o
    ôj	w  ã                *   @   sä  d dl mZmZmZmZmZ d dlZd dlmZ d dlm	Z	 ddl
mZmZmZmZmZmZmZmZmZmZmZmZmZmZmZmZmZ ddgZG d	d„ deƒZd
de› de› de› de› de› d� e_dee dee dee dee dee dee dee dee dede de deee f de de dedededef$d d!„Z!dee dee dee dee dee dee dee dee dede de deee f de de dedededef$d"d#„Z"dee dee dee dee dee dee dee dee dede de deee f de de dedededed$df&d%d&„Z#ee!d'�		(	(				(d,dee dee dee dee dee dee d)ee deded*ee dee dee dedede de dee ef de de def(d+d„ƒZ$dS )-é    )ÚcastÚListÚOptionalÚTupleÚUnionN)ÚTensor)Ú$_get_fused_kernels_supported_devicesé   )Ú_capturable_docÚ_default_to_fused_or_foreachÚ_differentiable_docÚ_disable_dynamo_if_unsupportedÚ_dispatch_sqrtÚ_foreach_docÚ
_fused_docÚ!_get_capturable_supported_devicesÚ_get_scalar_dtypeÚ
_get_valueÚ_maximize_docÚ_stack_if_compilingÚ_use_grad_for_differentiableÚ_view_as_realÚ
DeviceDictÚ	OptimizerÚParamsTÚAdamWÚadamwc                       sž   e Zd Z					dddddddœded	eeef d
eeef dedededede	e dedede	e f‡ fdd„Z
‡ fdd„Zdd„ Zeddd„ƒZ‡  ZS )r   çü©ñÒMbP?©gÍÌÌÌÌÌì?g+‡ÙÎ÷ï?ç:Œ0âŽyE>ç{®Gáz„?FN)ÚmaximizeÚforeachÚ
capturableÚdifferentiableÚfusedÚparamsÚlrÚbetasÚepsÚweight_decayÚamsgradr!   r"   r#   r$   r%   c                   s6  d|kst d|› �ƒ‚t|tƒr|r|	st dƒ‚d|ks#t d|› �ƒ‚d|d   kr/dk s9n t d|d › �ƒ‚d|d   krEdk sOn t d	|d › �ƒ‚d|ksZt d
|› �ƒ‚t||||||||	|
|d�
}tƒ  ||¡ |r—|
rwtdƒ‚d| _tƒ ‰ t	‡ fdd„| j
D ƒƒs‘tdˆ › d�ƒ‚|r™tdƒ‚d S d S )Nç        zInvalid learning rate: úElr as a Tensor is not supported for capturable=False and foreach=TruezInvalid epsilon value: r   ç      ð?z#Invalid beta parameter at index 0: r	   z#Invalid beta parameter at index 1: zInvalid weight_decay value: )
r'   r(   r)   r*   r+   r"   r!   r#   r$   r%   z)`fused` does not support `differentiable`Tc                 3   s4   � | ]}|d  D ]}|j jˆ v ot |¡V  qqdS )r&   N)ÚdeviceÚtypeÚtorchÚis_floating_point)Ú.0ÚpgÚp©Úfused_supported_devices© úb/var/www/html/academy/tools/voices/clone/sv-venv/lib/python3.10/site-packages/torch/optim/adamw.pyÚ	<genexpr>S   s   € ýþÿz!AdamW.__init__.<locals>.<genexpr>zX`fused=True` requires all the params to be floating point Tensors of supported devices: Ú.z0`fused` and `foreach` cannot be `True` together.)Ú
ValueErrorÚ
isinstancer   ÚdictÚsuperÚ__init__ÚRuntimeErrorÚ_step_supports_amp_scalingr   ÚallÚparam_groups)Úselfr&   r'   r(   r)   r*   r+   r!   r"   r#   r$   r%   Údefaults©Ú	__class__r6   r9   r@      sX   ÿöþÿÿízAdamW.__init__c                    sâ   t ƒ  |¡ | jD ]e}| dd¡ | dd¡ | dd ¡ | dd¡ | dd¡ | dd ¡}|d D ]:}| j |g ¡}t|ƒd	krmt |d
 ¡smt	|d
 ƒ}|d sW|d rctj
|t|d�|jd�ntj
|tƒ d�|d
< q3q	d S )Nr+   Fr!   r"   r#   r$   r%   r&   r   Ústep©Úis_fused©Údtyper/   ©rM   )r?   Ú__setstate__rD   Ú
setdefaultÚstateÚgetÚlenr1   Ú	is_tensorÚfloatÚtensorr   r/   )rE   rQ   Úgroupr%   r5   Úp_stateÚstep_valrG   r8   r9   rO   _   s2   
ûûýù€üùzAdamW.__setstate__c	                 C   sl  d}	|d D ]­}
|
j d u rq|	t |
¡O }	| |
¡ |
j jr"tdƒ‚| |
j ¡ | j|
 }t|ƒdkrs|d s;|d rItjdt	|d d�|
j
d	�ntjd
t	ƒ d�|d< tj|
tjd�|d< tj|
tjd�|d< |rstj|
tjd�|d< | |d ¡ | |d ¡ |d rŒ| |d ¡ |d r™|d jr™tdƒ‚|d r¬t|d tƒr¬|d s¬tdƒ‚| |d ¡ q|	S )NFr&   z'AdamW does not support sparse gradientsr   r#   r%   r8   rJ   rL   r,   rN   rI   )Úmemory_formatÚexp_avgÚ
exp_avg_sqÚmax_exp_avg_sqr+   r$   zB`requires_grad` is not supported for `step` in differentiable moder"   r'   r-   )Úgradr1   Ú
is_complexÚappendÚ	is_sparserA   rQ   rS   Úzerosr   r/   rV   Ú
zeros_likeÚpreserve_formatÚrequires_gradr=   r   )rE   rW   Úparams_with_gradÚgradsr+   Úexp_avgsÚexp_avg_sqsÚmax_exp_avg_sqsÚstate_stepsÚhas_complexr5   rQ   r8   r8   r9   Ú_init_groupv   sd   


