o hTQ!@s8ddlmZmZmZddlZddlmZddlmZmZm Z m Z m Z m Z m Z mZmZmZmZmZmZddgZGdddeZd d ed e d e d e d e_       d)deedeedeedeedeedeedeededeedededededededef ddZd d!Zdeedeedeedeedeedeededededededededefd"d#Zdeedeedeedeedeedeededededededededefd$d%Zdeedeedeedeedeedeededededededededed&dfd'd(ZdS)*)castOptionalUnionN)Tensor) _default_to_fused_or_foreach_device_dtype_check_for_fused_differentiable_doc _foreach_doc_get_scalar_dtype _get_value _maximize_doc _params_doc _to_scalar_use_grad_for_differentiable _view_as_real OptimizerParamsTAdagradadagradcseZdZ      ddddddedeeefd ed ed ed ed eedededeeffddZ fddZ ddZ ddZ e dddZZS)r{Gz?r绽|=NF)maximizedifferentiablefusedparamslrlr_decay weight_decayinitial_accumulator_valueepsforeachrrrc sdt|tr|dkrtdd|kstd|d|ks%td|d|ks0td|d|ks;td|d|ksFtd|t|||||||| | d } t|| | rk| rbtd |rhtd d |_|j D]A} | d D]:} |j | }| drt j dt | dd| jdnt jdt d|d<t | rt||n|}t j| |t jd|d<qtqndS)NrzTensor lr must be 1-elementgzInvalid learning rate: zInvalid lr_decay value: zInvalid weight_decay value: z)Invalid initial_accumulator_value value: zInvalid epsilon value: ) rrr rrr!rrrz)`fused` does not support `differentiable`z0`fused` and `foreach` cannot be `True` together.TrrZis_fused)dtypedevicer$step)Z memory_formatsum) isinstancerZnumel ValueErrordictsuper__init__ RuntimeError"_need_device_dtype_check_for_fused param_groupsstatetorchZzerosr r%tensor is_complexcomplexZ full_likeZpreserve_format)selfrrrrrr r!rrrdefaultsgrouppr1Z init_value __class__r"Y/home/www/facesmatcher.com/frenv_anti/lib/python3.10/site-packages/torch/optim/adagrad.pyr-sh       zAdagrad.__init__cst|d}|jD]}|dd|dd|dd|dd}q t|j}t|dko;t |dd}|sS|D]}tj t |dt |dd |d<q@dSdS) Nr!rFrrrr'r#r&) r, __setstate__r0 setdefaultlistr1valueslenr2Z is_tensorr3floatr )r6r1rr8Z state_valuesZstep_is_tensorsr:r"r<r=bs$       zAdagrad.__setstate__cCs4|jD]}|dD] }|j|}|dq qdS)Nrr()r0r1Z share_memory_)r6r8r9r1r"r"r< share_memoryws   zAdagrad.share_memoryc Csd\}}|dD]E}|jdurM|dr"t|ddr"t|ddd|_||jjO}|t|O}||||j|j|} || d|| d q||fS) N)FFrrr/T)Zcuda_unsupportedFr(r') gradgetattrrr/ is_sparser2r4appendr1) r6r8params_with_gradgrads state_sums state_stepshas_sparse_grad has_complexr9r1r"r"r< _init_group}s&        zAdagrad._init_groupc Csd}|durt |}Wdn1swY|jD]A}g}g}g}g}||||||\}} t|||||d|d|d|d||d|d|d| |d t|d dt|d dd q |S) zPerform a single optimization step. Args: closure (Callable, optional): A closure that reevaluates the model and returns the loss. Nrrrr r!rrr grad_scale found_inf) rrrr rMr!rrrNrrPrQ)r2Z enable_gradr0rOrrF) r6closureZlossr8rIrJrKrLrMrNr"r"r<r's@     z Adagrad.step)rrrrrNN)__name__ __module__ __qualname__rrrBrrboolr-r=rDrOrr' __classcell__r"r"r:r<rsJ      Fa[Implements Adagrad algorithm. .. math:: \begin{aligned} &\rule{110mm}{0.4pt} \\ &\textbf{input} : \gamma \text{ (lr)}, \: \theta_0 \text{ (params)}, \: f(\theta) \text{ (objective)}, \: \lambda \text{ (weight decay)}, \\ &\hspace{12mm} \tau \text{ (initial accumulator value)}, \: \eta\text{ (lr decay)}\\ &\textbf{initialize} : state\_sum_0 \leftarrow \tau \\[-1.ex] &\rule{110mm}{0.4pt} \\ &\textbf{for} \: t=1 \: \textbf{to} \: \ldots \: \textbf{do} \\ &\hspace{5mm}g_t \leftarrow \nabla_{\theta} f_t (\theta_{t-1}) \\ &\hspace{5mm} \tilde{\gamma} \leftarrow \gamma / (1 +(t-1) \eta) \\ &\hspace{5mm} \textbf{if} \: \lambda \neq 0 \\ &\hspace{10mm} g_t \leftarrow g_t + \lambda \theta_{t-1} \\ &\hspace{5mm}state\_sum_t \leftarrow state\_sum_{t-1} + g^2_t \\ &\hspace{5mm}\theta_t \leftarrow \theta_{t-1}- \tilde{\gamma} \frac{g_t}{\sqrt{state\_sum_t}+\epsilon} \\ &\rule{110mm}{0.4pt} \\[-1.ex] &\bf{return} \: \theta_t \\[-1.ex] &\rule{110mm}{0.4pt} \\[-1.ex] \end{aligned} For further details regarding the algorithm we refer to `Adaptive Subgradient Methods for Online Learning and Stochastic Optimization`_. z Args: a lr (float, Tensor, optional): learning rate (default: 1e-2) lr_decay (float, optional): learning rate decay (default: 0) weight_decay (float, optional): weight decay (L2 penalty) (default: 0) initial_accumulator_value (float, optional): initial value of the sum of squares of gradients (default: 0) eps (float, optional): term added to the denominator to improve numerical stability (default: 1e-10) z a fused (bool, optional): whether the fused implementation (CPU only) is used. Currently, `torch.float64`, `torch.float32`, `torch.float16`, and `torch.bfloat16` are supported. (default: None). Please note that the fused implementations does not support sparse or complex gradients. .. _Adaptive Subgradient Methods for Online Learning and Stochastic Optimization: http://jmlr.org/papers/v12/duchi11a.html FrrJrKrLrrPrQrMr!rrNrrrr rc Cstdd|Ds td|dur|durt|| dd\}}|dur$d}|dur*d}|r5tjr5td|r@tjr@td|rJtjsJt}n |rTtjsTt}nt}|||||| | | |||| | ||d dS) ztFunctional API that performs Adagrad algorithm computation. See :class:`~torch.optim.Adagrad` for details. css|] }t|tjVqdSrS)r)r2r).0tr"r"r< szadagrad..zPAPI has changed, `state_steps` argument must contain a list of singleton tensorsNF)Z use_fusedz6torch.jit.script not supported with foreach optimizersz4torch.jit.script not supported with fused optimizers rrrr rMrrrNrPrQ) allr.rr2jit is_scripting_fused_adagrad_multi_tensor_adagrad_single_tensor_adagrad)rrJrKrLrrPrQrMr!rrNrrrr r_funcr"r"r<rsJ  cCs|}t|||SrS)sizer2Zsparse_coo_tensor)rE grad_indicesr@rer"r"r< _make_sparse=srgc Cs|dur|dus Jtjst|}t||||D]\}}}}|d7}t|}| s,|n| }|dkrA|jr:td|j||d}|d|d|}|jr| }| }| }| t |||d||}|  | }|j t ||||| dqt|}|rt|}t|}t|}|j||dd| r|| }n| | }|j||| d|rt|}t|}qdS)Nrrz;weight_decay option is not compatible with sparse gradientsalphavalue)r2r^r_rzipr rGr.addZcoalesceZ_indicesZ_valuesZadd_rgpowZ sparse_maskZsqrt_r4Z view_as_realZaddcmul_sqrtZaddcdiv_Zview_as_complex)rrJrKrLrPrQrrrr rMrrrNparamrEZ state_sumZstep_tr'ZclrrfZ grad_valuesstdZ std_valuesr4r"r"r<rbBsN        rbcs| rJd|dur|dusJt|dkrdStt||||g}|D]\\}}}}}ttt|}ttt|}ttt|}ttt|}| oWtdd|D}|rmt |||||| d| | | ||dq)| rut |||| r|t |}t j s|djrt j|t jddd dd nt |d |dkr| rt j|||d nt j|||d }fd d |D}t j|||d dt |}t || |dks| rt |||}nt ||}t |||q)dS)Nz#_foreach ops don't support autogradrcss|]}|jVqdSrS)rG)rYrEr"r"r<r[s z(_multi_tensor_adagrad..Tr\g?cpu)r%rhrcs&g|]} dt|dqS)r)r )rYr'rrr"r< sz)_multi_tensor_adagrad..rk)rArr"_group_tensors_by_device_and_dtyper@rr?ranyrbrr2Z _foreach_negcompilerZ is_compilingZis_cpu _foreach_add_r3Z _foreach_addZ_foreach_addcmul_Z _foreach_sqrtZ _foreach_mul_Z _foreach_mulZ_foreach_addcdiv_)rrJrKrLrPrQrrrr rMrrrNZgrouped_tensorlistsdevice_params_ device_grads_device_state_sums_device_state_steps_rc device_params device_gradsdevice_state_sumsdevice_state_stepsZdevice_has_sparse_gradZ minus_clrrr numeratorr"rtr<ras             rareturncCs~|sdS| s| r td| rtdt|}|dur|j|ind}|dur*|j|ind}t||||g}|D]\\}}\\}}}}}ttt|}ttt|}ttt|}ttt|}d\}}|dur~|dur~||vrz|j |dd||<||}|dur|dur||vr|j |dd||<||}t |dt j |||||||| | ||d |durt ||gt|q9dS)Nz5`fused` does not support sparse grad or complex paramzs* <'   8     J      A      m