inP&8dZddlmZddlZddlmZddlmZmZmZm Z m Z m Z m Z m Z mZmZmZmZmZmZddgZGd deZd d ed ed e d e d ed ze_deedeedeedeedeedeedededededededededededdf"d Zdeedeedeedeedeedeedededededededededededdf"d!Ze e" d%deedeedeedeedeedeed#edzdededededededededededdf$d$Zy)&z)Implementation for the RMSprop algorithm.)castN)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 _to_scalar_use_grad_for_differentiable _view_as_real OptimizerParamsTRMSproprmspropceZdZ ddedeezdededededed ed edzd ed ed dffd ZfdZdZ e ddZ xZ S)rNparamslralphaeps weight_decaymomentumcentered capturableforeachmaximizedifferentiablereturnc Zt|tr|jdk7r tdd|kstd|d|kstd|d|kstd|d|kstd|d|kstd||||||||| | | d } t ||| y) NrzTensor lr must be 1-elementgzInvalid learning rate: zInvalid epsilon value: zInvalid momentum value: zInvalid weight_decay value: zInvalid alpha value: ) rrrrrrrrr r!) isinstancernumel ValueErrorsuper__init__)selfrrrrrrrrrr r!defaults __class__s h/Volumes/fast/ai/experiments/voice-extract-mac/.venv/lib/python3.12/site-packages/torch/optim/rmsprop.pyr(zRMSprop.__init__s b& !bhhjAo:; ;by6rd;< <cz6se<= =h7zBC Cl";L>JK Ke|4UG<= =  ($ ,   *c|t|||jD]}|jdd|jdd|jdd|jdd|jdd|jdd|d D]}|jj |g}t |dk7s.tj|d rGt|d }|dr*tj|t|j ntj|t |d <y) NrrrFrr r!rrstepdtypedevicer1) r' __setstate__ param_groups setdefaultstategetlentorch is_tensorfloattensorr r2)r)r7grouppp_statestep_valr+s r,r4zRMSprop.__setstate__Hs U#&&E   Z +   Z /   Y -   Z /   -u 5   \5 18_**..B/w<1$U__WV_-M$WV_5H !. $,=,?#\\(:K:MN FO %'r-cd}|dD]} | j|tj| z}|j| | jjr t d|j| j|j | } t| dk(r|dr*tjdt| jntjdt| d <tj| tj | d <|d dkDr(tj| tj | d <|dr(tj| tj | d<|j| d |j| d |d dkDr|j| d |ds|j| d|S)NFrz)RMSprop does not support sparse gradientsrrr0r3r/) memory_format square_avgrmomentum_bufferrgrad_avg) gradr: is_complexappend is_sparse RuntimeErrorr7r9zerosr r2 zeros_likepreserve_format) r)r>params_with_gradgrads square_avgsmomentum_buffer_list grad_avgs state_steps has_complexr?r7s r, _init_groupzRMSprop._init_group]s xAvv~ 5++A. .K  # #A &vv"#NOO LL JJqME5zQ\*KK*;*=ahhOR/@/BCf ',&6&6U%:%:'l#$q(/4/?/?)>)>0E+,$(-(8(8)>)>)E*%   u\2 3   uV} -Z 1$$++E2C,DEZ   z!23I!Lr-cj|jd}|$tj5|}ddd|jD]_}g}g}g}g}g}g} |j ||||||| } t |||||| |d|d|d|d|d|d|d|d |d |d | a|S#1swYyxYw) zPerform a single optimization step. Args: closure (Callable, optional): A closure that reevaluates the model and returns the loss. Nrrrrrrrr r!r) rrrrrrrr r!rrV)'_accelerator_graph_capture_health_checkr: enable_gradr5rWr) r)closurelossr>rPrQrRrTrSrUrVs r,r/z RMSprop.steps  446  ""$y%&&E-/ "$E(*K&(I13 (*K** $K  $;Gn%L">2z*z*i(z*$%56 .'