	ûûýù

ÿ
ÿ
ÿÿÿþýÿzAdamW._init_groupc                 C   s  |   ¡  d}|dur!t ¡ � |ƒ }W d  ƒ n1 sw   Y  | jD ]]}g }g }g }g }g }g }	|d }
ttttf |d ƒ\}}|  ||||
||||	¡}t||||||	f|
|||d |d |d |d |d |d	 |d
 |d t	| ddƒt	| ddƒ|dœŽ q$|S )z°Perform 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$   r%   Ú
grad_scaleÚ	found_inf)r+   Úbeta1Úbeta2r'   r*   r)   r!   r"   r#   r$   r%   rn   ro   rl   )
Ú _cuda_graph_capture_health_checkr1   Úenable_gradrD   r   r   rU   rm   r   Úgetattr)rE   ÚclosureÚlossrW   rf   rg   rh   ri   rj   rk   r+   rp   rq   rl   r8   r8   r9   rI   ¿   sb   
ÿ
øú


ìz
AdamW.step)r   r   r   r    F©N)Ú__name__Ú
__module__Ú__qualname__r   r   rU   r   r   Úboolr   r@   rO   rm   r   rI   Ú__classcell__r8   r8   rG   r9   r      sN    ù	óþ
ý
üûúù	÷
öõôó@I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

