EiDSSKrSSKJr SSKJrJr SSKrSSKJr SSKJ r SSK J r SSK J r SS KJrJrJr SS KJr SS KJrJr /S QrS r"SS\R25rS\S\S\S\S\S\S\S\R:4SjrS8S\S\S\S\4SjjrS\S\ \4Sjr!"SS \RR25r"S!\S"S#.r#"S$S%\5r$"S&S'\5r%"S(S)\5r&"S*S+\5r'S\S,\\S-\(S.\S\"4 S/jr)\"5\"S0\$RT4S19SS2S3.S,\\$S-\(S.\S\"4S4jj55r+\"5\"S0\%RT4S19SS2S3.S,\\%S-\(S.\S\"4S5jj55r,\"5\"S0\&RT4S19SS2S3.S,\\&S-\(S.\S\"4S6jj55r-\"5\"S0\'RT4S19SS2S3.S,\\'S-\(S.\S\"4S7jj55r.g)9N)partial)AnyOptional)Tensor)ImageClassification)_log_api_usage_once)register_modelWeights WeightsEnum)_IMAGENET_CATEGORIES)_ovewrite_named_paramhandle_legacy_interface) MNASNetMNASNet0_5_WeightsMNASNet0_75_WeightsMNASNet1_0_WeightsMNASNet1_3_Weights mnasnet0_5 mnasnet0_75 mnasnet1_0 mnasnet1_3g0U0*3?c`^\rSrSrSS\S\S\S\S\S\SS 4U4S jjjrS \S\4S jrS r U=r $)_InvertedResidual"in_chout_ch kernel_sizestrideexpansion_factor bn_momentumreturnNc>[TU]5 US;a[SU35eUS;a[SU35eX-nX:H=(a US:HUl[R "[R "XSSS9[R"XvS9[R"S S 9[R "XwX3S -XGSS 9[R"XvS9[R"S S 9[R "XrSSS9[R"X&S95Ul g) Nr rz#stride should be 1 or 2 instead of )z(kernel_size should be 3 or 5 instead of r F)biasmomentumTinplacerpaddingr groupsr() super__init__ ValueErrorapply_residualnn SequentialConv2d BatchNorm2dReLUlayers) selfrrrr r!r"mid_ch __class__s W/home/mande/repo/quber/.venv/lib/python3.13/site-packages/torchvision/models/mnasnet.pyr1_InvertedResidual.__init__#s   B6(KL L f $G }UV V)#o=&A+mm IIeQU 3 NN6 8 GGD ! IIfk!;KTZpu v NN6 8 GGD ! IIfae 4 NN6 8   inputcnUR(aURU5U-$URU5$)Nr3r9)r:r@s r=forward_InvertedResidual.forward;s.   ;;u%- -;;u% %r?rB)g?) __name__ __module__ __qualname____firstlineno__intfloatr1rrC__static_attributes__ __classcell__r<s@r=rr"s`ru  "% 47 AD X[ jo   0&V&&&r?rrrrr exp_factorrepeatsr"r#c US:a[SU35e[XX#XFS9n/n[SU5Hn UR[XUSXFS95 M [R "U/UQ76$)z&Creates a stack of inverted residuals.r z$repeats should be >= 1, instead got )r")r2rrangeappendr4r5) rrrr rNrOr"first remaining_s r=_stackrVBsp{?yIJJ e[* fEI 1g *6;:op == + ++r?valdivisor round_up_biascSUs=:aS:dO [SU35e[U[XS- -5U-U-5nX2U-:aU$X1-$)zAsymmetric rounding to make `val` divisible by `divisor`. With default bias, will round up, unless the number is no more than 10% greater than the smaller divisible value, i.e. (83, 8) -> 80, but (84, 8) -> 88.?zIround_up_bias should be greater than 0.0 and smaller than 1.0 instead of r)r2maxrI)rWrXrYnew_vals r=_round_to_multiple_ofr_Ps`  $ $derdstuu'3sq[01W>> model = MNASNet(1.0, num_classes=1000) >>> x = torch.rand(1, 3, 224, 224) >>> y = model(x) >>> y.dim() 2 >>> y.nelement() 1000 rr` num_classesdropoutr#Nc>[TU]5 [U5 US::a[SU35eXlX l[ U5n[R"SUSSSSSS9[R"US[S 9[R"S S 9[R"USUSSSSUSSS 9[R"US[S 9[R"S S 9[R"USUSSSSSS9[R"US[S 9[USUSSSSS[5[USUSS SSS[5[USUSS SSS[5[USUS SSSS[5[US USS SSS[5[USUSSSSS[5[R"USSSSSSS9[R"S[S 9[R"S S 9/n[R"U6Ul[R"[R"US