o ,i w*@sddlmZmZmZmZmZddlZddlmZddlm Z ddl m Z m Z m Z mZmZmZmZmZmZmZmZmZmZmZmZmZmZddgZGd ddeZd d ed ed e d e d ed e_deedeedeedeedeedeedeedeedede de deee fde de dedededef$d d!Z!deedeedeedeedeedeedeedeedede de deee fde de dedededef$d"d#Z"deedeedeedeedeedeedeedeedede de deee fde de dedededed$df&d%d&Z#ee!d'  ( (    (d,deedeedeedeedeedeed)eededed*eedeedeededede de dee efde de def(d+dZ$dS)-)castListOptionalTupleUnionN)Tensor)$_get_fused_kernels_supported_devices)_capturable_doc_default_to_fused_or_foreach_differentiable_doc_disable_dynamo_if_unsupported_dispatch_sqrt _foreach_doc _fused_doc!_get_capturable_supported_devices_get_scalar_dtype _get_value _maximize_doc_stack_if_compiling_use_grad_for_differentiable _view_as_real DeviceDict OptimizerParamsTAdamWadamwcseZdZ     dddddddded eeefd eeefd ed ed edede ededede effddZ fddZ ddZ e dddZZS)rMbP?g?g+?:0yE>{Gz?FN)maximizeforeach capturabledifferentiablefusedparamslrbetaseps weight_decayamsgradr!r"r#r$r%c s6d|ks td|t|tr|r| stdd|ks#td|d|dkr/dks9ntd|dd|dkrEdksOntd |dd|ksZtd |t|||||||| | | d } t|| | r| rwtd d |_tt fdd|j Dstdd|rtddSdS)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: r z#Invalid beta parameter at index 1: zInvalid weight_decay value: ) r'r(r)r*r+r"r!r#r$r%z)`fused` does not support `differentiable`Tc3s4|]}|dD]}|jjvot|VqqdS)r&N)devicetypetorchis_floating_point).0pgpfused_supported_devicesO/home/ubuntu/SoloSpeech/.venv/lib/python3.10/site-packages/torch/optim/adamw.py Ss z!AdamW.__init__..zX`fused=True` requires all the params to be floating point Tensors of supported devices: .z0`fused` and `foreach` cannot be `True` together.) ValueError isinstancerdictsuper__init__ RuntimeError_step_supports_amp_scalingrall param_groups) selfr&r'r(r)r*r+r!r"r#r$r%defaults __class__r6r9r@sX  zAdamW.__init__cst||jD]e}|dd|dd|dd|dd|dd|dd}|dD]:}|j|g}t|d krmt|d smt |d }|dsW|drctj |t |d |j d ntj |t d |d <q3q dS)Nr+Fr!r"r#r$r%r&rstepis_fuseddtyper/rM) r? __setstate__rD setdefaultstategetlenr1 is_tensorfloattensorrr/)rErQgroupr%r5p_statestep_valrGr8r9rO_s2          zAdamW.__setstate__c Csld} |dD]} | jdurq| t| O} || | jjr"td|| j|j| } t| dkrs|ds;|drItjdt |dd| j d ntj d t d | d <tj | tj d | d<tj | tj d | d<|rstj | tj d | d<|| d|| d|dr|| d|dr| d jrtd|drt|dtr|dstd|| d q| S)NFr&z'AdamW does not support sparse gradientsrr#r%r8rJrLr,rNrI) memory_formatexp_avg exp_avg_sqmax_exp_avg_sqr+r$zB`requires_grad` is not supported for `step` in differentiable moder"r'r-)gradr1 is_complexappend is_sparserArQrSzerosrr/rV zeros_likepreserve_format requires_gradr=r) rErWparams_with_gradgradsr+exp_avgs exp_avg_sqsmax_exp_avg_sqs state_steps has_complexr5rQr8r8r9 _init_groupvsd             zAdamW._init_groupcCs|d}|dur!t |}Wdn1swY|jD]]}g}g}g}g}g}g} |d} ttttf|d\} } ||||| |||| } t|||||| f| | | |d|d|d|d|d|d |d |d t |d dt |d d| dq$|S)zPerform a single optimization step. Args: closure (Callable, optional): A closure that reevaluates the model and returns the loss. Nr+r(r'r*r)r!r"r#r$r% grad_scale found_inf)r+beta1beta2r'r*r)r!r"r#r$r%rnrorl) _cuda_graph_capture_health_checkr1 enable_gradrDrrrUrmrgetattr)rEclosurelossrWrfrgrhrirjrkr+rprqrlr8r8r9rIsb      