inR#vddlmZddlZddlmZddlmZmZmZmZm Z m Z m Z m Z m Z mZmZmZmZddgZGddeZd d e d ed e d ed ze_ d$deedeedeedeededzdedzdedzdededzdedededededededdf"dZd Zdeedeedeedeededzdedzdededededededededdfd!Zdeedeedeedeededzdedzdededededededededdfd"Zdeedeedeedeededzdedzdededededededededdfd#Zy)%)castN)Tensor) _default_to_fused_or_foreach_device_dtype_check_for_fused_differentiable_doc _foreach_doc_get_scalar_dtype _get_value _maximize_doc _params_doc _to_scalar_use_grad_for_differentiable _view_as_real OptimizerParamsTAdagradadagradceZdZ ddddddedeezdededed ed edzd ed ed edzddffdZfdZddZ dZ e ddZ xZ S)rNF)maximizedifferentiablefusedparamslrlr_decay weight_decayinitial_accumulator_valueepsforeachrrrreturnc 6t|tr|jdk7r tdd|kstd|d|kstd|d|kstd|d|kstd|d|kstd||||||||| | d } t||| | r!| r t d |r t d d |_|jD]} | d D]} |j| }| dr/tjdt| d| jntjdt|d<tj| r t!||n|}tj"| |tj$|d<y)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: ) rrrrrrrrrz)`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)selfrrrrrrrrrrdefaultsgrouppr3 init_value __class__s h/Volumes/fast/ai/experiments/voice-extract-mac/.venv/lib/python3.12/site-packages/torch/optim/adagrad.pyr/zAdagrad.__init__s b& !bhhjAo:; ;by6rd;< <h7zBC Cl";L>JK K//;r3s rA share_memoryzAdagrad.share_memoryws=&&E8_ 1 e **,%'rBcd\}}|dD]}|j|dr!t|ddrt|dd|_||jjz}|t j |z}|j||j|j|j|} |j| d|j| d ||fS) N)FFrrr1T)cuda_unsupportedFr*r() gradgetattrrr1 is_sparser4r7appendr3) r;r=params_with_gradgrads state_sums state_stepshas_sparse_grad has_complexr>r3s rA _init_groupzAdagrad._init_group~s'3$xAvv!>g8' 2!dK>CD;166#3#33u//22  ''* QVV$ 1 !!%,/""5=1!" ++rBcbd}|$tj5|}ddd|jD]k}g}g}g}g}|j|||||\}} t |||||d|d|d|d||d|d|d| |d t |d dt |d d m|S#1swYxYw) zPerform a single optimization step. Args: closure (Callable, optional): A closure that reevaluates the model and returns the loss. Nrrrrrrrr grad_scale found_inf) rrrrr[rrrr\rr_r`)r4 enable_gradr2r]rrT) r;closurelossr=rWrXrYrZr[r\s rAr(z Adagrad.steps  ""$y%&&E-/ "$E')J(*K+/+;+;' K, (O[  ;">2z*%L /i(z*$%56'Gn"4t<!$ T:! ': A%$s B%%B.)g{Gz?rrrg|=N)r NN)__name__ __module__ __qualname__rrJrboolr/rDrPr]rr( __classcell__)r@s@rArrs"+,#D$!DD FND D  D $) DDDDDd{D DL*-,*"*"*rBa[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 rrXrYrZrr_r`r[rrr\rrrrrr c td|Ds td||t|| d\}}|d}|d}|r)tjj r td|r)tjj r td|r%tjj st }n-|r%tjj st}nt}|||||| | | |||| | ||y) ztFunctional API that performs Adagrad algorithm computation. See :class:`~torch.optim.Adagrad` for details. c3PK|]}t|tj ywrd)r+r4r).0ts rA zadagrad.. 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\r_r`) allr0rr4jit is_scripting_fused_adagrad_multi_tensor_adagrad_single_tensor_adagrad)rrXrYrZrr_r`r[rrr\rrrrr_funcs rArrs2 @K@ @ ^   }1 Ne 7 }599))+STT '')QRR UYY++- //1$%  ! '%rBcP|j}tj|||Srd)sizer4sparse_coo_tensor)rS grad_indicesrGrzs rA _make_sparser}>s" 99;D  " "< >>rBc || tdtjjs t |}t ||||dD]\}}}}|dz }t |}| s|n| }|dk7r*|jr td|j||}|d|dz |zzz }|jr|j}|j}|j}|jt|||jd|j!|}|jj#j| }|jt||||z | &tj$|}|r?tj&|}tj&|}tj&|}|j)||d | r|j+| z}n|j+j| }|j-||| |stj.|}tj.|} y) N,Expected grad_scale and found_inf to be NoneT)strictrrz;weight_decay option is not compatible with sparse gradientsalphavalue)AssertionErrorr4rrrsrzipr rUr0addcoalesce_indices_valuesadd_r}pow sparse_masksqrt_r7 view_as_realaddcmul_sqrtaddcdiv_view_as_complex)rrXrYrZr_r`rrrrr[rrr\paramrS state_sumstep_tr(clrr| grad_valuesstd std_valuesr7s rArvrvCs!"!6KLL 99 ! ! # ^*-z;t++&tY ! &!#t$ 1 ~~"Q88E86DAX--. >>==?D==?L,,.K NN<lKOOA.s9 '3tDNN|sTrpg?cpu)r&rrr)rrHrr"_group_tensors_by_device_and_dtyperGrrFranyrvrr4 _foreach_negcompiler is_compilingis_cpu _foreach_add_r6 _foreach_addr _foreach_addcmul_ _foreach_sqrt _foreach_mul_ _foreach_mul_foreach_addcdiv_)rrXrYrZr_r`rrrrr[rrr\grouped_tensorlistsdevice_params_ device_grads_device_state_sums_device_state_steps_rw device_params device_gradsdevice_state_sumsdevice_state_stepsdevice_has_sparse_gradr( minus_clrr numerators rArurus"BCC!6KLL 6{a BB#FF  K0 & & (   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 s Jc"|sy| s| r td| r tdt|}||j|ind}||j|ind}tj||||g}|j D]\\}}\\}}}}}t tt|}t tt|}t tt|}t tt|}d\}}|!|||vr|j|d||<||}|!|||vr|j|d||<||}tj|dtj|||||||| | || |tj||gt|zy)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 = !rB)NNNFNFF)typingrr4r optimizerrrrr r r r r rrrrr__all__r__doc__rFrhrJrr}rvrurtr"rBrArs " 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  !MrB