"EinR#SSKJr SSKrSSKJr SSKJrJrJrJrJ r J r J r J r J r JrJrJrJr SS/r"SS\5rS S \ S \S \ S \S 3 -\lS$S\\S\\S\\S\\S\S-S\S-S\S-S\S\S-S\S\S\S\S\S\S\SS4"SjjrS rS\\S\\S\\S\\S\S-S\S-S\S\S\S\S\S\S\S\SS4S!jrS\\S\\S\\S\\S\S-S\S-S\S\S\S\S\S\S\S\SS4S"jrS\\S\\S\\S\\S\S-S\S-S\S\S\S\S\S\S\S\SS4S#jrg)%)castN)Tensor) _default_to_fused_or_foreach_device_dtype_check_for_fused_differentiable_doc _foreach_doc_get_scalar_dtype _get_value _maximize_doc _params_doc _to_scalar_use_grad_for_differentiable _view_as_real OptimizerParamsTAdagradadagradc^\rSrSrSSSSS.S\S\\-S\S\S \S \S \S-S \S \S\S-SS4U4SjjjjrU4Sjr SSjr Sr \ SSj5r SrU=r$)rNF)maximizedifferentiablefusedparamslrlr_decay weight_decayinitial_accumulator_valueepsforeachrrrreturnc p>[U[5(aUR5S:wa [S5eSU::d[SU35eSU::d[SU35eSU::d[SU35eSU::d[SU35eSU::d[SU35eUUUUUUUU U S . n [TU]X5 U (a+U (a [ S 5eU(a [ S 5eS UlURHn U S Hn URU nU S(a*[R"S[U SS9U RS9O[R"S[5S9US'[R"U 5(a [!XU5OUn[R""X[R$S9US'M M g)NrzTensor lr must be 1-elementgzInvalid learning rate: zInvalid lr_decay value: zInvalid weight_decay value: z)Invalid initial_accumulator_value value: zInvalid epsilon value: ) rrrrrr rrrz)`fused` does not support `differentiable`z0`fused` and `foreach` cannot be `True` together.Trris_fused)dtypedevicer&step) memory_formatsum) isinstancernumel ValueErrorsuper__init__ RuntimeError"_need_device_dtype_check_for_fused param_groupsstatetorchzerosr r'tensor is_complexcomplex full_likepreserve_format)selfrrrrrrr rrrdefaultsgrouppr4 init_value __class__s P/home/mande/repo/quber/.venv/lib/python3.13/site-packages/torch/optim/adagrad.pyr0Adagrad.__init__s b& ! !bhhjAo:; ;by6rd;< <h7zBC Cl";L>JK K//;[TU]U5 SnURHKnURSS5 URSS5 URSS5 URSS5nMM [ UR R 55n[U5S:g=(a [R"USS5nU(d5UH.n[R"[US5[US9S 9US'M0 gg) Nr rFrrrr)r$r() r/ __setstate__r3 setdefaultlistr4valueslenr5 is_tensorr7floatr )r<r4rr> state_valuesstep_is_tensorsrAs rBrFAdagrad.__setstate__bs U#&&E   Y -   Z /   -u 5$$Wd3E ' DJJ--/0 l+q0 eoo OF #7 !!LL!F)$,=u,M& "rDcURH1nUSH%nURUnUSR5 M' M3 g)z6Calls tensor.share_memory_() on the state sum tensors.rr+N)r3r4 share_memory_)r<r>r?r4s rB share_memoryAdagrad.share_memoryws=&&E8_ 1 e **,%'rDcSupgUSHnURcMUS(a#[USS5(a[USS9 SUlXhRR-nU[ R "U5-nURU5 URUR5 URUn URU S5 URU S 5 M Xg4$) N)FFrrr2T)cuda_unsupportedFr+r)) gradgetattrrr2 is_sparser5r8appendr4) r<r>params_with_gradgrads state_sums state_stepshas_sparse_grad has_complexr?r4s rB _init_groupAdagrad._init_group~s'3$xAvv!>g8'' 2!dK>CD;66#3#33u//22  ''* QVV$ 1 !!%,/""5=1!"