z AdamW.step)rrrr FN)__name__ __module__ __qualname__rrrUrrboolrr@rOrmrrI __classcell__r8r8rGr9rsN         @Ia 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 r&rgrhrirjrkrnror+rprqr'r*r)r!r#r$rlc !Cs|dur|dus Jtjrt| tsJt|D]6\}}|s%||n|| }||}||}||}tjsV|rVt}|j j |j j krN|j j |vsVJd|dt |rzt |}t |}t |}|rut ||||<t |}|d7}| d| | ||d| | | j||d| d|s|r|}d| |}d| |}| |}|}|}|r|r||}n||}||t||||||| |} n |||| |} ||| nEt|}d| |}d| |}| |}t|}|r+tj|||||d|||| } n ||| } |j|| | d|rQt ||rQt||||<qdS)NIIf capturable=True, params and state_steps must be on supported devices: r;r )value)out)r1jit is_scriptingr=rU enumerate_utils is_compilingrr/r0r_ view_as_realmul_lerp_addcmul_negsqrtclonecopy_maximumadd_addcdiv_rrview_as_complex)!r&rgrhrirjrkrnror+rprqr'r*r)r!r#r$rliparamr^r[r\step_tcapturable_supported_devicesrIbias_correction1bias_correction2 step_size step_size_negbias_correction2_sqrtr]denomr8r8r9_single_tensor_adamwBst             rc st|dkrdSttr|stdtjs5|r5tddtfddt ||Ds5Jdd|r;Jd |durC|dusEJt ||||||g}| D]\\}}}}}}}|rs|rlt |||||nt |||||rzt|}|djrtj|tjd d d d d nt|d| dkrt|d| t||dt|t|||d~|rt|}t|}t|dt|dt|t|t|t||}|}|rt||t|}nt|}t||t|| t||t|||qTfdd|D}fdd|D}tfdd|D}dd|D}|rOt||t|}nt|}t||t|| t||||qTdS)Nrr-F) supports_xlac3s0|]\}}|jj|jjko|jjvVqdSrw)r/r0)r3r5rI)rr8r9r:s   z&_multi_tensor_adamw..r}r;z#_foreach ops don't support autogradr.cpu)r/)alphar cg|] }dt|qSr rr3rI)rpr8r9 Dz'_multi_tensor_adamw..crrrr)rqr8r9rGrcsg|]}|dqS)r8r3bc)r'r8r9rKscSsg|]}t|qSr8)rrr8r8r9rMs)rSr=rrAr1rrrrCzipr"_group_tensors_by_device_and_dtypevaluesr _foreach_negis_cpu _foreach_add_rV _foreach_mul__foreach_lerp__foreach_addcmul_ _foreach_pow _foreach_sub_ _foreach_neg_ _foreach_div__foreach_reciprocal__foreach_sqrt__foreach_maximum_ _foreach_sqrt_foreach_addcdiv_r)r&rgrhrirjrkrnror+rprqr'r*r)r!r#r$rlgrouped_tensors device_params device_gradsdevice_exp_avgsdevice_exp_avg_sqsdevice_max_exp_avg_sqsdevice_state_steps_rrrrexp_avg_sq_sqrtr8)rprqrr'r9_multi_tensor_adamws                               rreturnc Csb|sdS|r td|dur|j|ini}|dur|j|ini}t| tr1t| jdkr1| j| ind}t||||||g}|D]l\\}}\\}}}}}}}d\}}|durc|||j |dd}|durr|||j |dd}|dur||vr||| j |dd} t |dt j |||||||| | | | | |||d|durt ||gt|qBdS) Nz9Adam with fused=True does not support differentiable=Truer)NNT) non_blocking)r/rr ) r+r'rprqr*r)r!rnro)rAr/r=rstrrritemsrPtor1r _fused_adamw_rrS) r&rgrhrirjrkrnror+rprqr'r*r)r!r#r$rlgrad_scale_dictfound_inf_dictlr_dictrr/rrrrrrrdevice_grad_scaledevice_found_infr8r8r9 _fused_adamwdsx$  r)single_tensor_fnFr"r%c Cstjstdd|Dstd| dur.|dur.t||dd\}}|r.t|tr.|s.d}| dur4d} |dur:d}|rEtj rEtd| rPtj rPtd| rZtj sZt }n |rdtj sdt }nt }|||||||| ||||||||| | | d dS) zpFunctional API that performs AdamW algorithm computation. See :class:`~torch.optim.AdamW` for details. css|] }t|tjVqdSrw)r=r1r)r3tr8r8r9r:s  zadamw..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+rprqr'r*r)r!r#r$rnrorl) r1rrrCrAr r=rrrrrr)r&rgrhrirjrkr"r#r$r%rnrorlr+rprqr'r*r)r!rfuncr8r8r9rsZ  )NFFNNNF)%typingrrrrrr1rtorch.utils._foreach_utilsr optimizerr r r r rrrrrrrrrrrrr__all__r__doc__r{rUrrrrr8r8r8r9sr  La'E     w     ,     Z