hF<SSKJr SSKJrJrJr SSKrSSKJr SSKJ r SSK J r J r J r JrJrJrJr SSKJr SS KJrJrJr SS KJr SS KJrJr S S KJrJrJ r /SQr!"SS\ 5r""SS\ 5r#"SS\ 5r$S\%\\"\#4S\&\'S\\S\(S\(S\S\$4Sjr)S\SSS S!.r*"S"S#\5r+"S$S%\5r,"S&S'\5r-"S(S)\5r.\"S*S+9\"S,S-4S.9SS/S0S1.S\\\+\4S\(S\(S\S\$4 S2jj55r/\"S3S+9\"S,S44S.9SS/S0S1.S\\\,\4S\(S\(S\S\$4 S5jj55r0\"S6S+9\"S,S74S.9SS/S0S1.S\\\-\4S\(S\(S\S\$4 S8jj55r1\"S9S+9\"S,S:4S.9SS/S0S1.S\\\.\4S\(S\(S\S\$4 S;jj55r2g)<)partial)AnyOptionalUnionN)Tensor) BasicBlock BottleneckResNetResNet18_WeightsResNet50_WeightsResNeXt101_32X8D_WeightsResNeXt101_64X4D_Weights)ImageClassification)register_modelWeights WeightsEnum)_IMAGENET_CATEGORIES)_ovewrite_named_paramhandle_legacy_interface) _fuse_modules _replace_reluquantize_model) QuantizableResNetResNet18_QuantizedWeightsResNet50_QuantizedWeights!ResNeXt101_32X8D_QuantizedWeights!ResNeXt101_64X4D_QuantizedWeightsresnet18resnet50resnext101_32x8dresnext101_64x4dcf^\rSrSrS\S\SS4U4SjjrS\S\4SjrS S \\ SS4S jjr S r U=r $) QuantizableBasicBlock%argskwargsreturnNc>[TU]"U0UD6 [RRR 5UlgN)super__init__torchnn quantizedFloatFunctionaladd_reluselfr(r) __class__s ^/var/www/auris/envauris/lib/python3.13/site-packages/torchvision/models/quantization/resnet.pyr.QuantizableBasicBlock.__init__&s/ $)&)**::< xc&UnURU5nURU5nURU5nURU5nUR U5nUR bUR U5nUR R X25nU$r,)conv1bn1reluconv2bn2 downsampler3r5r:identityouts r7forwardQuantizableBasicBlock.forward*syjjmhhsmiinjjohhsm ?? &q)Hmm$$S3 r9is_qatc|[U/SQSS//USS9 UR(a[URSS/USS9 gg)Nr<r=r>r?r@Tinplace01rrAr5rGs r7 fuse_model QuantizableBasicBlock.fuse_model;s>d57GH&Z^_ ?? $//C:vt L r9)r3r, __name__ __module__ __qualname____firstlineno__rr.rrErboolrP__static_attributes__ __classcell__r6s@r7r&r&%sP=c=S=T=F"M$M4MMr9r&cf^\rSrSrS\S\SS4U4SjjrS\S\4SjrS S \\ SS4S jjr S r U=r $) QuantizableBottleneckAr(r)r*Nc>[TU]"U0UD6 [RR 5Ul[R "SS9Ul[R "SS9Ulg)NFrJ) r-r.r0r1r2 skip_add_reluReLUrelu1relu2r4s r7r.QuantizableBottleneck.__init__BsJ $)&)\\99;WWU+ WWU+ r9r:cUnURU5nURU5nURU5nURU5nUR U5nUR U5nUR U5nURU5nURbURU5nURRX25nU$r,) r<r=rar?r@rbconv3bn3rAr_r3rBs r7rEQuantizableBottleneck.forwardHsjjmhhsmjjojjohhsmjjojjohhsm ?? &q)H  ))#8 r9rGc[U/SQ/SQSS//USS9 UR(a[URSS/USS9 gg) N)r<r=ra)r?r@rbrerfTrJrLrMrNrOs r7rP QuantizableBottleneck.fuse_modelZsH ,.G'SXIYZ\blp  ?? $//C:vt L r9)rarbr_r,rRrZs@r7r\r\AsP,c,S,T, F$M$M4MMr9r\cf^\rSrSrS\S\SS4U4SjjrS\S\4SjrS S \\ SS4S jjr S r U=r $) rbr(r)r*Nc>[TU]"U0UD6 [RRR 5Ul[RRR5Ulgr,) r-r.r/ao quantization QuantStubquant DeQuantStubdequantr4s r7r.QuantizableResNet.__init__csI $)&)XX**446 xx,,88: r9r:clURU5nURU5nURU5nU$r,)rp _forward_implrr)r5r:s r7rEQuantizableResNet.forwardis3 JJqM   q ! LLOr9rGc[U/SQUSS9 UR5H:n[U5[Ld[U5[LdM)UR U5 M< g)zFuse conv/bn/relu modules in resnet models Fuse conv+bn+relu/ Conv+relu/conv+Bn modules to prepare for quantization. Model is modified in place. Note that this operation does not change numerics and the model after modification is in floating point rITrJN)rmodulestyper\r&rP)r5rGms r7rPQuantizableResNet.fuse_modelrsH d4fdKAAw//47>S3S V$ r9)rrrpr,rRrZs@r7rrbsL;c;S;T; F %$ %4 % %r9rblocklayersweightsprogressquantizer)r*c `UbM[US[URS55 SUR;a[USURS5 URSS5n[ X40UD6n[ U5 U(a [ Xv5 UbURURUSS95 U$)N num_classes categoriesbackendfbgemmT)r check_hash) rlenmetapoprrrload_state_dictget_state_dict)r|r}r~rrr)rmodels r7_resnetrsfmSl9S5TU  $ !