sgA: dZddlmZmZmZddlZddlmcmZ ddl m Z gdZ GddZ Gdd ZGd d Zed e Z ddededededeedef dZddededefdZy)z=Spectral Normalization from https://arxiv.org/abs/1802.05957.)AnyOptionalTypeVarN)Module) SpectralNorm SpectralNormLoadStateDictPreHookSpectralNormStateDictHook spectral_normremove_spectral_normc eZdZUdZeed<eed<eed<eed<eed< ddededededd f d Zde jde jfd Z d e d e de jfdZd e dd fdZd e dedd fdZdZed e dededededdf dZy )r_versionnamedimn_power_iterationsepsweightreturnNcb||_||_|dkrtd|||_||_y)NrzGExpected n_power_iterations to be positive, but got n_power_iterations=)rr ValueErrorrr)selfrrrrs O/var/www/html/venv/lib/python3.12/site-packages/torch/nn/utils/spectral_norm.py__init__zSpectralNorm.__init__"sH   "**<)=? #5c|}|jdk7rR|j|jgt|jDcgc]}||jk7s|c}}|jd}|j |dScc}w)Nr)rpermuterangesizereshape)rr weight_matdheights rreshape_weight_to_matrixz%SpectralNorm.reshape_weight_to_matrix3s} 88q=+++',Z^^-='>P!!txx-APJ#!!&"--Qs B B moduledo_power_iterationc>t||jdz}t||jdz}t||jdz}|j|}|rtj5t |j D]~}tjtj|j|d|j|}tjtj||d|j|}|j dkDr@|jtj}|jtj}dddtj|tj||}||z }|S#1swY:xYw)N_orig_u_vr)rrout) memory_format)getattrrr$torchno_gradrrF normalizemvtrclonecontiguous_formatdot) rr%r&ruvr!_sigmas rcompute_weightzSpectralNorm.compute_weight=sO<W!45 FDII, - FDII, -226:  Gt667YA 3aA EHHZ$;VWXAY**Q.e.E.EFAe.E.EFA G !UXXj!45% ! G Gs 4C&FFctj5|j|d}dddt||jt||jdzt||jdzt||jdz|j |jtj jjy#1swYxYw)NFr&r)r*r() r.r/r;delattrrregister_parameternn Parameterdetach)rr%rs rremovezSpectralNorm.removess ]]_ K((E(JF K " D() D() G+,!!$))UXX-?-? -PQ  K Ks CC#inputscht||j|j||jy)Nr=)setattrrr;training)rr%rDs r__call__zSpectralNorm.__call__|s+  II   6??  K rc ltjj|jj |j |j|j dgjd}|j|tj|tj||z S)Nr ) r.linalg multi_dotr3mmpinverse unsqueezesqueezemul_r6r2)rr!r7 target_sigmar8s r_solve_v_and_rescalez!SpectralNorm._solve_v_and_rescales LL " " \\^  z * 3 3 5z||~q{{ST~ V '!* vvlUYYq%((:q2I%JJKKrc|jjD]0}t|ts|j|k(s$t d|t||||}|j |}|td|dt|tjjjr tdtj5|j|}|j\} } tj |j#| j%ddd|j&} tj |j#| j%ddd|j&} dddt)||j|j+|jdz|t-||j|j.|j1|jd z |j1|jd z |j3||j5t7||j9t;||S#1swYxYw) Nz>Cannot register two spectral_norm hooks on the same parameter z/`SpectralNorm` cannot be applied as parameter `z ` is NonezThe module passed to `SpectralNorm` can't have uninitialized parameters. Make sure to run the dummy forward before applying spectral normalizationrr )rrr(r)r*)_forward_pre_hooksvalues isinstancerr RuntimeError _parametersrr.r@ parameterUninitializedParameterr/r$rr0r1 new_emptynormal_rr>r?rFdataregister_bufferregister_forward_pre_hook_register_state_dict_hookr "_register_load_state_dict_pre_hookr) r%rrrrhookfnrr!hwr7r8s rapplyzSpectralNorm.applys --446 D$ -$))t2C"TUYTZ[  $ 2C =##D) >A$yQ  fehh00GG H\  ]]_ R44Vz.sT Q*4Ts)r(r)r*F)r(r)Tr(r)r*) rcgetrallappendr.r/popmeanr$rR)rr~prefixlocal_metadatastrict missing_keysunexpected_keys error_msgsrcversionhas_missing_keyssuffixkey weight_origrr:r!r7r8rs ` @rrHz)SpectralNormLoadStateDictPreHook.__call__sUWW $$_b9== GGj $  ?gk"'')JT>STTj0 $ - 1 6)j('+$$++C0  1   2(g)=> # 3$v-33588E zD01++J5A01 :,- 2 2)*( 2 2s 0A!DD$rNrhrirjrrHrrrrrrs)2 )2rrceZdZddZddZy)r Nc||_yrurvrws rrz"SpectralNormStateDictHook.