o #iU@sjUddlmZmZmZmZddlZddlmmZ ddlm Z gZ ee e d<ejjGdddZdS))DictListOptionalTupleN)Tensor__all__c@seZdZ         ddeededeeefd ed ed ed ed ededefddZdede efddZ dee efddZ dS)_FunctionalAdamMbP?g?g+?:0yE>Fparamslrbetaseps weight_decayamsgradmaximizeforeachfused_allow_empty_param_listc Csd|ks td|d|kstd|d|dkr"dks,ntd|dd|dkr8dksBntd|dd|ksMtd ||||d|d|d |_||_||_||_| |_tjt tj t t tj ffi|_ t |dkr| std d |i|_dS) Nr 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: )rrbeta1beta2rz%optimizer got an empty parameter listr ) ValueErrordefaultsrrrrtorchjitannotaterrstrstatelen param_group) selfr rrrrrrrrrr$e/home/ubuntu/SoloSpeech/.venv/lib/python3.10/site-packages/torch/distributed/optim/functional_adam.py__init__s0 $z_FunctionalAdam.__init__paramgradc Csvg}g}g}g}g}g}t|} |dur||||||jvrVi|j|<|j|} td| d<tj|tjd| d<tj|tjd| d<|jrVtj|tjd| d<|j|} || d|| d|jrs|| d|| dt3t j |||||||j| |j |j d|j d |j d |j d |j d |j |jddd WddS1swYdS)zo Similar to step, but operates on a single parameter and optionally a gradient tensor. Nr step memory_formatexp_avg exp_avg_sqmax_exp_avg_sqrrrrr r has_complexrrrrrrrr grad_scale found_inf)r is_complexappendr tensor zeros_likepreserve_formatrno_gradFadamrrrr) r#r'r(params_with_gradgradsexp_avgs exp_avg_sqsmax_exp_avg_sqs state_stepsr0r r$r$r% step_paramBsh          "z_FunctionalAdam.step_param gradientsc Cs|jd}g}g}g}g}g}g}d} t|t|kr.tddt|ddt|t|jd|D]t\} } | dur| t| O} || || | |jvri|j| <|j| } td| d<tj | tj d | d <tj | tj d | d <|j rtj | tj d | d <|j| } || d || d |j r|| d || dq6t 3t j|||||||j | |j|jd |jd|jd|jd|jd|j|jdddWddS1swYdS)Nr FzEthe gradients passed in does not equal to the size of the parameters!zParams length: z. zGradients length: r r)r*r,r-r.rrrrrr/)r"r!rziprr3r4r r5r6r7rr8r9r:rrrr) r#rBr r;r<r=r>r?r@r0r'gradientr r$r$r%r)~s           "z_FunctionalAdam.stepN) r r r r FFFFF) __name__ __module__ __qualname__rrfloatrboolr&rrAr)r$r$r$r%rsB    ,s