# %'L S%$s B))B2) g{Gz?gGz?g:0yE>rrFFNFFN) __name__ __module__ __qualname__rr<rboolr(r4rWrr/ __classcell__)r+s@r,rrs" #$'+'+ FN'+ '+  '+  '+'+'+'+'+'+'+ '+R*1f"4"4r-aj Implements RMSprop algorithm. .. math:: \begin{aligned} &\rule{110mm}{0.4pt} \\ &\textbf{input} : \alpha \text{ (alpha)}, \: \gamma \text{ (lr)}, \: \theta_0 \text{ (params)}, \: f(\theta) \text{ (objective)} \\ &\hspace{13mm} \lambda \text{ (weight decay)},\: \mu \text{ (momentum)}, \: centered, \: \epsilon \text{ (epsilon)} \\ &\textbf{initialize} : v_0 \leftarrow 0 \text{ (square average)}, \: \textbf{b}_0 \leftarrow 0 \text{ (buffer)}, \: g^{ave}_0 \leftarrow 0 \\[-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 \alpha v_{t-1} + (1 - \alpha) g^2_t \hspace{8mm} \\ &\hspace{5mm} \tilde{v_t} \leftarrow v_t \\ &\hspace{5mm}if \: centered \\ &\hspace{10mm} g^{ave}_t \leftarrow g^{ave}_{t-1} \alpha + (1-\alpha) g_t \\ &\hspace{10mm} \tilde{v_t} \leftarrow \tilde{v_t} - \big(g^{ave}_{t} \big)^2 \\ &\hspace{5mm}if \: \mu > 0 \\ &\hspace{10mm} \textbf{b}_t\leftarrow \mu \textbf{b}_{t-1} + g_t/ \big(\sqrt{\tilde{v_t}} + \epsilon \big) \\ &\hspace{10mm} \theta_t \leftarrow \theta_{t-1} - \gamma \textbf{b}_t \\ &\hspace{5mm} else \\ &\hspace{10mm}\theta_t \leftarrow \theta_{t-1} - \gamma g_t/ \big(\sqrt{\tilde{v_t}} + \epsilon \big) \hspace{3mm} \\ &\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 `lecture notes `_ by G. Hinton. and centered version `Generating Sequences With Recurrent Neural Networks `_. The implementation here takes the square root of the gradient average before adding epsilon (note that TensorFlow interchanges these two operations). The effective learning rate is thus :math:`\gamma/(\sqrt{v} + \epsilon)` where :math:`\gamma` is the scheduled learning rate and :math:`v` is the weighted moving average of the squared gradient. z Args: a0 lr (float, Tensor, optional): learning rate (default: 1e-2) alpha (float, optional): smoothing constant (default: 0.99) 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) momentum (float, optional): momentum factor (default: 0) centered (bool, optional) : if ``True``, compute the centered RMSProp, the gradient is normalized by an estimation of its variance z z rrQrRrTrSrUrrrrrrr r!rrVr"c tjjs t|}t |D]+\}}||}tj j s`|r^t}|jj|jjk(r|jj|vstd|d||}| s|n| }||}|dz }| dk7r|j|| }tj|}|r?tj|}tj|}tj|}|j|j||d|z | rT||}|rtj|}|j!