use a float LR if you are not also specifying fused=True or capturable=True. 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): if True, this optimizer is equivalent to AdamW and the algorithm will not accumulate weight decay in the momentum nor variance. (default: False) amsgrad (bool, optional): whether to use the AMSGrad variant of this algorithm from the paper `On the Convergence of Adam and Beyond`_ (default: False) z a= .. Note:: A prototype implementation of Adam and AdamW for MPS supports `torch.float32` and `torch.float16`. .. _Adam\: A Method for Stochastic Optimization: https://arxiv.org/abs/1412.6980 .. _On the Convergence of Adam and Beyond: https://openreview.net/forum?id=ryQu7f-RZ r&rdrerfrgrhrmrnr+rirorpr'r*r)r!r"r#r%c  UcUbe[RR5(aF[U [5(de[U [5(de[U [5(deO [ U 5n [U [ 5(aU RU R4U 0nOSn[U5GHunnU(dUUOUU*nUUnUUnUUn[RR5(ddU(a][5nURRURR:XaURRU;d SUS35eUS- nU S:waU(aURSX-- 5 OoU(aX[U [ 5(aCU R(a!UR!UR#5U 5nO!UR%UU S9nOUR%UU S9n[R&"U5(a{[R("U5n[R("U5n[R("U5nU(a[R("UU5UU'[R("U5nURnUb0URnUU4nUU;aU R+UUSS9UU'UUnOU nUR-USU- 5 U(a[U [ 5(akU R(a(UR-[R."U5SU - S9 OUURU 5R!UU[1[SU - 5S 9 O#URU 5R!UUSU - S 9 U(dU(GaUnU(aF[U [ 5(a1U R(aSU UR#5-- n OSU U-- n OSU U-- n U(aF[U [ 5(a1U R(aSU UR#5-- n!OSU U-- n!OSU U-- n!U U - n"U"R35n#U!R55n$U(auU(aUUR#5n%OUUn%UUR7[R8"U%U55 UUR55U$U#-- R;UU#- 5n&O(UR55U$U#-- R;UU#- 5n&U(a!UR=UR#5U&5 OUR=UU&5 O[?U5nSU U-- n SU U-- n!U U - n"U!S -n$U(aB[R8"UUUUUS 9 UUR55U$- R;U5n&O"UR55U$- R;U5n&UR=UU&U"*S 9 U(dGMH[R&"UU5(dGMi[R@"UU5UU'GM g) NIIf capturable=True, params and state_steps must be on supported devices: .rralphaT)rErD non_blocking)weight)value?)out)!rMjit is_scriptingr1r4rrrErD enumeratecompiler is_compilingrtypemul_rbaddcmul_cloneaddr\ view_as_realtolerp_squarernegsqrtcopy_maximumadd_addcdiv_rview_as_complex)'r&rdrerfrgrhrmrnr+rirorpr'r*r)r!r"r#r% beta1_dictiparamr[rXrYstep_tcapturable_supported_devicesrErDkey device_beta1r@bias_correction1bias_correction2 step_size step_size_negbias_correction2_sqrtrZdenoms' r<_single_tensor_adamrYs,  )"33 3 yy"e$$$$%''''%'''' ^%  27,, 1Le0T  f%5'uQxeAhY1+ ^ Q~~**,,+L+N ( !!V]]%7%77LL%%)EE \\x[yyz{  F !  