    r&   rg   rh   ri   rj   rk   rn   ro   r+   rp   rq   r'   r*   r)   r!   r#   r$   rl   c       
   !      C   s¨  |d u r|d u s
J ‚t j ¡ rt|tƒsJ ‚t| ƒD �]6\}}|s%|| n||  }|| }|| }|| }t j ¡ sV|rVtƒ }|j	j
|j	j
krN|j	j
|v sVJ d|› d�ƒ‚t  |¡rzt  |¡}t  |¡}t  |¡}|rut  || ¡||< t  |¡}|d7 }| d||  ¡ | |d|	 ¡ | |
¡j||d|
 d� |s |rø|}d|	|  }d|
|  }|| }| ¡ }| ¡ }|rä|rÅ||  ¡ }n|| }||  t  ||¡¡ ||  ¡ ||   || ¡} n| ¡ ||   || ¡} | || ¡ nEt|ƒ}d|	|  }d|
|  }|| }t|ƒ}|�r+t j|| ||| d� ||  ¡ |  |¡} n	| ¡ |  |¡} |j|| | d� |�rQt  | | ¡�rQt  || ¡||< qd S )NúIIf capturable=True, params and state_steps must be on supported devices: r;   r	   )Úvalue)Úout)r1   ÚjitÚis_scriptingr=   rU   Ú	enumerateÚ_utilsÚis_compilingr   r/   r0   r_   Úview_as_realÚmul_Úlerp_Úaddcmul_ÚnegÚsqrtÚcloneÚcopy_ÚmaximumÚadd_Úaddcdiv_r   r   Úview_as_complex)!r&   rg   rh   ri   rj   rk   rn   ro   r+   rp   rq   r'   r*   r)   r!   r#   r$   rl   ÚiÚparamr^   r[   r\   Ústep_tÚcapturable_supported_devicesrI   Úbias_correction1Úbias_correction2Ú	step_sizeÚstep_size_negÚbias_correction2_sqrtr]   Údenomr8   r8   r9   Ú_single_tensor_adamwB  st   

ÿ





þ
þ€©r›   c       
            sÖ  t | ƒdkrd S tˆtƒr|stdƒ‚tj ¡ s5|r5tdd�‰t‡fdd„t	| |ƒD ƒƒs5J dˆ› d�ƒ‚|r;J d	ƒ‚|d u rC|d u sEJ ‚t
 | |||||g¡}| ¡ D �]\\}}}}}}}|rs|rlt|||||ƒ nt||||ƒ |rzt |¡}|d jr�tj|tjd
dd�d
d� nt |d¡ |dkr¡t |dˆ|  ¡ t ||dˆ  ¡ t |ˆ¡ t |||dˆ ¡ ~|�rt ˆ |¡}t ˆ|¡}t |d¡ t |d¡ t |¡ t |ˆ¡ t |¡ t |¡ |}|}|rýt ||¡ t |¡}nt |¡}t ||¡ t ||¡ t ||¡ t |||¡ qT‡ fdd„|D ƒ}‡fdd„|D ƒ}t‡fdd„|D ƒƒ}dd„ |D ƒ}|�rOt ||¡ t |¡}nt |¡}t ||¡ t ||¡ t ||||¡ qTd S )Nr   r-   F)Úsupports_xlac                 3   s0   � | ]\}}|j j|j jko|j jˆ v V  qd S rw   )r/   r0   )r3   r5   rI   )r”   r8   r9   r:   Û  s   € þ

þz&_multi_tensor_adamw.<locals>.<genexpr>r}   r;   z#_foreach ops don't support autogradr.   Úcpu)r/   )Úalphar	   c                    ó   g | ]
}d ˆ t |ƒ  ‘qS ©r	   ©r   ©r3   rI   )rp   r8   r9   Ú
<listcomp>D  ó    ÿz'_multi_tensor_adamw.<locals>.<listcomp>c                    rŸ   r    r¡   r¢   )rq   r8   r9   r£   G  r¤   c                    s   g | ]}ˆ | d  ‘qS )éÿÿÿÿr8   ©r3   Úbc)r'   r8   r9   r£   K  s    c                 S   s   g | ]}t |ƒ‘qS r8   )r   r¦   r8   r8   r9   r£   M  s    ÿ)rS   r=   r   rA   r1   rƒ   r„   r   rC   Úzipr   Ú"_group_tensors_by_device_and_dtypeÚvaluesr   Ú_foreach_negÚis_cpuÚ_foreach_add_rV   Ú_foreach_mul_Ú_foreach_lerp_Ú_foreach_addcmul_Ú_foreach_powÚ_foreach_sub_Ú_foreach_neg_Ú_foreach_div_Ú_foreach_reciprocal_Ú_foreach_sqrt_Ú_foreach_maximum_Ú_foreach_sqrtÚ_foreach_addcdiv_r   )r&   rg   rh   ri   rj   rk   rn   ro   r+   rp   rq   r'   r*   r)   r!   r#   r$   rl   Úgrouped_tensorsÚdevice_paramsÚdevice_gradsÚdevice_exp_avgsÚdevice_exp_avg_sqsÚdevice_max_exp_avg_sqsÚdevice_state_stepsÚ_r•   r–   r—   r™   Úexp_avg_sq_sqrtr8   )rp   rq   r”   r'   r9   Ú_multi_tensor_adamw¹  sÄ   ÿÿ
ý
üÿ

