
    Ng                     B    d Z ddlZddlZddlmZ  G d de          ZdS )z Adan Optimizer

Adan: Adaptive Nesterov Momentum Algorithm for Faster Optimizing Deep Models[J]. arXiv preprint arXiv:2208.06677, 2022.
    https://arxiv.org/abs/2208.06677

Implementation adapted from https://github.com/sail-sg/Adan
    N)	Optimizerc                        e Zd ZdZ	 	 	 	 	 d fd	Z ej                    d             Z ej                    dd
            Z xZ	S )Adanae  
    Implements a pytorch variant of Adan
    Adan was proposed in
    Adan: Adaptive Nesterov Momentum Algorithm for Faster Optimizing Deep Models[J]. arXiv preprint arXiv:2208.06677, 2022.
    https://arxiv.org/abs/2208.06677
    Arguments:
        params (iterable): iterable of parameters to optimize or dicts defining parameter groups.
        lr (float, optional): learning rate. (default: 1e-3)
        betas (Tuple[float, float, flot], optional): coefficients used for computing
            running averages of gradient and its norm. (default: (0.98, 0.92, 0.99))
        eps (float, optional): term added to the denominator to improve
            numerical stability. (default: 1e-8)
        weight_decay (float, optional): decoupled weight decay (L2 penalty) (default: 0)
        no_prox (bool): how to perform the decoupled weight decay (default: False)
    MbP?g\(\?gq=
ףp?gGz?:0yE>        Fc                    d|k    s"t          d                    |                    d|k    s"t          d                    |                    d|d         cxk    rdk     s*n t          d                    |d                             d|d         cxk    rdk     s*n t          d                    |d                             d|d	         cxk    rdk     s*n t          d
                    |d	                             t          |||||          }t          t          |                               ||           d S )Nr	   zInvalid learning rate: {}zInvalid epsilon value: {}r         ?z%Invalid beta parameter at index 0: {}   z%Invalid beta parameter at index 1: {}   z%Invalid beta parameter at index 2: {})lrbetasepsweight_decayno_prox)
ValueErrorformatdictsuperr   __init__)	selfparamsr   r   r   r   r   defaults	__class__s	           K/var/www/html/ai-engine/env/lib/python3.11/site-packages/timm/optim/adan.pyr   zAdan.__init__!   s_    byy8??CCDDDczz8??DDEEEeAh$$$$$$$$DKKERSHUUVVVeAh$$$$$$$$DKKERSHUUVVVeAh$$$$$$$$DKKERSHUUVVV2U,X_```dD""6844444    c                     | j         D ]k}d|d<   |d         D ][}|j        rR| j        |         }t          j        |          |d<   t          j        |          |d<   t          j        |          |d<   \ld S )Nr   stepr   exp_avg
exp_avg_sqexp_avg_diff)param_groupsrequires_gradstatetorch
zeros_like)r   grouppr%   s       r   restart_optzAdan.restart_opt7   s    & 	@ 	@EE&M8_ 
@ 
@? 	@ JqME (-'7':':E)$*/*:1*=*=E,',1,<Q,?,?E.)
@	@ 	@r   Nc                    d}|5t          j                    5   |            }ddd           n# 1 swxY w Y   | j        D ]}|d         \  }}}d|v r|dxx         dz  cc<   nd|d<   d||d         z  z
  }d||d         z  z
  }d||d         z  z
  }	|d         D ]4}
|
j        |
j        }| j        |
         }t          |          dk    r\t          j        |
          |d<   t          j        |
          |d	<   t          j        |
          |d
<   |                                |d<   |d         |d	         |d
         }}}||d         z
  }|                    |d|z
             |                    |d|z
             |||z  z   }|	                    |          
                    ||d|z
             |                                t          j        |	          z                      |d                   }||z  ||z  |z  z                       |          }|d         rK|
j        	                    d|d         |d         z  z
             |
                    ||d                     nJ|
                    ||d                     |
j                            d|d         |d         z  z              |d                             |           6|S )z. Performs a single optimization step.
        Nr   r   r   r   r   r   r    r"   r!   pre_grad)valuer   r   r   r   )alpha)r&   enable_gradr#   gradr%   lenr'   clonelerp_mul_addcmul_sqrtmathadd_div_datacopy_)r   closurelossr(   beta1beta2beta3bias_correction1bias_correction2bias_correction3r)   r0   r%   r    r!   r"   	grad_diffupdatedenoms                      r   r   z	Adan.stepG   sM    "$$ ! !wyy! ! ! ! ! ! ! ! ! ! ! ! ! ! ! & *	. *	.E"'.E5% f" !f"UeFm%;;"UeFm%;;"UeFm%;;8_ . .6>v
1u::??','7':':E)$,1,<Q,?,?E.)*/*:1*=*=E,'(,

E*%49)4DeNF[]bco]p\ 5#44	dBJ///""9b5j999	 11&&//b5j/QQQ#**TY7G-H-HHNNuUZ|\\!$44u|7KN^7^^ddejkk# IFKKE$K%2G$G GHHHFF6%+F6666FF6%+F666FKKE$K%2G$G GHHHj!''----;.> s   /33)r   r   r   r	   F)N)
__name__
__module____qualname____doc__r   r&   no_gradr*   r   __classcell__)r   s   @r   r   r      s         & $5 5 5 5 5 5, U]__@ @ _@ U]__4 4 4 _4 4 4 4 4r   r   )rJ   r7   r&   torch.optimr   r    r   r   <module>rO      ss       ! ! ! ! ! !l l l l l9 l l l l lr   