1 % 1r001"jv&F&F#11#}}U[[]LI#xx\xB88E8>D   E " "%%d+D((1G++J7J%*%7%78J%K"&&u-E  !KKE5/C*$"'((!T#+# 32)NO$s]*+w}}6w.f%D 5$;  5$; --I$4c$9 ! oa0*/RSBTU)+0025JJPPQTU#*-BBHHM NN7E)N < 7u''q 22!&!6!6q7I!JOA q&r>c 4^-[U5S:Xag[U [5(a1U(d [S5eU R 5S:wa [ S5e[U [5(a1U(d [ S5eU R 5S:wa [ S5e[U [5(a1U(d [ S5eU R 5S:wa [ S5e[ RR5(d>U(a7[S S 9m-[U-4S j[X555(d S T-S 35eUcUbeU(aS5e[U 5n [R"XX#XE/5n[U [5(a'[U R 5S:waU R U 0OSnUR#5GH3uunnnnnnn[%[&[U5n[%[&[U5n[%[&[U5n[%[&[U5n[%[&[U5n USR n!UbU!U;aU R)U!SS9UU!'U(aUU!OU n"U (a<U(a'[%[&[U5n#[+UUUUU#5 O[+UUUU5 U(a[ R,"U5n[ RR5(d>U SR.(a*[ R0"U [ R2"SSS9SS9 O[ R0"U S5 U S:waXU(a[ R4"USX-- 5 O4U(a[ R0"UUU S9 O[ R6"UUU S9n[ R8"UU[%[:SU"- 55 [ R4"UU 5 [U [ R5(a[ R<"USU - 5n$Sn%OUn$SU - n%[ R>"UU$UU%5 AA$U(Ga{[ R@"U U 5n&[ R@"U U 5n'[ RB"U&S5 [ RB"U'S5 [ RD"U'5 [ RF"U&U 5 [ RH"U&5 [ RJ"U'5 U&n(U'n)U(aE[%[&[U5n#[ RL"U#U5 [ RN"U#5n*O[ RN"U5n*[ RF"U*U)5 [ R0"U*U5 [ RF"U*U(5 [ RP"UUU*5 GMU V+s/sHn+SU [SU+5-- PM n&n+U V+s/sHn+SU [SU+5-- PM n'n+[UU&V,s/sH n,U U,- S-PM sn,5n(U'V,s/sHn,U,S-PM n)n,U(aE[%[&[U5n#[ RL"U#U5 [ RN"U#5n*O[ RN"U5n*[ RF"U*U)5 [ R0"U*U5 [ RP"UUU*U(5 GM6 gs sn+fs sn+fs sn,fs sn,f)Nrr-rr.zHbeta1 as a Tensor is not supported for capturable=False and foreach=TruezTensor beta1 must be 1-elementzHbeta2 as a Tensor is not supported for capturable=False and foreach=TruezTensor beta2 must be 1-elementF) supports_xlac3># UHTupURRURR:H=(a URRT;v MV g7frw)rEr).0rQr@rs r< %_multi_tensor_adam..VsN 4 HHMMT[[-- - > !== >3sAArrz#_foreach ops don't support autogradcpuTrErr0)rErr)+rLr1rr7r3r2rMrrrallziprr"_group_tensors_by_device_and_dtypestrrEvaluesrlistrr _foreach_negis_cpu _foreach_add_rO _foreach_mul_ _foreach_add_foreach_lerp_r4 _foreach_mul_foreach_addcmul_ _foreach_pow _foreach_sub_ _foreach_neg_ _foreach_div__foreach_reciprocal__foreach_sqrt__foreach_maximum_ _foreach_sqrt_foreach_addcdiv_rr).r&rdrerfrgrhrmrnr+rirorpr'r*r)r!r"r#r%grouped_tensorsrdevice_params_ device_grads_device_exp_avgs_device_exp_avg_sqs_device_max_exp_avg_sqs_device_state_steps__ device_params device_gradsdevice_exp_avgsdevice_exp_avg_sqsdevice_state_stepsrErdevice_max_exp_avg_sqsscaled_device_gradsrrrrrexp_avg_sq_sqrtr@bcrs. @r<_multi_tensor_adamr sp, 6{a"fW  88:?:; ;%  Z  ;;=A => >%  Z  ;;=A => > >> & & ( (Z'H( $ v3    XXtWuuv w     )"33 3DDD  BB BB LO eV $ $U\\):e)C u   " " $  T&\>: DL-8 tF|-=>!