++rDchSnUb%[R"5 U"5nSSS5 URHfn/n/n/n/nURX4XVU5up[ UUUUUSUSUSUSUUSUSUSU US [ US S5[ US S5S 9 Mh U$!,(df  N=f) zPerform a single optimization step. Args: closure (Callable, optional): A closure that reevaluates the model and returns the loss. Nrrrrr rrr grad_scale found_inf) rrrrr_r rrr`rrdre)r5 enable_gradr3rarrX) r<closurelossr>r[r\r]r^r_r`s rBr) Adagrad.steps  ""$y%&&E-/ "$E')J(*K+/+;+;K, (O  ;">2z*%L /i(z*$%56'Gn"4t<!$ T:! ': A%$s B## B1)r2)g{Gz?rrrg|=N)r!NN)__name__ __module__ __qualname____firstlineno__rrLrboolr0rFrSrarr)__static_attributes__ __classcell__)rAs@rBrrs"+,#D$!DD FND D  D $) DDDDDd{D DDL*-,*"*"*rDa[Implements Adagrad algorithm. .. math:: \begin{aligned} &\rule{110mm}{0.4pt} \\ &\textbf{input} : \gamma \text{ (lr)}, \: \theta_0 \text{ (params)}, \: f(\theta) \text{ (objective)}, \: \lambda \text{ (weight decay)}, \\ &\hspace{12mm} \tau \text{ (initial accumulator value)}, \: \eta\text{ (lr decay)}\\ &\textbf{initialize} : state\_sum_0 \leftarrow \tau \\[-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} \tilde{\gamma} \leftarrow \gamma / (1 +(t-1) \eta) \\ &\hspace{5mm} \textbf{if} \: \lambda \neq 0 \\ &\hspace{10mm} g_t \leftarrow g_t + \lambda \theta_{t-1} \\ &\hspace{5mm}state\_sum_t \leftarrow state\_sum_{t-1} + g^2_t \\ &\hspace{5mm}\theta_t \leftarrow \theta_{t-1}- \tilde{\gamma} \frac{g_t}{\sqrt{state\_sum_t}+\epsilon} \\ &\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 `Adaptive Subgradient Methods for Online Learning and Stochastic Optimization`_. z Args: a lr (float, Tensor, optional): learning rate (default: 1e-2) lr_decay (float, optional): learning rate decay (default: 0) weight_decay (float, optional): weight decay (L2 penalty) (default: 0) initial_accumulator_value (float, optional): initial value of the sum of squares of gradients (default: 0) eps (float, optional): term added to the denominator to improve numerical stability (default: 1e-10) z a fused (bool, optional): whether the fused implementation (CPU only) is used. Currently, `torch.float64`, `torch.float32`, `torch.float16`, and `torch.bfloat16` are supported. (default: None). Please note that the fused implementations does not support sparse or complex gradients. .. _Adaptive Subgradient Methods for Online Learning and Stochastic Optimization: http://jmlr.org/papers/v12/duchi11a.html rr\r]r^rrdrer_r rr`rrrrrr!c L[SU55(d [S5eUcUc [X SS9unnUcSnUcSnU(a.[RR 5(a [S5eU(a.