|d|z |j#||dj%}n|j'}| r|j|}n|j)|}| dkDrS||}|rtj|}|j| j+|||j)|| |j+||| .y)NIIf capturable=True, params and state_steps must be on supported devices: .rrrvalue)r:jit is_scriptingr enumeratecompiler is_compilingr r2typeAssertionErroraddrI view_as_realmul_addcmul_lerp_addcmulsqrt_sqrtadd_addcdiv_)rrQrRrTrSrUrrrrrrr r!rrViparamr/capturable_supported_devicesrHrEis_complex_paramrGavgbufs r,_single_tensor_rmspropr s'& 99 ! ! # ^f%51~~~**,+L+N ( !!T[[%5%55LL%%)EE$_`|_}}~Qx#t$ ^    1 88E86D ++E2 &&u-E%%d+D++J7J''d!e)'D  |H --h7 NN4U +$$Xxr$BHHJC//#C ''#,C((3-C a<&q)C((- HHX  ' 'c 2 JJs2#J & NN4RCN 0i&r-c \!t|dk(ry| r tdtjj s;|r9t !t !fdt||dDstd!dt|}tj||||||g}|jD]|\\}}}}}}}ttt|}ttt|}ttt|}ttt|}|rg||g}| dkDr(ttt|}|j|| r(ttt|}|j|t!|g|| rtj"|}tjj s=|dj$r.tj&|tj(dd d ntj&|d | dk7r3| rtj&||| ntj*||| }tj,||tj.|||d |z | rvttt|}tj0||d |z tj2|||d }tj4|tj&||n+tj6|}tj&||| dkDrttt|}tj,|| tj8||||rIt;|tjr/tj<|| } tj&|| tj&||| |rJt;|tjr0tj>|| tj8|||ctj8|||| y)Nrz#_foreach ops don't support autogradc3K|]N\}}|jj|jjk(xr|jjvPywr])r2ro).0r?r/r}s r, z(_multi_tensor_rmsprop..rsR A4 HHMMT[[-- - > !== >@sAAT)strictrdreg?cpu)r2rfrrgri) r9rpr:rmrnr allziprr"_group_tensors_by_device_and_dtypevaluesrlistrrJr _foreach_negis_cpu _foreach_add_r= _foreach_add _foreach_mul__foreach_addcmul__foreach_lerp__foreach_addcmul_foreach_sqrt_ _foreach_sqrt_foreach_addcdiv_r$ _foreach_mul _foreach_div_)"rrQrRrTrSrUrrrrrrr r!rrVgrouped_tensorsgrouped_params_grouped_grads_grouped_square_avgs_grouped_grad_avgs_grouped_momentum_buffer_list_grouped_state_steps__grouped_params grouped_gradsgrouped_square_avgsgrouped_state_stepsstate_and_gradsgrouped_momentum_buffer_listgrouped_grad_avgsr momentum_lrr}s" @r,_multi_tensor_rmsproprVs& 6{aBCC >> & & (Z'H'J$ v{4@  ![\x[yyz{  BBBB  Y0DkRO  " " $     ) d6lO<T&\>: "4<1EF"4<1EF ,.ABO!|/3L"?0, &&'CD$(f7I$J!&&'89 . ;? ; !..}=M ~~**,1DQ1G1N1N   #U\\#e%DC     3Q 7 1 ##M>V % 2 2!>!  /7   QY   $T&\3E F   !2M1u9 M((#%68IQSC   %   S )%%&9:C   S ) a<+/V ;, (    .s5-8 1ell#[s$&zPAPI has changed, `state_steps` argument must contain a list of singleton tensorsNF) use_fusedz6torch.jit.script not supported with foreach optimizers) rrrrrrr rr!rV) r:rmrnrrLrrjrkrr)rrQrRrTrSrUrr r!rrVrrrrrrrfuncs r,rrs: >> & & (5-852 ^  1 Ne 7599))+STTuyy--/$%   !%!r-)NFFFF)__doc__typingrr:r optimizerrrrr r r r r rrrrrr__all__rrr<rarrrrCr-r,rs^0 $ i gigV+X       Y<BJ1 LJ1 <J1fJ1F| J1 v, J1 f J1 J1 J1 J1J1J1J1J1J1 !J1"#J1$ %J1ZCW LCW <CWfCWF| CW v, CW f CW CW CW CWCWCWCWCWCW !CW"#CW$ %CWL 1GH  A LA <AfAF| A v, A f AD[AAAAA A !A" #A$%A&'A()A* +AIAr-