$v,0CD!$v,0CDq!((  !fJ&>!&d!KJv -7z&)U  )-d6l@RS#("5"56L"M"'"5"56H"I   1F G    5    ;  # #M?O T;M :L$EZ---:L  ;M :L$EZ---:L  ,FV,WFVb2g^FV,WXI7G$H7GRW7G !$H)-d6l@RS#("5"56L"M"'"5"56H"I   1F G    5  # #  k %|  -X$Hs"\\ )\ \returnc :U(dgU(a [S5eUbURU0O0nUbURU0O0n[U [5(a'[ U R5S:waU RU 0OSn[ R "XX#XE/5nUR5GHouunnuunnnnnnn[[[U5n[[[U5n [[[U5n![[[U5n"[[[U5n#Sun$n%Ub URUURUSS95n$Ub URUURUSS95n%UbUU;aU RUSS9UU'UUn [R"U#S5 U(d[RO[Rn&U&"UU U!U"UU#UU U U U UUU$U%S9 U%cGMK[R "U#U%/[#U#5-5 GMr g) Nz9Adam with fused=True does not support differentiable=Truer)NNT)rrr) r+r'rorpr*r)r!rmrn)r7rEr1rrrritemsrrrIrrMr _fused_adam_ _fused_adamw_rrL)'r&rdrerfrgrhrmrnr+rirorpr'r*r)r!r"r#r%grad_scale_dictfound_inf_dictlr_dictrrErrrrrrrrrrrrdevice_grad_scaledevice_found_inffuncs' r< _fused_adamrsC, VWW,6+A  J'r*3)>  9%B&b&11c"))n6MBSW  BB LO            "  T&\>: DL-8 tF|-=>!$v,0CD!$v,0CD.8++  ! / : : f4 @!   -88 V$ ?    6#8 ee6eEGFOB .2)?u!!UEXEX     " %(& "  '   "%5$6=O9P$P O !r>)single_tensor_fnr r$cU c5Uc2[XSS9unnU(a[U[5(a U(dSnU cSn UcSn[RR 5(d"[ SU55(d [S5eU(a.[RR5(a [S5eU (a.[RR5(a [S5eU (a*[RR5(d[nO7U(a*[RR5(d[nO[nU"UUUUUU4UU UUUUUUUUU U U S. 6 g) zfFunctional API that performs Adam algorithm computation. See :class:`~torch.optim.Adam` for details. NF) use_fusedc3V# UHn[U[R5v M! g7frw)r1rMr)rts r<radam..s!5-8 1ell##[s')zPAPI has changed, `state_steps` argument must contain a list of singleton tensorsz6torch.jit.script not supported with foreach optimizersz4torch.jit.script not supported with fused optimizers) r+rirorpr'r*r)r!r"r#rmrnr%) r r1rrMrrrr7rrrrr)r&rdrerfrgrhr r"r#r$rmrnrir%r+rorpr'r*r)r!rrs r<rrusRF }1 e 7 z"f--jG } >> & & ( (5-8522 ^  599))++STT ''))QRR UYY++-- //11!"   ! %5'r>)NFFNNNFF)%typingrrrrMr optimizerrr r r r r rrrrrrrrrrrrrr__all__r__doc__rr}r4rrrrrVr>r<rs6(( 0 6 m9mb$J         +KB NDK LDK <DK6lDKf DK &\ DK f DK DKDKDKDK  DK  DK eVmDKDK !DK"#DK$%DK&'DK(!)DKNr Lr <r6lrf r &\ r f r rrrr  r  r eVmrr !r"#r$%r&'r(!)rj] L] <]6l]f ] &\ ] f ] ]]]] ] ] eVm]] !]"#]$%]&'](!)]* +]@ 1DE#  #'"&#(!W LW <W6lWf W &\ W f Wd^WWW D>W WWW !!W$%W& 'W( )W* eVm+W,-W. /W01WFWr>