&)W\\)5L MjjH-G e 6v 6E%u& g44hSW4XY Lr9)rrrzdhttps://github.com/pytorch/vision/tree/main/references/classification#post-training-quantized-modelsz These weights were produced by doing Post Training Quantization (eager mode) on top of the unquantized weights listed below. )min_sizerrrecipe_docsc d\rSrSr\"S\"\SS90\ES\RSSSS .0S S S .ES 9r \ r Sr g)rzJhttps://download.pytorch.org/models/quantized/resnet18_fbgemm_16fa66dd.pth crop_sizei(^ ImageNet-1KgV-_Q@g r8V@zacc@1zacc@5g/$?g`"y&@ num_params unquantized_metrics_ops _file_sizeurl transformsrN) rSrTrUrVrrr _COMMON_METAr IMAGENET1K_V1IMAGENET1K_FBGEMM_V1DEFAULTrXrr9r7rrs[" X.#>    "+99##     "#Gr9rc \rSrSr\"S\"\SS90\ES\RSSSS .0S S S .ES 9r \"S\"\SSS90\ES\RSSSS .0S SS .ES 9r \ r Srg)rzJhttps://download.pytorch.org/models/quantized/resnet50_fbgemm_bf931d71.pthrri(rg{GR@gjt4W@rgB`"[@gM8@rrzJhttps://download.pytorch.org/models/quantized/resnet50_fbgemm-23753f79.pthr resize_sizeg5^I T@gX9vW@g8@rN)rSrTrUrVrrrrr rr IMAGENET1K_V2IMAGENET1K_FBGEMM_V2rrXrr9r7rrs" X.#>    "+99##     "# X.#3O    "+99##     "#Gr9rc \rSrSr\"S\"\SS90\ES\RSSSS .0S S S .ES 9r \"S\"\SSS90\ES\RSSSS .0S SS .ES 9r \ r Srg)rzQhttps://download.pytorch.org/models/quantized/resnext101_32x8_fbgemm_09835ccf.pthrri(JrgvS@gQW@rgDli0@gV-U@rrzQhttps://download.pytorch.org/models/quantized/resnext101_32x8_fbgemm-ee16d00c.pthrrg~jT@g rX@gzGU@rN)rSrTrUrVrrrrr rrrrrrXrr9r7rrs" _.#>    "3AA##     "# _.#3O    "3AA##     "#Gr9rc h\rSrSr\"S\"\SSS90\ESS\RSS S S .0S S S.ES9r \ r Sr g)r izRhttps://download.pytorch.org/models/quantized/resnext101_64x4d_fbgemm-605a1cb3.pthrrri(mz+https://github.com/pytorch/vision/pull/5935rgxT@g/X@rgQ.@g$cT@)rrrrrrrrN) rSrTrUrVrrrrrrrrrXrr9r7r r s`" `.#3O    "C3AA##     $#Gr9r quantized_resnet18)name pretrainedcpURSS5(a[R$[R$NrF)getrrr rr)s r7r1zz*e,, & : : 0!// 0r9)r~TF)r~rrc pU(a[O[RU5n[[/SQXU40UD6$)aCResNet-18 model from `Deep Residual Learning for Image Recognition `_ .. note:: Note that ``quantize = True`` returns a quantized model with 8 bit weights. Quantized models only support inference and run on CPUs. GPU inference is not yet supported. Args: weights (:class:`~torchvision.models.quantization.ResNet18_QuantizedWeights` or :class:`~torchvision.models.ResNet18_Weights`, optional): The pretrained weights for the model. See :class:`~torchvision.models.quantization.ResNet18_QuantizedWeights` 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. quantize (bool, optional): If True, return a quantized version of the model. Default is False. **kwargs: parameters passed to the ``torchvision.models.quantization.QuantizableResNet`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.quantization.ResNet18_QuantizedWeights :members: .. autoclass:: torchvision.models.ResNet18_Weights :members: :noindex: )rrrr)rr verifyrr&r~rrr)s r7r!r!4^-5(:JRRSZ[G (,8 ^W] ^^r9quantized_resnet50cpURSS5(a[R$[R$r)rrrr rrs r7rrSrr9c pU(a[O[RU5n[[/SQXU40UD6$)aCResNet-50 model from `Deep Residual Learning for Image Recognition `_ .. note:: Note that ``quantize = True`` returns a quantized model with 8 bit weights. Quantized models only support inference and run on CPUs. GPU inference is not yet supported. Args: weights (:class:`~torchvision.models.quantization.ResNet50_QuantizedWeights` or :class:`~torchvision.models.ResNet50_Weights`, optional): The pretrained weights for the model. See :class:`~torchvision.models.quantization.ResNet50_QuantizedWeights` below for