o m9:jY#@sDddlmZddlZddlmZddlmZmZmZmZm Z m Z m Z m Z m Z mZmZmZmZmZddgZGdddeZd d e d ed e d ed e_       d)deedeedeedeededBdedBdedBdededBdedededededededdf"d dZd!d"ZdeedeedeedeededBdedBdededededededededdfd#d$ZdeedeedeedeededBdedBdededededededededdfd%d&ZdeedeedeedeededBdedBdeeBdedededededededdfd'd(ZdS)*)castN)Tensor)_default_to_fused_or_foreach_device_dtype_check_for_fused_differentiable_doc _foreach_doc_get_scalar_dtype _get_value _maximize_doc _params_doc _to_scalar_use_grad_for_differentiable _view_as_real DeviceDict OptimizerParamsTAdagradadagradcseZdZ      ddddddedeeBd ed ed ed ed edBdedededBddffddZfddZdddZ ddZ e dddZ Z S)r{Gz?r绽|=NF)maximizedifferentiablefusedparamslrlr_decay weight_decayinitial_accumulator_valueepsforeachrrrreturnc sht|tr|dkrtdd|kstd|d|ks%td|d|ks0td|d|ks;td|d|ksFtd||||||||| | d } t|| | rm| ratd |rgtd d |_d |_|j D]A} | d D]:} |j | }| drt j dt | dd| jdnt jdt d|d<t | rt||n|}t j| |t jd|d<qvqpdS)NrzTensor lr must be 1-elementzInvalid learning rate: zInvalid lr_decay value: zInvalid weight_decay value: z)Invalid initial_accumulator_value value: zInvalid epsilon value: ) rrrrrr rrrz)`fused` does not support `differentiable`z0`fused` and `foreach` cannot be `True` together.Trris_fuseddtypedevicer'step memory_formatsum) isinstancernumel ValueErrorsuper__init__ RuntimeError"_need_device_dtype_check_for_fused_step_supports_amp_scaling param_groupsstatetorchzerosr r(tensor is_complexcomplex full_likepreserve_format)selfrrrrrrr rrrdefaultsgrouppr7 init_value __class__r#Z/home/nk/hobo-godmode/plappi-mvp/.venv/lib/python3.10/site-packages/torch/optim/adagrad.pyr2sh       zAdagrad.__init__c s$t|d}|jD]U}|dd|dd|dd|dd}|dD]6}|j|g}t|dkr_t|ds_t |d}|drUtj |t |d |j d ntj |t d |d<q)q t |j}t|dkovt|dd}|s|D]} tj t | dt |d d | d<q{dSdS) Nr rFrrrrr*r$r&r))r1 __setstate__r6 setdefaultr7getlenr8 is_tensorfloatr:r r(listvalues) r?r7rrArBp_statestep_val state_valuesstep_is_tensorsrDr#rFrGds@          zAdagrad.__setstate__cCs4|jD]}|dD] }|j|}|dq qdS)z6Calls tensor.share_memory_() on the state sum tensors.rr-N)r6r7 share_memory_)r?rArBr7r#r#rF share_memorys   zAdagrad.share_memoryc Cs,d\}}|dD]}|jdur|dr t|ddr t|d|_||jjO}|t|O}||||j|j|} t | dkr|drKt||dr]tj dt |dd |j d ntj d t d | d <|jd} t|rvt| | n| } tj|| tjd| d<|| d|| d q||fS)N)FFrrr4TFrr#r$r&r"r)r*rr+r-)gradgetattrrr4 is_sparser8r;appendr7rJr9r r(r:r@r<r=r>) r?rAparams_with_gradgrads state_sums state_stepshas_sparse_grad has_complexrBr7rrCr#r#rF _init_groupsP            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. Nrrrrr rrr grad_scale found_inf) rrrrr^r rrr_rrarb)r8 enable_gradr6r`rrW) r?closurelossrArZr[r\r]r^r_r#r#rFr*s@     z Adagrad.step)rrrrrN)r!NN)__name__ __module__ __qualname__rrLrboolr2rGrUr`rr* __classcell__r#r#rDrFrsN      G #.a[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 and CUDA 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 Frr[r\r]rrarbr^r rr_rrrrrr!c 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|tjVqdSrf)r.r8r).0tr#r#rF 6szadagrad..zPAPI has changed, `state_steps` argument must contain a list of singleton tensorsNF) use_fusedz6torch.jit.script not supported with foreach optimizersz4torch.jit.script not supported with fused optimizers rrrrr^rrr_rarb) allr3rr8jit is_scripting_fused_adagrad_multi_tensor_adagrad_single_tensor_adagrad)rr[r\r]rrarbr^r rr_rrrrr_funcr#r#rFrsJ  cCs|}t|||Srf)sizer8sparse_coo_tensor)rV grad_indicesrNryr#r#rF _make_sparsegsr|c Cs|dus|dur tdtjst|}t||||ddD]\}}}}|d7}t|}| s0|n| }|dkrE|jr>td|j ||d}|d|d|}|jr| }| }| }| t|||d||}|  | }|j t||||| dqt|}|rt|}t|}t|}|j||dd | r|| }n| | }|j||| d |rt|}t|}qdS) N,Expected grad_scale and found_inf to be NoneT)strictrrz;weight_decay option is not compatible with sparse gradientsalphavalue)AssertionErrorr8rrrsr zipr rXr3addcoalesce_indices_valuesadd_r|pow sparse_masksqrt_r; view_as_realaddcmul_sqrtaddcdiv_view_as_complex)rr[r\r]rarbrrrrr^rrr_paramrV state_sumstep_tr*clrr{ grad_valuesstd std_valuesr;r#r#rFrvlsT         rvcs| rtd|dus|durtdt|dkrdStt||||g}|D]\\}}}}}ttt|}ttt|}ttt|}ttt|}| oYt dd|D}|rot |||||| d| | | ||dq+| rwt |||| r~t |}t js|djrt j|t jdd d 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 autogradr}rcss|]}|jVqdSrf)rX)rlrVr#r#rFrns z(_multi_tensor_adagrad..Trpg?cpu)r(rrcs&g|]} dt|dqS)r)r )rlr*rrr#rF sz)_multi_tensor_adagrad..r)rrJr r"_group_tensors_by_device_and_dtyperNrrMranyrvrr8 _foreach_negcompiler is_compilingis_cpu _foreach_add_r: _foreach_add_foreach_addcmul_ _foreach_sqrt _foreach_mul_ _foreach_mul_foreach_addcdiv_)rr[r\r]rarbrrrrr^rrr_grouped_tensorlistsdevice_params_ device_grads_device_state_sums_device_state_steps_rw device_params device_gradsdevice_state_sumsdevice_state_stepsdevice_has_sparse_grad minus_clrr numeratorr#rrFrus            rucCs|sdS| s| r td| rtd|dur|j|ini}|dur&|j|ini}t|tr9t|jdkr9|j|ind}t||||g}|D]\\}}\\}}}}}tt t|}tt t|}tt t|}tt t|}d\}}|dur| ||j |dd}|dur| ||j |dd}|dur||vr|j |dd||<||}t |dt j|||||||| | ||d |durt ||gt|qHdS) Nz5`fused` does not support sparse grad or complex paramzs6  @P   8     J      D      o