o hcl'@sddlZddlmZddlmZmZmZmZmZmZm Z m Z m Z m Z m Z mZddlmZmZmZmZddlmZddgZGd ddeZd d e d e d ed e d e d e_      d*deedeedeedeedeedeedeedededeedeedeededededeeefded ed!ef&d"dZdeedeedeedeedeedeedeedeededededeeefded ed!ededef"d#d$Zdeedeedeedeedeedeedeedeededededeeefded ed!ededef"d%d&Zdeedeedeedeedeedeedeedeededededeeefded ed!ededed'df$d(d)ZdS)+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_foreachparams_t)ListOptionalTupleUnion)$_get_fused_kernels_supported_devicesAdamWadamwcseZdZ     dddddddded eeefd eeefd ed ed edede ededede effddZ fddZ ddZ e dddZZS)rMbP?g?g+?:0yE>{Gz?FN)maximizeforeach capturabledifferentiablefusedparamslrbetaseps weight_decayamsgradrrrrrc s6d|ks td|t|tr|r| stdd|ks#td|d|dkr/dks9ntd|dd|dkrEdksOntd |dd|ksZtd |t|||||||| | | d } t|| | r| rwtd d |_tt fdd|j Dstdd|rtddSdS)NzInvalid learning rate: Elr as a Tensor is not supported for capturable=False and foreach=TruezInvalid epsilon value: rg?z#Invalid beta parameter at index 0: rz#Invalid beta parameter at index 1: zInvalid weight_decay value: ) r!r"r#r$r%rrrrrz)`fused` does not support `differentiable`Tc3s4|]}|dD]}|jjvot|VqqdS)r N)devicetypetorchis_floating_point).0pgpfused_supported_devicesG/var/www/html/ai/venv/lib/python3.10/site-packages/torch/optim/adamw.py ?s  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$r%rrrrrdefaults __class__r/r2r9 sP   zAdamW.__init__cst||jD]&}|dd|dd|dd|dd|dd|ddq t|j}t|dkoEt |dd }|sX|D]}t t |d |d <qJdSdS) Nr%Frrrrrrstep) r8 __setstate__r= setdefaultliststatevalueslenr* is_tensortensorfloat)r>rFgroup state_valuesstep_is_tensorsr@r1r2rCIs"        zAdamW.__setstate__c CsL|dD]} | jdur q|| | jjrtd|| j|j| } t| dkrc|ds2|drrLparams_with_gradgradsr%exp_avgs exp_avg_sqsmax_exp_avg_sqs state_stepsr.rFr1r1r2 _init_groupZsF        zAdamW._init_groupc Cs|d}|dur!t |}Wdn1swY|jD]U}g}g}g}g}g}g} |d} |d\} } ||||| |||| t|||||| f| | | |d|d|d|d|d|d |d |d t|d dt|d dd q$|S)zPerforms 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#rrrrrrcrd) _cuda_graph_capture_health_checkr* enable_gradr=rbrgetattr) r>closurelossrLr\r]r^r_r`rar%rerfr1r1r2rBs`       z AdamW.step)rrrrFN)__name__ __module__ __qualname__rrrKrrboolrr9rCrbrrB __classcell__r1r1r@r2r sN         <:a 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 Fr r]r^r_r`rarrrrrcrdr%rerfr!r$r#rc Cstjstdd|Dstd| dur.|dur.t||dd\}}|r.t|tr.