sgUddlmZmZmZddlZddlmcmZddlm Z gZ ee e d<ejjGddZy))DictListOptionalN)Tensor__all__cbeZdZ d deedededededededefd Zd eeefd Z y )_FunctionalAdadeltaparamslrrhoeps weight_decayforeachmaximize_allow_empty_param_listc .||||d|_||_||_t|dk(r |s t dd|i|_t jjtt jttt jffi|_ y)N)r r r rrz%optimizer got an empty parameter listr ) defaultsrrlen ValueError param_grouptorchjitannotaterrstrstate) selfr r r r rrrrs ^/var/www/html/venv/lib/python3.12/site-packages/torch/distributed/optim/functional_adadelta.py__init__z_FunctionalAdadelta.__init__s(      v;! $;DE E%f-YY''U\\4U\\@Q;R-R(SUWX gradientscT|jd}g}g}g}g}g}|jd}|jd} |jd} |jd} t|t|k7r*tddt|dzd t|zd } t ||D]"\} }| | t j | z} |j| |j|| |jvri|j| <|j| }t jd |d <t j| t j |d<t j| t j |d<|j| }|j|d|j|d|j|d %t j5tj||||||| | | |j|j |  dddy#1swYyxYw)Nr r r r rzEthe gradients passed in does not equal to the size of the parameters!zParams length: z. zGradients length: Fstep) memory_format square_avg acc_delta)r r r rrr has_complex)rrrrzipr is_complexappendrtensor zeros_likepreserve_formatno_gradFadadeltarr)rr r params_with_gradgrads square_avgs acc_deltas state_stepsr r r rr'paramgradientrs rr#z_FunctionalAdadelta.step4s!!(+   ]]4 mmE"mmE"}}^4 v;#i. (W#CK=34&s9~&678   "695 2OE8#u//66  ''. X& *(*DJJu% JJu-E$)LL$5E&M*/*:*:U-B-B+E,'*/)9)9U-B-B*E+& 5)""5#67!!% "45""5=1) 2,]]_  JJ ) '    s 6HH'N)g?g?gư>r"FFF) __name__ __module__ __qualname__rrfloatboolrrr#rrr r s !(-YV Y Y Y  Y  YYY"&Y:7d8F#347rr )typingrrrrtorch.optim._functionaloptim _functionalr/rrr__annotations__rscriptr r=rrrDsM'' ##cUUUr