Q/j7}*:ddlmZmZmZmZmZddlZddlmZddlm Z m Z m Z m Z m Z mZmZmZmZmZmZmZmZmZmZmZmZddgZGddeZd d ed ed e d e d ed ze_d eedeedeedeedeedeedeedeedededededeeefdededededef$dZd eedeedeedeedeedeedeedeedededededeeefdededededef$d Z d eedeedeedeedeedeedeedeedededededeeefdededededed!df&d"Z!e e# d'd eedeedeedeedeedeed$eededed%eedeedeedededededeeefdededef(d&Z"y)()castListOptionalTupleUnionN)Tensor)_capturable_doc_default_to_fused_or_foreach_device_dtype_check_for_fused_differentiable_doc_disable_dynamo_if_unsupported _foreach_doc _fused_doc!_get_capturable_supported_devices_get_scalar_dtype _get_value _maximize_doc_stack_if_compiling_use_grad_for_differentiable _view_as_real DeviceDict OptimizerParamsTAdamadamceZdZ ddddddddedeeefdeeefdeded ed e ed ed ed ede effdZ fdZ dZ e ddZxZS)rFN)foreachmaximize capturabledifferentiablefusedparamslrbetaseps weight_decayamsgradrrr r!r"c t|tr-|r | s td|jdk7r tdd|kstd|d|kstd|d|dcxkrdksntd |dd|dcxkrdksntd |dd|kstd |t |||||||| | | } t ||| | r"| r td d|_|r tdyy)NElr as a Tensor is not supported for capturable=False and foreach=Truer zTensor lr must be 1-elementzInvalid learning rate: zInvalid epsilon value: r?z#Invalid beta parameter at index 0: z#Invalid beta parameter at index 1: zInvalid weight_decay value: ) r$r%r&r'r(rrr r!r"z)`fused` does not support `differentiable`Tz0`fused` and `foreach` cannot be `True` together.) isinstancer ValueErrornumeldictsuper__init__ RuntimeError_step_supports_amp_scaling)selfr#r$r%r&r'r(rrr r!r"defaults __class__s S/root/repos/pocs/video-evals/.venv/lib/python3.12/site-packages/torch/optim/adam.pyr2z Adam.__init__!sL b& !z [xxzQ !>??by6rd;< <cz6se<= =eAh$$B58*MN NeAh$$B58*MN Nl";L>JK K%!)   * "#NOO.2D + "#UVV ct|||jD]#}|jdd|jdd|jdd|jdd|jdd|jdd}|dD]}|jj |g}t |d k7s.tj|d rGt|d }|ds|dr,tj|t| |j ntj|t |d <&y)Nr(Frrr r!r"r#rstepis_fuseddtypedevicer?) r1 __setstate__ param_groups setdefaultstategetlentorch is_tensorfloattensorrr@)r5rEgroupr"pp_statestep_valr7s r8rBzAdam.__setstate__[s% U#&& E   Y .   Z /   Y -   \5 1   -u 5$$Wd3E8_ **..B/w<1$U__WV_-M$WV_5H!.%.  $"3U"C#$88 #\\(:K:MNFO  r9cRd}|dD]} | j|tj| z}|j| | jjr t d|j| j|j | } t| dk(r|dr t| |ds|dr/tjdt|d| j ntjd t | d <tj| tj | d<tj| tj | d<|dr(tj| tj | d<|j| d|j| d|dr|j| d|dr| d jr t d|dr(tj |dr|ds t d|j| d |S)NFr#zJAdam does not support sparse gradients, please consider SparseAdam insteadrr"r r<r>r+rAr;) memory_formatexp_avg exp_avg_sqr(max_exp_avg_sqr!zB`requires_grad` is not supported for `step` in differentiable moderr$r*)gradrH is_complexappend is_sparser3rErGr zerosrr@rK zeros_likepreserve_format requires_gradrI) r5rLparams_with_gradgradsexp_avgs exp_avg_sqsmax_exp_avg_sqs state_steps has_complexrMrEs r8 _init_groupzAdam._init_grouprs x= 2Avv!u//22  ''*66##&d QVV$ 1 u:?W~5a8!.%.  "3U7^"L#$88 #\\#5F5HI&M(-'7'7)>)>(E)$+0*:*:)>)>+E,'Y'272B2BU-B-B3./i 01""5#67##**51A+BC)*uV}/J/J&\ )$d 4!,/&_""5=1{= 2|r9c|jd}|$tj5|}ddd|jD]|}g}g}g}g}g}g} |d\} } |j ||||||| } t |||||| f|d| | | |d|d|d|d|d|d |d |d t |d dt |d dd~|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&rrr r!r" grad_scale found_inf)r(rdbeta1beta2r$r'r&rrr r!r"rgrh) _cuda_graph_capture_health_checkrH enable_gradrCrergetattr) r5closurelossrLr^r_r`rarbrcrirjrds r8r;z Adam.steps> --/  ""$ !y !&&( E-/ "$E%'H(*K,.O(*K >LE5** K    i(';">2%Lz*i( .$%56Gn"4t<!$ T:) '( T [ ! !s CC)gMbP?)g?g+?g:0yE>rFN)__name__ __module__ __qualname__rrrJrrboolrr2rBrerr; __classcell__)r7s@r8rr s$(%18W#' $ $8W8W %- 8WUE\" 8W  8W  8W8W$8W8W8W8W~8Wt.IV"8"8r9a Implements Adam 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)} \\ &\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}\textbf{if} \: \lambda \neq 0 \\ &\hspace{10mm} g_t \leftarrow g_t + \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-1} - \gamma \widehat{m_t}/ \big(\sqrt{\widehat{v_t}^{max}} + \epsilon \big) \\ &\hspace{5mm}\textbf{else} \\ &\hspace{10mm}\theta_t \leftarrow \theta_{t-1} - \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 `Adam: A Method for Stochastic Optimization`_. 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 (L2 penalty) (default: 0) 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 a= .. Note:: A prototype implementation of Adam and AdamW for MPS supports `torch.float32` and `torch.float16`. .. _Adam\: A Method for Stochastic Optimization: https://arxiv.org/abs/1412.6980 .. _On the Convergence of Adam and Beyond: https://openreview.net/forum?id=ryQu7f-RZ r#r_r`rarbrcrgrhr(rdrirjr$r'r&rr r!c "||Jtjjrt| tsJt |D]J\}}|s||n|| }||}||}||}tj js\|rZt}|jj|jjk(r|jj|vs Jd|d|dz }| dk7r|j|| }tj|rqtj|}tj|}tj|}|rtj||||<tj|}|j|d| z |j| 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 } |j1|| nt3|}d| |zz }d| |zz }| |z }|dz}|rDtj,|||||||j'|z j/|} n"|j'|z j/|} |j1|| | |stj||s0tj4||||<My) NIIf capturable=True, params and state_steps must be on supported devices: .r ralpha)value?)out)rHjit is_scriptingr-rJ enumerate_utils is_compilingrr@typeaddrW view_as_reallerp_mul_addcmul_conjnegsqrtclonecopy_maximumadd_addcdiv_rview_as_complex)!r#r_r`rarbrcrgrhr(rdrirjr$r'r&rr r!iparamrVrSrTstep_tcapturable_supported_devicesr;bias_correction1bias_correction2 step_size step_size_negbias_correction2_sqrtrUdenoms! r8_single_tensor_adamr@s*  )"33 3 yy"e$$$f%WK5'uQxeAhY1+ ^ Q||((*z+L+N ( !!V]]%7%77LL%%)EE {[[wZxxyz {F !  1 88E86D   E "%%d+D((1G++J7J%*%7%78J%K"&&u-E  dAI&''diikU'K D 5$;  5$; --I%MMOM$4$9$9$; !!%4Q%7%=%=%?N%4Q%7N"((~z)RS $A&++-1F1VW$s]*+ OO%)>)NO$s]*+ NN7E *f%D 5$;  5$; --I$4c$9 ! oa0*/RSBTU)+0025JJPPQTU#*-BBHHM NN7E)N < u''q 2!&!6!6q7I!JOA oWKr9c  't|dk(ryt| tr |s tdtj j s7|r5td't'fdt||Ds Jd'd||J|rJdtj||||||g}|jD]9\\}}}}}}}ttt|}ttt|}ttt|}ttt|}ttt|}| r7|r'ttt|}t|||||nt|||||rt j |}tj j s=|dj"r.t j$|t j&d d d nt j$|d | dk7r3|rt j$||| nt j(||| }t j*||d | z t j,|| t j.|||d | z ~|rft j0| |} t j0| |}!t j2| d t j2|!d t j4|!t j6| | t j8| t j:|!| }"|!}#|rCttt|}t j<||t j>|}$nt j>|}$t j6|$|#t j$|$|t j6|$|"t j@|||$/|D%cgc]}%d | tC|%zz } }%|D%cgc]}%d | tC|%zz }!