S9[R "SU55UlUR%5GHmn['U[R5(ac[R(R+UR,SSS9 UR.b+[R(R1UR.5 MM['U[R5(aT[R(R3UR,5 [R(R1UR.5 M['U[R 5(dGM[R(R5UR,SSS9 [R(R1UR.5 GMp g)Nr[z,alpha should be greater than 0.0 instead of r&rr rFr.r r(r)Tr+r-r'i)pr,fan_outrelu)mode nonlinearitysigmoid)r0r1r r2r`rorlr4r6r7 _BN_MOMENTUMr8rVr5r9DropoutLinear classifiermodules isinstanceinitkaiming_normal_weightr(zeros_ones_kaiming_uniform_)r:r`rorprjr9mr<s r=r1MNASNet.__init__ps D! C<KE7ST T &U# IIaAq G NN6!9| < GGD ! IIfQiAq6RS9[` a NN6!9| < GGD ! IIfQiAq O NN6!9| < 6!9fQiAq!\ B 6!9fQiAq!\ B 6!9fQiAq!\ B 6!9fQiAq!\ B 6!9fQiAq!\ B 6!9fQiAq!\ B IIfQiq!AE J NN4, 7 GGD !+ .mmV, -- Wd(KRYYW[]hMijA!RYY''''yv'V66%GGNN166*&Ar~~.. ahh'qvv&Aryy))(( PY(Zqvv& r?xclURU5nURSS/5nURU5$)Nrr&)r9meanr)r:rs r=rCMNASNet.forwards/ KKN FFAq6Nq!!r? state_dictprefixlocal_metadatastrict missing_keysunexpected_keys error_msgsc>URSS5nUS;a[SU35eUS:XGa:URS:XGd)[UR5n [R "SSSSSS S 9[R "S[S 9[R"S S 9[R "SSSSSSS S9[R "S[S 9[R"S S 9[R "SSSSSS S 9[R "S[S 9[SU SSSSS[5/ n [U 5HupXRU 'M SUl [R"S[5 [ T U]EXX4XVU5 g)Nversionr%z+version should be set to 1 or 2 instead of r r\r&rbrFrrr)Tr+r-rcraA new version of MNASNet model has been implemented. Your checkpoint was saved using the previous version. This checkpoint will load and work as before, but you may want to upgrade by training a newer model or transfer learning from an updated ImageNet checkpoint.)getr2r`rlr4r6r7r|r8rV enumerater9_versionwarningswarn UserWarningr0_load_from_state_dict)r:rrrrrrrrrjv1_stemidxlayerr<s r=rMNASNet._load_from_state_dictsQ!$$Y5 & J7)TU U a< c 1 !,F !RAaeDrL9% "b!Qq%PrL9% "b!QquErL9r6!9aAq,? G(0 #( C 1DM MMI    %  Wa r?)rr`rr9ro)ig?)rErFrGrH__doc__rrJrIr1rrCdictstrboollistrrKrLrMs@r=rras H,'e,'#,'u,'W[,','\""F" / / /  /  / 3i / c/ I/  / / r?r)r r z(https://github.com/1e100/mnasnet_trainer)min_size categoriesrecipec P\rSrSr\"S\"\SS90\ESSSSS .0S S S S .ES9r\r Sr g)rzIhttps://download.pytorch.org/models/mnasnet0.5_top1_67.823-3ffadce67e.pth crop_sizei! ImageNet-1Kg"P@g(\U@zacc@1zacc@5g9v?g;O.!@9These weights reproduce closely the results of the paper. num_params_metrics_ops _file_size_docsurl transformsmetaN rErFrGrHr rr _COMMON_META IMAGENET1K_V1DEFAULTrKrr?r=rrsT W.#>    !##  T  M"Gr?rc T\rSrSr\"S\"\SSS90\ESSSS S S .0S S SS.ES9r\r Sr g)rz`_. rrrrrrrrNrrr?r=rrs[ J.#3O  C!##    M*Gr?rc P\rSrSr\"S\"\SS90\ESSSSS .0S S S S .ES9r\r Sr g)ri zIhttps://download.pytorch.org/models/mnasnet1.0_top1_73.512-f206786ef8.pthrriPBrgw/]R@gq= ףV@rgjt?g ףp=0@rrrrNrrr?r=rr sT W.#>    !##   T  M"Gr?rc T\rSrSr\"S\"\SSS90\ESSSS S S .0S S SS.ES9r\r Sr g)ri z;https://download.pytorch.org/models/mnasnet1_3-a4c69d6f.pthrrrri_rgMb S@g rhaW@rgE?g">8@rrrrNrrr?r=rr s[ I.#3O  C!