__init__rxrcd|vri|d<|jjdz}||dvrtd||jj|d|<y)Nr rzz-Unexpected key in metadata['spectral_norm']: )rcrrWr)rr%r~rrrs rrHz"SpectralNormStateDictHook.__call__s_ . 0.0N? +ggllZ' .1 1!NseTU U/3ww/?/?',rrrrrrrr r s @rr T_module)boundr%rrrrrc|\t|tjjtjjtjj frd}nd}t j||||||S)aAApply spectral normalization to a parameter in the given module. .. math:: \mathbf{W}_{SN} = \dfrac{\mathbf{W}}{\sigma(\mathbf{W})}, \sigma(\mathbf{W}) = \max_{\mathbf{h}: \mathbf{h} \ne 0} \dfrac{\|\mathbf{W} \mathbf{h}\|_2}{\|\mathbf{h}\|_2} Spectral normalization stabilizes the training of discriminators (critics) in Generative Adversarial Networks (GANs) by rescaling the weight tensor with spectral norm :math:`\sigma` of the weight matrix calculated using power iteration method. If the dimension of the weight tensor is greater than 2, it is reshaped to 2D in power iteration method to get spectral norm. This is implemented via a hook that calculates spectral norm and rescales weight before every :meth:`~Module.forward` call. See `Spectral Normalization for Generative Adversarial Networks`_ . .. _`Spectral Normalization for Generative Adversarial Networks`: https://arxiv.org/abs/1802.05957 Args: module (nn.Module): containing module name (str, optional): name of weight parameter n_power_iterations (int, optional): number of power iterations to calculate spectral norm eps (float, optional): epsilon for numerical stability in calculating norms dim (int, optional): dimension corresponding to number of outputs, the default is ``0``, except for modules that are instances of ConvTranspose{1,2,3}d, when it is ``1`` Returns: The original module with the spectral norm hook .. note:: This function has been reimplemented as :func:`torch.nn.utils.parametrizations.spectral_norm` using the new parametrization functionality in :func:`torch.nn.utils.parametrize.register_parametrization`. Please use the newer version. This function will be deprecated in a future version of PyTorch. Example:: >>> m = spectral_norm(nn.Linear(20, 40)) >>> m Linear(in_features=20, out_features=40, bias=True) >>> m.weight_u.size() torch.Size([40]) r r)rVr.r@ConvTranspose1dConvTranspose2dConvTranspose3drrf)r%rrrrs rr r sip {  ((((((  CCvt%7cB MrcZ|jjD]E\}}t|ts|j|k(s'|j ||j|=nt d|d||jjD]>\}}t|ts|jj|k(s1|j|=n|jjD]?\}}t|ts|jj|k(s1|j|=|S|S)aRemove the spectral normalization reparameterization from a module. Args: module (Module): containing module name (str, optional): name of weight parameter Example: >>> m = spectral_norm(nn.Linear(40, 10)) >>> remove_spectral_norm(m) zspectral_norm of 'z' not found in ) rTitemsrVrrrCr_state_dict_hooksr rc_load_state_dict_pre_hooksr)r%rkrbs rr r Qs,,224M4 dL )dii4.? KK ))!,  M -dV?6(KLL++1134 d5 6477<<4;O((+  44::<4 d< =$'',,RVBV11!4  M  Mr)rr rgN)r)__doc__typingrrrr.torch.nn.functionalr@ functionalr0torch.nn.modulesr__all__rrr rrmrkrnr r rrrrrsC)) # eeT6262v @ @ :V ,  E E EE  E # E  EPHr