}%tE| D&cgc] }&| |&z dz c}&}"|!D&cgc]}&|&dz }#}&|rCttt|}t j<||t j>|}$nt j>|}$t j6|$|#t j$|$|t j@|||$|"<ycc}%wcc}%wcc}&wcc}&w)Nrr*F) supports_xlac3K|]N\}}|jj|jjk(xr|jjvPywrp)r@r).0rMr;rs r8 z%_multi_tensor_adam..sQ 4 HHMMT[[-- - > !== > sAArwrxz#_foreach ops don't support autogradr,cpu)r@ryr r|)#rGr-rr3rHrrrallzipr"_group_tensors_by_device_and_dtypevaluesrrr _foreach_negis_cpu _foreach_add_rK _foreach_add_foreach_lerp_ _foreach_mul__foreach_addcmul_ _foreach_pow _foreach_sub_ _foreach_neg_ _foreach_div__foreach_reciprocal__foreach_sqrt__foreach_maximum_ _foreach_sqrt_foreach_addcdiv_rr)(r#r_r`rarbrcrgrhr(rdrirjr$r'r&rr r!grouped_tensorsdevice_params_ device_grads_device_exp_avgs_device_exp_avg_sqs_device_max_exp_avg_sqs_device_state_steps__ device_params device_gradsdevice_exp_avgsdevice_exp_avg_sqsdevice_state_stepsdevice_max_exp_avg_sqsrrrrexp_avg_sq_sqrtr;bcrs( @r8_multi_tensor_adamrs* 6{a"fj S  << $ $ &:'H( $ v{3   wWWsVttu v  w  )"33 3DDD BB + LO  " " $C  T&\>: DL-8 tF|-=>!$v,0CD!$v,0CD )-d6l@RS#("5"56L"M"'"5"56H"I   1F G    5    ;  # #M?O T;M 26EZ---   ;M 26EZ---   ,FV,Wb2g^,WXI7G$HRW$H !$H)-d6l@RS#("5"56L"M"'"5"56H"I   1F G    5  # # CCT  -X$Hs0S;T1T T returnc |sy|r td||j|ini}||j|ini}t| tr&t | jdk7r| j| ind}t j ||||||g}|jD]l\\}}\\}}}}}}}ttt|}ttt|}ttt|} ttt|}!ttt|}"|jdk(r||Jd\}#}$|#|j||j|d}#|#|j||j|d}$|||vr| j|d||<||} tj|"dtj||| |!||"|| | | | |||#|$ |$Jtj |"|$gt#|"zoy) Nz9Adam with fused=True does not support differentiable=Truermps)NNT) non_blocking)r@rr ) r(r$rirjr'r&rrgrh)r3r@r-rstrrritemsrrrrDtorHr _fused_adam_rrG)%r#r_r`rarbrcrgrhr(rdrirjr$r'r&rr r!grad_scale_dictfound_inf_dictlr_dictrr@rrrrrrrrrrrrdevice_grad_scaledevice_found_infs% r8 _fused_adamrlsY* VWW,6+A  J'r*3)>  9%B&b&1c"))n6MBSW  BB + LO    5        "  T&\>: DL-8 tF|-=>!$v,0CD!$v,0CD ;;% $); ;;.8++  ! / : : f4 @!   -88 V$ ?    6#8 ee6eEGFOB .2      " %(& "  '   "%5$6=O9P$P g5r9)single_tensor_fnrr"c h| )|'t||d\}}|rt|tr|sd}| d} |d}tjj st d|Ds td|r)tjjr td| r)tjjr td| r%tjjst}n-|r%tjjst}nt}|||||||| | ||||||||| | y) znFunctional API that performs Adam algorithm computation. See :class:`~torch.optim.Adam` for details. NF) use_fusedc3PK|]}t|tj ywrp)r-rHr)rts r8rzadam..s 3() 1ell#3s$&zPAPI has changed, `state_steps` argument must contain a list of singleton tensorsz6torch.jit.script not supported with foreach optimizersz4torch.jit.script not supported with fused optimizers) r(rdrirjr$r'r&rr r!rgrh) r r-rrHrrrr3r~rrrr)r#r_r`rarbrcrr r!r"rgrhrdr(rirjr$r'r&rrfuncs r8rrs3D }1 Ne 7 z"f-jG } << $ $ &s3-830 ^  599))+STT '')QRR UYY++- //1!"  ! %%r9)NFFNNNF)#typingrrrrrrHr optimizerr r r r rrrrrrrrrrrrr__all__r__doc__rtrJrrrrrQr9r8rs65 * 6 V9Vt&N       'OB NtK LtK <tK6ltKf tK &\ tK f tK tKtKtKtK tK tK eVmtKtK !tK"#tK$%tK&'tKnr Lr <r6lrf r &\ r f r rrrr r r eVmrr !r"#r$%r&'rj^ L^ <^6l^f ^ &\ ^ f ^ ^^^^ ^ ^ eVm^^ !^"#^$%^&'^( )^B 1DE#  #'"&U LU <U6lUf U &\ U f Ud^UUU D>U UUU"#U$ %U& 'U( eVm)U*+U, -U./UFUr9