##    M*Gr?rweightsprogresskwargsc Ub#[US[URS55 [U40UD6nU(aUR UR USS95 U$)NrorT)r check_hash)rlenrrload_state_dictget_state_dict)r`rrrmodels r=_mnasnetr9sXfmSl9S5TU E $V $E g44hSW4XY Lr? pretrained)rT)rrc F[RU5n[SX40UD6$)anMNASNet with depth multiplier of 0.5 from `MnasNet: Platform-Aware Neural Architecture Search for Mobile `_ paper. Args: weights (:class:`~torchvision.models.MNASNet0_5_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.MNASNet0_5_Weights` below for more details, and possible values. By default, no pre-trained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to the ``torchvision.models.mnasnet.MNASNet`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.MNASNet0_5_Weights :members: g?)rverifyrrrrs r=rrE&.!''0G C 5f 55r?c F[RU5n[SX40UD6$)arMNASNet with depth multiplier of 0.75 from `MnasNet: Platform-Aware Neural Architecture Search for Mobile `_ paper. Args: weights (:class:`~torchvision.models.MNASNet0_75_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.MNASNet0_75_Weights` below for more details, and possible values. By default, no pre-trained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to the ``torchvision.models.mnasnet.MNASNet`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.MNASNet0_75_Weights :members: g?)rrrrs r=rras&."((1G D' 6v 66r?c F[RU5n[SX40UD6$)anMNASNet with depth multiplier of 1.0 from `MnasNet: Platform-Aware Neural Architecture Search for Mobile `_ paper. Args: weights (:class:`~torchvision.models.MNASNet1_0_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.MNASNet1_0_Weights` below for more details, and possible values. By default, no pre-trained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to the ``torchvision.models.mnasnet.MNASNet`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.MNASNet1_0_Weights :members: r\)rrrrs r=rr}rr?c F[RU5n[SX40UD6$)anMNASNet with depth multiplier of 1.3 from `MnasNet: Platform-Aware Neural Architecture Search for Mobile `_ paper. Args: weights (:class:`~torchvision.models.MNASNet1_3_Weights`, optional): The pretrained weights to use. See :class:`~torchvision.models.MNASNet1_3_Weights` below for more details, and possible values. By default, no pre-trained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. **kwargs: parameters passed to the ``torchvision.models.mnasnet.MNASNet`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.MNASNet1_3_Weights :members: g?)rrrrs r=rrrr?)g?)/r functoolsrtypingrrtorchtorch.nnr4rtransforms._presetsrutilsr _apir r r _metar_utilsrr__all__r|ModulerrIrJr5rVr_rrlrrrrrrrrrrrrrrr?r=rs 5'66'B  & &@ ,  , ,*- ,7: ,HK ,VY ,hm ,]] ,LuLsL5LSVLIuIcIr ehhoor l&8 *+2*2 E H[$9 T UX ]d ,0B0P0P!QR:>QU68$676$6ad6ip6S64,0C0Q0Q!RS<@SW7H%897D7cf7kr7T74,0B0P0P!QR:>QU68$676$6ad6ip6S64,0B0P0P!QR:>QU68$676$6ad6ip6S6r?