ùûÿ

ÿÿ




ÿ
ÿÿ
üŒrÃ   Úreturnc       
          C   sb  | sd S |r
t dƒ‚|d ur|j|ini }|d ur|j|ini }t|tƒr1t|jƒdkr1|j|ind }t | |||||g¡}| ¡ D ]l\\}}\\}}}}}}}d\}}|d urc| ||j	|dd�¡}|d urr| ||j	|dd�¡}|d ur…||vr…| ||j	|dd�¡}t
 |d¡ t
j|||||||||	|
|||||d� |d ur®t
 ||gt|ƒ ¡ qBd S )	Nz9Adam with fused=True does not support differentiable=Truer�   )NNT)Únon_blocking)r/   rÅ   r	   )	r+   r'   rp   rq   r*   r)   r!   rn   ro   )rA   r/   r=   r   Ústrr   r©   ÚitemsrP   Útor1   r­   Z_fused_adamw_r²   rS   ) r&   rg   rh   ri   rj   rk   rn   ro   r+   rp   rq   r'   r*   r)   r!   r#   r$   rl   Úgrad_scale_dictÚfound_inf_dictÚlr_dictrº   r/   rÁ   r»   r¼   r½   r¾   r¿   rÀ   Údevice_grad_scaleÚdevice_found_infr8   r8   r9   Ú_fused_adamwd  sx   ÿÿ$ÿÿöÿÿÿñÿ€ÕrÎ   )Úsingle_tensor_fnFr"   r%   c                C   sü   t j ¡ stdd„ |D ƒƒstdƒ‚|	du r.|du r.t| |dd�\}}|r.t|tƒr.|s.d}|	du r4d}	|du r:d}|rEt j 	¡ rEtdƒ‚|	rPt j 	¡ rPtdƒ‚|	rZt j 	¡ sZt
}n|rdt j 	¡ sdt}nt}|| |||||||||||||||
||d	� dS )
zpFunctional API that performs AdamW algorithm computation.

    See :class:`~torch.optim.AdamW` for details.
    c                 s   s   � | ]	}t |tjƒV  qd S rw   )r=   r1   r   )r3   Útr8   r8   r9   r:   Ü  s   € 
ÿzadamw.<locals>.<genexpr>zPAPI has changed, `state_steps` argument must contain a list of singleton tensorsNF)Ú	use_fusedz6torch.jit.script not supported with foreach optimizersz4torch.jit.script not supported with fused optimizers)r+   rp   rq   r'   r*   r)   r!   r#   r$   rn   ro   rl   )r1   rƒ   r„   rC   rA   r   r=   r   r€   r�   rÎ   rÃ   r›   )r&   rg   rh   ri   rj   rk   r"   r#   r$   r%   rn   ro   rl   r+   rp   rq   r'   r*   r)   r!   rÁ   Úfuncr8   r8   r9   r   ¾  sZ   ÿÿ
ÿ
î)NFFNNNF)%Útypingr   r   r   r   r   r1   r   Útorch.utils._foreach_utilsr   Ú	optimizerr
   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   r   Ú__all__r   Ú__doc__r{   rU   r›   rÃ   rÎ   r   r8   r8   r8   r9   Ú<module>   sr  L a'ñðïîíÙÿEÿþýüûúùø
öõô
óòñðïî
íwÿþýüûúùø
öõô
óòñðïî
í ,ÿþýüûúùø
öõô
óòñðïîí
ìZ
ñÿþýüûú	÷
öõôóòñïîí
ìëêé