iA ddlmZmZddlZddlmZddlmZmZmZm Z m Z m Z m Z m Z mZmZmZmZmZmZddgZGddeZd d ed e d ed e d ed ze_deedeedeedeedeedededededededededdfdZdeedeedeedeedeedededededededededdfdZe e d!deedeedeedeedeedededzdedededededededdfd Zy)")AnycastN)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 OptimizerParamsTAdadeltaadadeltaceZdZ ddddddedeezdededed edzd ed ed ed dffdZfdZde e e fde ede ede ede ede ef dZ eddZxZS)rNF) capturablemaximizedifferentiableparamslrrhoeps weight_decayforeachrrrreturnc @t|tr|jdk7r tdd|kstd|d|cxkrdksntd|d|kstd|d|kstd||||||||| d } t ||| y) NrzTensor lr must be 1-elementgzInvalid learning rate: ?zInvalid rho value: zInvalid epsilon value: zInvalid weight_decay value: )rrrrrrr r) isinstancernumel ValueErrorsuper__init__) selfrrrrrr rrrdefaults __class__s i/Volumes/fast/ai/experiments/voice-extract-mac/.venv/lib/python3.12/site-packages/torch/optim/adadelta.pyr(zAdadelta.__init__s b& !bhhjAo:; ;by6rd;< <c S 23%89 9cz6se<= =l";L>JK K( $,   *c0t|||jD]}|jdd|jdd|jdd|jdd|dD]}|jj |g}t |dk7s.tj|drGt|d}|dr*tj|t|j ntj|t |d<y) Nr rFrrrrstepdtypedevicer1) r' __setstate__ param_groups setdefaultstategetlentorch is_tensorfloattensorr r2)r)r7grouppp_statestep_valr+s r,r4zAdadelta.__setstate__As U#&&E   Y -   Z /   -u 5   \5 18_**..B/w<1$U__WV_-M$WV_5H !. $,=,?#\\(:K:MN FO % 'r-r>params_with_gradgrads square_avgs acc_deltas state_stepscd}|dD]o}|j|tj|z}|j||jjr t d|j|j|j |} t| dk(r|dr*tjdt|jntjdt| d <tj|tj | d <tj|tj | d <|j| d |j| d |j| d r|S) NFrz*Adadelta does not support sparse gradientsrrr0r3r/) memory_format square_avg acc_delta) gradr: is_complexappend is_sparse RuntimeErrorr7r9zerosr r2 zeros_likepreserve_format) r)r>rBrCrDrErF has_complexr?r7s r, _init_groupzAdadelta._init_groupTsN xAvv~ 5++A. .K  # #A &vv"#OPP LL JJqME5zQ\*KK*;*=ahhOR/@/BCf ',&6&6U%:%:'l#&+%5%5U%:%:&k"   u\2 3   eK0 1   uV} -9!<r-cx|jd}|$tj5|}ddd|jD]f}g}g}g}g}g}|d|d|d|d|d|d|d|d f\} } } } } }}}|j ||||||}t |||||| | | | | |||| h|S#1swYxYw) zPerform a single optimization step. Args: closure (Callable, optional): A closure that reevaluates the model and returns the loss. Nrrrrr rrr) rrrrr rrrrT)'_accelerator_graph_capture_health_checkr: enable_gradr5rUr)r)closurelossr>rBrCrDrErFrrrrr rrrrTs r,r/z Adadelta.steps' 446  ""$y%&&E-/ "$E(*K')J(*Kd e e n%i j!&'l#  **' ZK  )!-%' ='^ e%$s B00B9)r#g?gư>rNN)__name__ __module__ __qualname__rr<rboolr(r4dictstrrlistrUrr/ __classcell__)r+s@r,rrs !#"+!$"+"+ FN"+ "+  "+  "+"+"+"+"+ "+H&)CH~)v,)F| ) &\ ) L )&\)V"="=r-aImplements Adadelta algorithm. .. math:: \begin{aligned} &\rule{110mm}{0.4pt} \\ &\textbf{input} : \gamma \text{ (lr)}, \: \theta_0 \text{ (params)}, \: f(\theta) \text{ (objective)}, \: \rho \text{ (decay)}, \: \lambda \text{ (weight decay)} \\ &\textbf{initialize} : v_0 \leftarrow 0 \: \text{ (square avg)}, \: u_0 \leftarrow 0 \: \text{ (accumulate variables)} \\[-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 v_{t-1} \rho + g^2_t (1 - \rho) \\ &\hspace{5mm}\Delta x_t \leftarrow \frac{\sqrt{u_{t-1} + \epsilon }}{ \sqrt{v_t + \epsilon} }g_t \hspace{21mm} \\ &\hspace{5mm} u_t \leftarrow u_{t-1} \rho + \Delta x^2_t (1 - \rho) \\ &\hspace{5mm}\theta_t \leftarrow \theta_{t-1} - \gamma \Delta x_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 `ADADELTA: An Adaptive Learning Rate Method`_. z Args: ar lr (float, Tensor, optional): coefficient that scale delta before it is applied to the parameters (default: 1.0) rho (float, optional): coefficient used for computing a running average of squared gradients (default: 0.9). A higher value of `rho` will result in a slower average, which can be helpful for preventing oscillations in the learning process. eps (float, optional): term added to the denominator to improve numerical stability (default: 1e-6). weight_decay (float, optional): weight decay (L2 penalty) (default: 0) z zd .. _ADADELTA\: An Adaptive Learning Rate Method: https://arxiv.org/abs/1212.5701 rrCrDrErFrrrrrrrrTr!c.tjjs=| r;tdt fdt ||dDst ddtjjs t|}t |||||dD]{\} }}}}|dz }| s|n| }|d k7r|j| | }tj| r?tj|}tj|}tj|}|j|j||d|z |j|j}|j|j}| r|j!}|j#|j||j|j||d|z tj| rtj$|}| j'|| ~y) NF supports_xlac3K|]N\}}|jj|jjk(xr|jjvPywr[r2type.0r?r/capturable_supported_devicess r, z*_single_tensor_adadelta.. R A4 HHMMT[[-- - > !== >@AATstrictIIf capturable=True, params and state_steps must be on supported devices: .rralphavalue)r:compiler is_compilingr allzipAssertionErrorjit is_scriptingraddrM view_as_realmul_addcmul_sqrt_clonediv_view_as_complexadd_)rrCrDrErFrrrrrrrrTparamrLrJrKr/stddeltarls @r,_single_tensor_adadeltars" >> & & (Z'H( $ v{4@  ![\x[yyz{  99 ! ! # ^47{J D550tZD  #t$ 1 88E86D   E "++J7J**95I%%d+D%%dDC%@nnS!'') c"((* KKME 3T"s$$UES$A   E "))%0E 5 $15r-c| r tdtjjs=| r;t dt fdt ||dDstddt|d k(ryt|}tj|||||g} | jD]\\}}}}}}ttt|}ttt|}ttt|}ttt|}ttt|}| rt||||tjjs=|d j r.tj"|tj$d d d ntj"|d| rtj&|}|d k7r3| rtj"||| ntj(||| }tj*||tj,|||d|z tj(||}tj.|tj(||}tj.|tj0||tj*||tj*||tj,|||d|z | rIt3|tjr/tj*|| tj"||tj"||| y)Nz#_foreach ops don't support autogradFrec3K|]N\}}|jj|jjk(xr|jjvPywr[rhrjs r,rmz)_multi_tensor_adadelta..IrnroTrprrrsrr#cpu)r2rtrrv)r|r:rxryr rzr{r9rr"_group_tensors_by_device_and_dtypevaluesrrbrris_cpu _foreach_add_r= _foreach_neg _foreach_add _foreach_mul__foreach_addcmul__foreach_sqrt_ _foreach_div_r$)rrCrDrErFrrrrrrrrTgrouped_tensorsdevice_params_ device_grads_device_square_avgs_device_acc_deltas_device_state_steps__ device_params device_gradsdevice_square_avgsdevice_acc_deltasdevice_state_stepsrdeltasrls @r,_multi_tensor_adadeltar1s BCC >> & & (Z'H( $ v{4@  ![\x[yyz{  6{a BBBB  Z=O  " " $   T&\>: DL-8 !$v,0CD f/AB!$v,0CD  |-?AR ~~**,1CA1F1M1M   "ELLU$C3     2A 6  --l;L 1 ##L-|T$11 -|   .4   l!c'   !3S9 S!##$5s; V$ FC( FL1 -s3  166SQ *R6    ,    v 6    vbS Aq %r-)single_tensor_fnr c ztjjstd|Ds t d|t ||d\}}|r)tj jr t d|r%tj jst}nt}||||||| | | | | ||| y)zvFunctional API that performs Adadelta algorithm computation. See :class:`~torch.optim.Adadelta` for details. c3PK|]}t|tj ywr[)r$r:r)rkts r,rmzadadelta..s5-8 1ell#[s$&zPAPI has changed, `state_steps` argument must contain a list of singleton tensorsNF) use_fusedz6torch.jit.script not supported with foreach optimizers)rrrrrrrrT) r:rxryrzrPrr}r~rr)rrCrDrErFrr rrTrrrrrrfuncs r,rrs6 >> & & (5-852 ^  1 Ne 7599))+STTuyy--/%&   !%r-)FNFF)typingrrr:r optimizerrrr r r r r rrrrrrr__all__r__doc__rbr<r_rrrrHr-r,rs $ z "ayaJ8       90 j9% L9% <9%f9%V 9% f 9% 9% 9% 9%9%9%9%9%9% 9%xgB LgB <gBfgBV gB f gB gB gB gBgBgBgBgBgB gBT 1HI = L= <=f=V = f ==D[=== = = = !="#=$ %=J=r-