Phq)ddlZddlmZddlmZmZmZmZmZmZm Z m Z m Z m Z m Z mZmZddlmZmZmZmZddlmZddgZGd deZd d e d e d ed e d e d ze_ d'deedeedeedeedeedeedeedededeedeedeedededededeeefded ed!ef(d"Zdeedeedeedeedeedeedeedeededededeeefded ed!edededef$d#Zdeedeedeedeedeedeedeedeededededeeefded ed!edededef$d$Zdeedeedeedeedeedeedeedeededededeeefded ed!edededed%df&d&Zy)(N)Tensor) Optimizer_use_grad_for_differentiable _get_value_dispatch_sqrt_stack_if_compiling_capturable_doc_differentiable_doc _foreach_doc _fused_doc _maximize_doc_default_to_fused_or_foreachParamsT _view_as_real)ListOptionalTupleUnion)$_get_fused_kernels_supported_devicesAdamWadamwceZdZ ddddddddedeeefdeeefdeded ed ed e ed ed ede effdZ fdZ dZ e ddZxZS)rFN)maximizeforeach capturabledifferentiablefusedparamslrbetaseps weight_decayamsgradrrrrrc < d|kstd|t|tr|r | s tdd|kstd|d|dcxkrdksntd|dd|dcxkrdksntd |dd|kstd |t|||||||| | | } t||| | rY| r t d d |_t t fd|jDst d d|r t dyy)NzInvalid learning rate: Elr as a Tensor is not supported for capturable=False and foreach=TruezInvalid epsilon value: r?z#Invalid beta parameter at index 0: rz#Invalid beta parameter at index 1: zInvalid weight_decay value: ) r r!r"r#r$rrrrrz)`fused` does not support `differentiable`Tc3K|]=}|dD]3}|jjvxrtj|5?yw)rN)devicetypetorchis_floating_point).0pgpfused_supported_devicess \C:\Users\daisl\Desktop\realtime-object-detection\venv\Lib\site-packages\torch/optim/adamw.py z!AdamW.__init__..?sQ,BR\ !88+''*+5A++sAAzX`fused=True` requires all the params to be floating point Tensors of supported devices: .z0`fused` and `foreach` cannot be `True` together.) ValueError isinstancerdictsuper__init__ RuntimeError_step_supports_amp_scalingrall param_groups)selfrr r!r"r#r$rrrrrdefaultsr1 __class__s @r2r9zAdamW.__init__ sxby6rd;< < b& !gjde ecz6se<= =eAh$$B58*MN NeAh$$B58*MN Nl";L>JK K%!)   * "#NOO.2D + 'K&L #++ #$99P8QQR$TUU"#UVV! c6t|||jD]n}|jdd|jdd|jdd|jdd|jdd|jddpt |j j }t|dk7xrtj|dd }|s<|D]6}tjt|d tj |d <8yy) Nr$Frrrrrrstepdtype) r8 __setstate__r= setdefaultliststatevalueslenr, is_tensortensorfloatfloat32)r>rIgroup state_valuesstep_is_tensorsr@s r2rFzAdamW.__setstate__Is U#&&E   Y .   Z /   Y -   \5 1   -u 5   Wd + 'DJJ--/0 l+q0 eoo OF #7 !!LLqy)9O& "rAc 0d} |dD] } | j| tj| z} |j| | jjr t d|j| j|j | } t| dk(r|ds|dr0tjdtj| jn$tjd tj | d <tj| tj | d <tj| tj | d<|r(tj| tj | d<|j| d |j| d|dr|j| d|dr| d jr t d|dr#t|dt r|ds t d|j| d  | S)NFrz'AdamW does not support sparse gradientsrrr)rEr*r&rDrC) memory_formatexp_avg exp_avg_sqmax_exp_avg_sqr$rzB`requires_grad` is not supported for `step` in differentiable moderr r')gradr, is_complexappend is_sparser:rIrKzerosrOr*rM zeros_likepreserve_format requires_gradr6r) r>rPparams_with_gradgradsr$exp_avgs exp_avg_sqsmax_exp_avg_sqs state_steps has_complexr0rIs r2 _init_groupzAdamW._init_groupZs xAvv~ 5++A. .K  # #A &vv"#LMM LL JJqME5zQ \*eGnKK%--Ic?f $)#3#3U%:%:$i ',&6&6U%:%:'l#.3.>.>)>)>/E*+ OOE), -   u\2 3Y&&u-='>?%&5=+F+F"#ghhYJuT{F$CER^L_"#jkk   uV} -]!