[RR 5(a [S5eU(a*[RR 5(d[ nO7U(a*[RR 5(d[nO[nU"UUUUU U U UUUU U UUS9 g) zlFunctional API that performs Adagrad algorithm computation. See :class:`~torch.optim.Adagrad` for details. c3V# UHn[U[R5v M! g7frj)r,r5r).0ts rB adagrad.. s@Kqz!U\\**Ks')zPAPI has changed, `state_steps` argument must contain a list of singleton tensorsNF) use_fusedz6torch.jit.script not supported with foreach optimizersz4torch.jit.script not supported with fused optimizers rrrrr_rrr`rdre) allr1rr5jit is_scripting_fused_adagrad_multi_tensor_adagrad_single_tensor_adagrad)rr\r]r^rrdrer_r rr`rrrrr_funcs rBrrs2 @K@ @ @ ^   }1 e 7 }599))++STT ''))QRR UYY++-- //11$%  ! '%rDcPUR5n[R"XU5$rj)sizer5sparse_coo_tensor)rW grad_indicesrIrs rB _make_sparser>s 99;D  " "< >>rDc UcUb [S5e[RR5(d [ U5n[ XX#SS9GH upnnUS- n[ U5nU (dUOU*nUS:wa+UR(a [S5eURXS9nUSUS- U--- nUR(aUR5nUR5nUR5nUR[UUURS555 UR!U5nUR5R#5RU 5nUR[UUUU- 5U*S9 GM/[R$"U5nU(aB[R&"U5n[R&"U5n[R&"U5nUR)XSS 9 U (aUR+5U -nOUR+5RU 5nUR-UUU*S 9 U(dGM[R."U5n[R."U5nGM# g) N,Expected grad_scale and found_inf to be NoneT)strictrrz;weight_decay option is not compatible with sparse gradientsalphavalue)AssertionErrorr5r{r|rzipr rYr1addcoalesce_indices_valuesadd_rpow sparse_masksqrt_r8 view_as_realaddcmul_sqrtaddcdiv_view_as_complex)rr\r]r^rdrerrrrr_rrr`paramrW state_sumstep_tr)clrr grad_valuesstd std_valuesr8s rBrrCs"!6KLL 99 ! ! # # ^*-zt+&Y ! &!#t$ 1 ~~"Q88E86DAX--. >>==?D==?L,,.K NN<lKOOAUSR (a*[R""U[R$"SSS 9SS 9 O[R""US 5 US:wa4U (a[R""UUUS 9 O[R&"UUUS 9nUVs/sHnU*S [)U5S - U--- PM nn[R*"UUUS S 9 [R,"U5n[R""UU 5 US:wdU (a[R."UU5 UnO[R0"UU5n[R2"UUU5 GMJ gs snf) Nz#_foreach ops don't support autogradrrc38# UHoRv M g7frj)rY)rtrWs rBrv(_multi_tensor_adagrad..s9 '3tNN|sTryg?cpu)r'rrr)rrJrr"_group_tensors_by_device_and_dtyperIrrHranyrrr5 _foreach_negcompiler is_compilingis_cpu _foreach_add_r7 _foreach_addr _foreach_addcmul_ _foreach_sqrt _foreach_mul_ _foreach_mul_foreach_addcdiv_)rr\r]r^rdrerrrrr_rrr`grouped_tensorlistsdevice_params_ device_grads_device_state_sums_device_state_steps_r device_params device_gradsdevice_state_sumsdevice_state_stepsdevice_has_sparse_gradr) minus_clrr numerators rBr~r~s"BCC!6KLL 6{a BB#FF  0 & & (   T&\>: DL-8  f/AB!$v,0CD!0" S9 '39 6  " "!")! $!-'%#    -7H I  --l;L ~~**,,1CA1F1M1M   "ELLU$C3     2A 6 1 ##L-|T$11 -|  GY FXdRC1 4(1,88 9FX    1<UVW!!"34 C% 1     i 8$I**<CI  y#>Q )p s0"J4cVU(dgU (dU (a [S5eU (a [S5e[U5nUbURU0OSnUbURU0OSn[R"XX#/5nUR 5GHuunnuunnnnn[ [[U5n[ [[U5n[ [[U5n[ [[U5nSunnUb!UbUU;aURUSS9UU'UUnUb UbX_;aURUSS9UU'UUn[R"US5 [R"UUUUUUUU U UUS9 UcM[R"UU/[U5-5 GM g)Nz5`fused` does not support sparse grad or complex paramz: DL-8  f/AB!$v,0CD.8++  !o&A_,*4--T-*R' / 7   ^%?.)2f4)Pv&-f5  .2     %(&   '   "%5$6=O9P$P = !rD)NNNFNFF)typingrr5r optimizerrrrr r r r r rrrrr__all__r__doc__rHrorLrrrr~r}r#rDrBrs " i cicN4      5.p $#" G LG <GV Gf G $; G  G}GGD[GGG G !G"#G$ %G&'G( )GT? A= LA= <A=V A=f A=  A= } A= A=A=A= A=A=A=A=A=  !A=Hl? Ll? <l?V l?f l?  l? } l? l?l?l? l?l?l?l?l?  !l?^M LM <MV Mf M  M } M MMM MMMMM  !MrD