|s.d}| dur4d} |dur:d}|rEtj rEtd| rPtj rPtd| rZtj sZt }n |rdtj sdt }nt }|||||||| | |||||||| | d dS) zpFunctional API that performs AdamW algorithm computation. See :class:`~torch.optim.AdamW` for details. css|] }t|tjVqdSrl)r6r*r)r,tr1r1r2r30szadamw..zPAPI has changed, `state_steps` argument must contain a list of singleton tensorsNF) use_fusedz6torch.jit.script not supported with foreach optimizersz4torch.jit.script not supported with fused optimizers) r%rerfr!r$r#rrrrcrd) r*_utils is_compilingr<r:rr6rjit is_scripting _fused_adamw_multi_tensor_adamw_single_tensor_adamw)r r]r^r_r`rarrrrrcrdr%rerfr!r$r#r_funcr1r1r2rsP c Cs|dur|dus Jtjrt| tsJt|D]-\}}|s%||n|| }||}||}||}tjsM|rM|jrC|jsM|j rI|j sMJdt |rqt |}t |}t |}|rlt ||||<t |}|d7}| d| | | |d| | | j||d| d|s|r|}d| |}d| |}| |}|}|}|r|r||}n||}||t||||||| |}n |||| |}|||nEt|}d| |}d| |}| |}t|}|r"tj|||||d|||| }n ||| }|j||| d|rHt ||rHt||||<qdS)NzGIf capturable=True, params and state_steps must be CUDA or XLA tensors.r)value)out)r*rvrwr6rK enumeratertruis_cudais_xla is_complex view_as_realmul_lerp_addcmul_negsqrtclonecopy_maximumadd_addcdiv_rrview_as_complex)r r]r^r_r`rarcrdr%rerfr!r$r#rrriparamrUrRrSstep_trBbias_correction1bias_correction2 step_size step_size_negbias_correction2_sqrtrTdenomr1r1r2rzds~            rzc st|dkrdSttr|stdtjs*|r*tddt||Ds*Jd|r0Jd|dur8|dus:Jt ||||||g}| D]\\}}}}}}}|r[t |}dd|D}d d|D}d d|D}d d|D}d d|D}t |d | dkrt|d | t||d t|t|||d ~|r t|}t|}t|d t|d t|t|t|t||}|}|rt||t|}nt|}t||t || t||t|||qIfdd|D}fdd|D}tfdd|D}dd|D}|r@t||t|}nt|}t||t || t||||qIdS)Nrr'css |] \}}|jo |jVqdSrl)r)r,r.rBr1r1r2r3s z&_multi_tensor_adamw..z@If capturable=True, params and state_steps must be CUDA tensors.z#_foreach ops don't support autogradcS$g|]}t|rt|n|qSr1r*rrr,xr1r1r2 $z'_multi_tensor_adamw..cSrr1rrr1r1r2r rcSrr1rrr1r1r2r cSrr1rrr1r1r2rrcSrr1rrr1r1r2rrrcg|] }dt|qSrrr,rB)rer1r2rIcrrrr)rfr1r2rJrcsg|]}|dqS)r1r,bc)r!r1r2rLscSsg|]}t|qSr1)rrr1r1r2rNs)rHr6rr:r*rtrur<zipr"_group_tensors_by_device_and_dtyperG _foreach_neg _foreach_add_ _foreach_mul__foreach_lerp__foreach_addcmul_ _foreach_pow _foreach_sub_ _foreach_neg_ _foreach_div__foreach_reciprocal__foreach_sqrt__foreach_maximum_ _foreach_sqrt_foreach_addcdiv_r)r r]r^r_r`rarcrdr%rerfr!r$r#rrrgrouped_tensors device_params device_gradsdevice_exp_avgsdevice_exp_avg_sqsdevice_max_exp_avg_sqsdevice_state_stepsr{rrrrexp_avg_sq_sqrtr1)rerfr!r2rys                           ryreturnc Cs~|sdS|r td|dur|j|ind}|dur|j|ind}t| tr1t| jdkr1| j| ind}t||||||g}|D]z\\}}\\}}}}}}}d\}}|duri||vre|j|dd||<||}|dur~||vrz|j|dd||<||}|dur||vr| j|dd||<||} t |dt j |||||||| | | | | |||d|durt ||gt |qBdS) Nz9Adam with fused=True does not support differentiable=Truecpu)NNT) non_blocking)r(rr) r%r!rerfr$r#rrcrd)r:r(r6rstrrritemstor*r _fused_adamw_rrH)r r]r^r_r`rarcrdr%rerfr!r$r#rrrgrad_scale_dictfound_inf_dictlr_dictrr(r{rrrrrrdevice_grad_scaledevice_found_infr1r1r2rx^sf& rx)NFFNNN)r*r optimizerrrrrrr r r r r rrtypingrrrrtorch.utils._foreach_utilsr__all__r__doc__rprKrrzryrxr1r1r1r2s\ 8 F&K       Q     u