^rAc|jd}|$tj5|}ddd|jD]}g}g}g}g}g}g} |d} |d\} } |j |||| |||| } t |||||| f| | | |d|d|d|d|d|d |d |d t |d dt |d d| d|S#1swYxYw)zPerform a single optimization step. Args: closure (Callable, optional): A closure that reevaluates the model and returns the loss. Nr$r!r r#r"rrrrr grad_scale found_inf)r$beta1beta2r r#r"rrrrrrkrlrh) _cuda_graph_capture_health_checkr, enable_gradr=rirgetattr)r>closurelossrPrbrcrdrerfrgr$rmrnrhs r2rCz AdamW.steps: --/  ""$y%&&E! EHK OKI&G >LE5**  K     ;">2%Lz*i( .$%56Gn"4t<!$ T:') +'X _%$s C  C)gMbP?)g?g+?g:0yE>g{Gz?FN)__name__ __module__ __qualname__rrrNrrboolrr9rFrirrC __classcell__)r@s@r2rr s$(%1":W"& $ $:W:W %- :WUE\" :W  :W  :W:W:W$:W:W:W~:WxP";z":":rAa Implements AdamW 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)}, \: \epsilon \text{ (epsilon)} \\ &\hspace{13mm} \lambda \text{(weight decay)}, \: \textit{amsgrad}, \: \textit{maximize} \\ &\textbf{initialize} : m_0 \leftarrow 0 \text{ (first moment)}, v_0 \leftarrow 0 \text{ ( second moment)}, \: \widehat{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} \theta_t \leftarrow \theta_{t-1} - \gamma \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}\widehat{v_t} \leftarrow v_t/\big(1-\beta_2^t \big) \\ &\hspace{5mm}\textbf{if} \: amsgrad \\ &\hspace{10mm}\widehat{v_t}^{max} \leftarrow \mathrm{max}(\widehat{v_t}^{max}, \widehat{v_t}) \\ &\hspace{10mm}\theta_t \leftarrow \theta_t - \gamma \widehat{m_t}/ \big(\sqrt{\widehat{v_t}^{max}} + \epsilon \big) \\ &\hspace{5mm}\textbf{else} \\ &\hspace{10mm}\theta_t \leftarrow \theta_t - \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 `Decoupled Weight Decay Regularization`_. a Args: params (iterable): iterable of parameters to optimize or dicts defining parameter groups 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 coefficient (default: 1e-2) 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 z .. _Decoupled Weight Decay Regularization: https://arxiv.org/abs/1711.05101 .. _On the Convergence of Adam and Beyond: https://openreview.net/forum?id=ryQu7f-RZ rrcrdrerfrgrrrrrkrlrhr$rmrnr r#r"rc htjjstd|Ds t d| )|'t ||d\}}|rt |tr|sd}| d} |d}|r)tjjr t d| r)tjjr t d| r%tjjst}n-|r%tjjst}nt}|||||||| ||||||||| | | y) zpFunctional API that performs AdamW algorithm computation. See :class:`~torch.optim.AdamW` for details. c3PK|]}t|tj ywrt)r6r,r)r.ts r2r3zadamw..4s2dXcST:a3NXcs$&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) r$rmrnr r#r"rrrrkrlrh) r,_utils is_compilingr<r:rr6rjit is_scripting _fused_adamw_multi_tensor_adamw_single_tensor_adamw)rrcrdrerfrgrrrrrkrlrhr$rmrnr r#r"r_funcs r2rrs$: << $ $ &s2dXc2d/d ^   }1&.TYZ 7 z"f-jG }599))+STT '')QRR UYY++- //1"#  ! %%rAc ||Jtjjrt| tsJt |D]\}}|s||n|| }||}||}||}tj js9|r7|jr |js|jr |jsJdtj|rqtj|}tj|}tj|}|rtj||||<tj|}|dz }|jd| | zz |j|d| z |j| j||d| z |s|r|}d| |zz }d| |zz }| |z }|j}|j!}|ro|r||j#}n||}||j%tj&||||j!||zz j)| |z }n(|j!||zz j)| |z }|j+||nt-|}d| |zz }d| |zz }| |z }t/|}|rDtj&|||||||j!|z j)| }n"|j!