more details, and possible values. By default, no pre-trained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. quantize (bool, optional): If True, return a quantized version of the model. Default is False. **kwargs: parameters passed to the ``torchvision.models.quantization.QuantizableResNet`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.quantization.ResNet50_QuantizedWeights :members: .. autoclass:: torchvision.models.ResNet50_Weights :members: :noindex: )rr)rr rrr\rs r7r"r"Orr9quantized_resnext101_32x8dcpURSS5(a[R$[R$r)rrrr rrs r7rr1zz*e,, . B B 8)77 8r9c U(a[O[RU5n[USS5 [USS5 [ [ /SQXU40UD6$)aResNeXt-101 32x8d model from `Aggregated Residual Transformation for Deep Neural Networks `_ .. note:: Note that ``quantize = True`` returns a quantized model with 8 bit weights. Quantized models only support inference and run on CPUs. GPU inference is not yet supported. Args: weights (:class:`~torchvision.models.quantization.ResNeXt101_32X8D_QuantizedWeights` or :class:`~torchvision.models.ResNeXt101_32X8D_Weights`, optional): The pretrained weights for the model. See :class:`~torchvision.models.quantization.ResNet101_32X8D_QuantizedWeights` 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. quantize (bool, optional): If True, return a quantized version of the model. Default is False. **kwargs: parameters passed to the ``torchvision.models.quantization.QuantizableResNet`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.quantization.ResNeXt101_32X8D_QuantizedWeights :members: .. autoclass:: torchvision.models.ResNeXt101_32X8D_Weights :members: :noindex: groups width_per_grouprrr)rr rrrr\rs r7r#r#M^5=0BZbbcjkG&(B/&"3Q7 (-H _X^ __r9quantized_resnext101_64x4dcpURSS5(a[R$[R$r)rr rrrrs r7rrrr9c U(a[O[RU5n[USS5 [USS5 [ [ /SQXU40UD6$)aResNeXt-101 64x4d model from `Aggregated Residual Transformation for Deep Neural Networks `_ .. note:: Note that ``quantize = True`` returns a quantized model with 8 bit weights. Quantized models only support inference and run on CPUs. GPU inference is not yet supported. Args: weights (:class:`~torchvision.models.quantization.ResNeXt101_64X4D_QuantizedWeights` or :class:`~torchvision.models.ResNeXt101_64X4D_Weights`, optional): The pretrained weights for the model. See :class:`~torchvision.models.quantization.ResNet101_64X4D_QuantizedWeights` below for more details, and possible values. By default, no pre-trained weights are used. progress (bool, optional): If True, displays a progress bar of the download to stderr. Default is True. quantize (bool, optional): If True, return a quantized version of the model. Default is False. **kwargs: parameters passed to the ``torchvision.models.quantization.QuantizableResNet`` base class. Please refer to the `source code `_ for more details about this class. .. autoclass:: torchvision.models.quantization.ResNeXt101_64X4D_QuantizedWeights :members: .. autoclass:: torchvision.models.ResNeXt101_64X4D_Weights :members: :noindex: r@rrr)r rrrrr\rs r7r$r$rr9)3 functoolsrtypingrrrr/torch.nnr0rtorchvision.models.resnetrr r r r r rtransforms._presetsr_apirrr_metar_utilsrrutilsrrr__all__r&r\rrylistintrWrrrrrr r!r"r#r$rr9r7rsA'' 777(C?? MJM8MJMB%%: +-BBC D Ik "     4&t  # #*## ##L## ##L# #,)*   MQ &_ e57GGH I&_&_ &_  &_  &_ +&_R)*   MQ &_ e57GGH I&_&_ &_  &_  &_ +&_R12   ]a (` e=?WWX Y(`(` (`  (`  (` 3(`V12   ]a (` e=?WWX Y(`(` (`  (`  (` 3(`r9