o H hL@sjUddlmZddlZddlmmZddlmZddlm Z gZ e e e d<ejjGdddZdS))OptionalN)Tensor)2_scripted_functional_optimizer_deprecation_warning__all__c@sleZdZ         ddeededed ed ed ed ed ededefddZdeeefddZ dS)_FunctionalRMSprop{Gz?Gz?:0yE>Fparamslralphaeps weight_decaymomentumcenteredforeachmaximize_allow_empty_param_listc Csxtdd|||||d|_||_||_| |_t|dkr#| s#tdd|i|_tj t tj t t tj ffi|_dS)N) stacklevel)r r rrrrz%optimizer got an empty parameter listr )rdefaultsrrrlen ValueError param_grouptorchjitannotatedictrstrstate) selfr r r rrrrrrrr"^/var/www/vscode/kcb/lib/python3.10/site-packages/torch/distributed/optim/functional_rmsprop.py__init__s  (z_FunctionalRMSprop.__init__ gradientscCs|jd}g}g}g}g}g}g}|jd} |jd} |jd} |jd} |jd} t|t|krEtddt|d d t|d }t||D]|\}}|dur|t|O}||||||jvri|j|<|j|}t d |d <tj |tj d|d<| dkrtj |tj d|d<|j rtj |tj d|d<|j|}||d| dkr||d|j r||d||d qLt tj||||||| | | | | |j |j|j|dWddS1swYdS)Nr r r rrrzEthe gradients passed in does not equal to the size of the parameters!zParams length: z. zGradients length: Fr step) memory_format square_avgrmomentum_buffergrad_avg) r r rrrrrr has_complex)rrrrzipr is_complexappendr tensor zeros_likepreserve_formatrno_gradFrmsproprr)r!r%r params_with_gradgrads square_avgs grad_avgsmomentum_buffer_list state_stepsr r rrrr+paramgradientr r"r"r#r&<s                "z_FunctionalRMSprop.stepN) rrr r r FFFF) __name__ __module__ __qualname__listrfloatboolr$rr&r"r"r"r#rs@   "r)typingrrtorch.optim._functionaloptim _functionalr3r,torch.distributed.optim._deprecation_warningrrr@r__annotations__rscriptrr"r"r"r#s