|z j)| }|j+||| |stj||stj0||||<!y)NzGIf capturable=True, params and state_steps must be CUDA or XLA tensors.r)value)out)r,rrr6rN enumerater~ris_cudais_xlar[ view_as_realmul_lerp_addcmul_negsqrtclonecopy_maximumadd_addcdiv_rrview_as_complex) rrcrdrerfrgrkrlr$rmrnr r#r"rrrrhiparamrZrWrXstep_trCbias_correction1bias_correction2 step_size step_size_negbias_correction2_sqrtrYdenoms r2rris^,  )"33 3 yy"e$$$f%5'uQxeAhY1+ ^ Q||((*z6>>u||  YX YV   E "%%d+D((1G++J7J%*%7%78J%K"&&u-E !   1rL(()  dAI&''d!e)'D D 5D=0  5D=0 --I%MMOM$4$9$9$; !!%4Q%7%=%=%?N%4Q%7N"((~z)RS $A&++-1F1VW$s]*+ OO%)>)NO$s]*+ NN7E *f%D 5D=0  5D=0 --I$23C$D ! oa0*/RSBTU)+0025JJPPQTU#*-BBHHM NN7E)N < u''q 2!&!6!6q7I!JOA k&rAc *t|dk(ryt| tr |s tdtj j s%|r#tdt||DsJd|rJd||Jtj||||||g}|jD]R\\}}}}}}}|rt j|}|r |rt|||||nt|||||djr.t j|t j ddd nt j|d | dk7rt j"|d | | zz t j$||d | z t j"|| t j&|||d | z ~|rOt j(| |}t j(| |}t j*|d t j*|d t j,|t j.|| t j0|t j2||}|}|r,t j4||t j6|}nt j6|}t j.||t j|| t j.||t j8|||Y|Dcgc]}d | t;|zz }}|Dcgc]}d | t;|zz }}t=|D cgc] } | | z d z c} }|D cgc] } t?| }} |r,t j4||t j6|}nt j6|}t j.||t j|| t j8||||Uycc}wcc}wcc} wcc} w) Nrr'c3VK|]!\}}|jxr |j#ywrt)r)r.r0rCs r2r3z&_multi_tensor_adamw..s) 6N71dAII &$,, &6Ns')z@If capturable=True, params and state_steps must be CUDA tensors.z#_foreach ops don't support autogradr(cpu)r*)alphar) rKr6rr:r,r~rr<zipr"_group_tensors_by_device_and_dtyperJ _foreach_negris_cpu _foreach_add_rM _foreach_mul__foreach_lerp__foreach_addcmul_ _foreach_pow _foreach_sub_ _foreach_neg_ _foreach_div__foreach_reciprocal__foreach_sqrt__foreach_maximum_ _foreach_sqrt_foreach_addcdiv_rr r)!rrcrdrerfrgrkrlr$rmrnr r#r"rrrrhgrouped_tensors device_params device_gradsdevice_exp_avgsdevice_exp_avg_sqsdevice_max_exp_avg_sqsdevice_state_stepsrrrrrexp_avg_sq_sqrtrCbcs! r2rrs* 6{a"fjbcc << $ $ &: 69&+6N   N M N DDD   )"33 3BBxo{DLMO # # %    --l;L m\?L^`vwm\?L^_ a ' '    2ELLU4S[^ _    2A 6 1     q2 3D/D E _lAIF .6  2L,PQTYPYZ  $11%9KL $11%9KL     0! 4    0! 4    0 1    0" 5  & &'7 8  !1 2 )I$4 !''(>@RS#("5"56L"M"'"5"56H"I   1F G    5    ;  # #M?O TJ\]J\$EZ-=$= =J\ ]J\]J\$EZ-=$= =J\ ]+FV,WFVb2g^FV,WXIBR$SBRB^B%7BR !$S''(>@RS#("5"56L"M"'"5"56H"I   1F G    5  # #M?OU^ _o&J ^],W$SsP%PP "Preturnc |sy|r td||j|ind}||j|ind}t| tr&t | jdk7r| j| ind}t j ||||||g}|jD]\\}}\\}}}}}}}d\}}|||vr|j|d||<||}|||vr|j|d||<||}|||vr| j|d||<||} tj|dtj|||||||| | | | | ||||tj||gt|zy) Nz9Adam with fused=True does not support differentiable=Truer)NNT) non_blocking)r*rr) r$r rmrnr#r"rrkrl)r:r*r6rstrrritemstor,r _fused_adamw_rrK) rrcrdrerfrgrkrlr$rmrnr r#r"rrrrhgrad_scale_dictfound_inf_dictlr_dictrr*rrrrrrrdevice_grad_scaledevice_found_infs r2rrgs* VWW9C9Oz((*5UYO6?6Ki&& 2QUN",B!7C Ners iiii0/K G FIFR&L       ' M? V#  #'"&O LO <O6lOf O &\ O f Od^OOO D>O OOO"#O$ %O& 'O( eVm)O*+O, -O./OdsK LsK <sK6lsKf sK &\ sK f sK sKsKsK sK sK femsKsK sK !sK"#sK$%sK&'sKlE` LE` <E`6lE`f E` &\ E` f E` E`E`E` E` E` femE`E` E` !E`"#E`$%E`&'E`PHb LHb <Hb6lHbf Hb &\ Hb f Hb HbHbHb Hb Hb eVmHbHb Hb !Hb"#Hb$%Hb&'Hb( )HbrA