a h@sjUddlmZddlZddlmmZddlmZddlm Z gZ e e e d<ejjGdddZdS))OptionalN)Tensor)2_scripted_functional_optimizer_deprecation_warning__all__c @sdeZdZdeeeeeefeeeeeeed ddZee ed d d Z ee ed d dZ dS)_FunctionalAdamMbP?g?g+?:0yE>F) paramslrbetaseps weight_decayamsgradmaximizeforeachfused_allow_empty_param_listc Cs$tddd|ks td|d|ks6td|d|dkrNdksbntd|dd|d krzdksntd |d d|kstd ||||d|d |d |_||_||_||_| |_tj t tj t t tj ffi|_ t|dkr| std d|i|_dS)N) stacklevelr zInvalid learning rate: zInvalid epsilon value: rg?z#Invalid beta parameter at index 0: z#Invalid beta parameter at index 1: zInvalid weight_decay value: )r rbeta1beta2rz%optimizer got an empty parameter listr )r ValueErrordefaultsrrrrtorchjitZannotatedictrstrstatelen param_group) selfr r r rrrrrrrr$U/var/www/auris/lib/python3.9/site-packages/torch/distributed/optim/functional_adam.py__init__s2 $z_FunctionalAdam.__init__)paramgradc Csvg}g}g}g}g}g}t|} |dur>||||||jvri|j|<|j|} td| d<tj|tjd| d<tj|tjd| d<|jrtj|tjd| d<|j|} || d|| d|jr|| d|| dtdt j |||||||j| |j |j d|j d |j d |j d |j d |j |jddd Wdn1sh0YdS)zo Similar to step, but operates on a single parameter and optionally a gradient tensor. Nr stepZ memory_formatexp_avg exp_avg_sqmax_exp_avg_sqrrr rr r has_complexrrrr rrrrZ grad_scaleZ found_inf)r is_complexappendr tensor zeros_likepreserve_formatrno_gradFadamrrrr) r#r'r(params_with_gradgradsexp_avgs exp_avg_sqsmax_exp_avg_sqs state_stepsr/r r$r$r% step_paramGsf           z_FunctionalAdam.step_param) gradientsc Cs|jd}g}g}g}g}g}g}d} t|t|kr\tddt|ddt|t|jd|D]\} } | durl| t| O} || || | |jvri|j| <|j| } td| d<tj | tj d | d <tj | tj d | d <|j rtj | tj d | d <|j| } || d || d |j rL|| d || dqlt dt j|||||||j | |j|jd |jd|jd|jd|jd|j|jdddWdn1s0YdS)Nr FzEthe gradients passed in does not equal to the size of the parameters!zParams length: z. zGradients length: r r)r*r+r,r-rrr rrr.)r"r!rziprr0r1r r2r3r4rr5r6r7rrrr) r#r?r r8r9r:r;r<r=r/r'Zgradientr r$r$r%r)s|            z_FunctionalAdam.stepN) rrr r FFFFF) __name__ __module__ __qualname__listrfloattupleboolr&rr>r)r$r$r$r%rs.  -s