U KcB@sdZddlZddlmZddlmmZddlmZddlm Z ddl m Z ddl m Z mZmZddlmZddlmmmZddlZdd d d d d gZGdddejZGdd d ejZGdd d ejZGdd d ejZGdd d ejZGdd d ejZdS)z*Dynamically quantized convolution modules.N)Tensor)ops) _size_1_t)_single_pair_triple)_reverse_repeat_paddingConv1dConv2dConv3dConvTranspose1dConvTranspose2dConvTranspose3dc sbeZdZdZejZdZdZde e e e e e e e e d fdd Z d d Zdee ed d dZZS)r aA dynamically quantized conv module with floating point tensors as inputs and outputs. For details on input arguments, parameters, and implementation see :class:`~torch.nn.Conv1d` and :class:`~torch.nn.quantized.dynamic.Conv1d` and Attributes: weight (Tensor): packed tensor derived from the learnable weight parameter. scale (Tensor): scalar for the output scale zero_point (Tensor): scalar for the output zero point See :class:`~torch.nn.Conv1d` for other attributes. Examples:: >>> m = nn.quantized.dynamic.Conv1d(16, 33, 3, stride=2) >>> input = torch.randn(20, 16, 100) >>> # xdoctest: +SKIP >>> output = m(input) NrTzeros) in_channels out_channels kernel_sizestridepaddingdilationgroupsbias padding_modec  sttd|| | d} t|}t|}t|tr<|nt|}t|}tt|j ||||||||| f | dSNfThe current implementation of the {} module has poor numerical accuracy and its use is not recommendeddevicedtype) warningswarnformat _get_namer isinstancestrsuperr __init__)selfrrrrrrrrrrr reduce_rangefactory_kwargs __class__N/tmp/pip-unpacked-wheel-gikjz4vx/torch/ao/nn/quantized/dynamic/modules/conv.pyr&-s.   zConv1d.__init__cCsdS)NZDynamicQuantizedConv1dr,r'r,r,r-r"IszConv1d._get_nameinputr(returncCsVt|jdkrtd|jdkrDt|jdd}tj|||jd}tj ||j |S)N Input shape must be `(N, C, L)`!rrmode) lenshape ValueErrorrrrFpadr quantizedZconv1d_dynamic_packed_paramsr'r0r(Z _reversed_padding_repeated_twicer,r,r-forwardLs zConv1d.forward) rrrrTrNNT)T)__name__ __module__ __qualname____doc__nnr _FLOAT_MODULE_NNIQAT_CONV_BN_MODULE_NNI_CONV_RELU_MODULEintrboolr$r&r"rr> __classcell__r,r,r*r-r s4csLeZdZdZejZdZdZdfdd Z d d Z de e e d d d Z ZS)r a|A dynamically quantized conv module with floating point tensors as inputs and outputs. For details on input arguments, parameters, and implementation see :class:`~torch.nn.Conv2d` and :class:`~torch.nn.quantized.dynamic.Conv2d` and Attributes: weight (Tensor): packed tensor derived from the learnable weight parameter. scale (Tensor): scalar for the output scale zero_point (Tensor): scalar for the output zero point See :class:`~torch.nn.Conv2d` for other attributes. Examples:: >>> # With square kernels and equal stride >>> m = nn.quantized.dynamic.Conv2d(16, 33, 3, stride=2) >>> # non-square kernels and unequal stride and with padding >>> m = nn.quantized.dynamic.Conv2d(16, 33, (3, 5), stride=(2, 1), padding=(4, 2)) >>> # non-square kernels and unequal stride and with padding and dilation >>> m = nn.quantized.dynamic.Conv2d(16, 33, (3, 5), stride=(2, 1), padding=(4, 2), dilation=(3, 1)) >>> input = torch.randn(20, 16, 50, 100) >>> # xdoctest: +SKIP >>> output = m(input) NrrTrc sftd|| | d} t|}t|}t|}t|}tt|j||||||||| f | dSr)rr r!r"rr%r r& r'rrrrrrrrrrrr)r*r,r-r&xs.  zConv2d.__init__cCsdS)NZDynamicQuantizedConv2dr,r.r,r,r-r"szConv2d._get_namer/cCsNt|jdkrtd|jdkrs  zConv2d.forward)rrrrTrNN)T)r?r@rArBrCr rDrErFr&r"rrHr>rIr,r,r*r-r YscsLeZdZdZejZdZdZdfdd Z d d Z de e e d d d Z ZS)r aA dynamically quantized conv module with floating point tensors as inputs and outputs. For details on input arguments, parameters, and implementation see :class:`~torch.nn.Conv3d` and :class:`~torch.nn.quantized.dynamic.Conv3d` and Attributes: weight (Tensor): packed tensor derived from the learnable weight parameter. scale (Tensor): scalar for the output scale zero_point (Tensor): scalar for the output zero point See :class:`~torch.nn.Conv3d` for other attributes. Examples:: >>> # With square kernels and equal stride >>> m = nn.quantized.dynamic.Conv3d(16, 33, 3, stride=2) >>> # non-square kernels and unequal stride and with padding >>> m = nn.quantized.dynamic.Conv3d(16, 33, (3, 5, 5), stride=(1, 2, 2), padding=(1, 2, 2)) >>> # non-square kernels and unequal stride and with padding and dilation >>> m = nn.quantized.dynamic.Conv3d(16, 33, (3, 5, 5), stride=(1, 2, 2), padding=(1, 2, 2), dilation=(1, 2, 2)) >>> input = torch.randn(20, 16, 56, 56, 56) >>> # xdoctest: +SKIP >>> output = m(input) NrrTrc s~td|| dks$td| | d} t|}t|}t|}t|}tt|j||||||dtd||| f | dS)NrZreflectz*Conv3d does not support reflection paddingrFr) rr r!r"AssertionErrorrr%r _initrJr*r,r-r&s4  zConv3d.__init__cCsdS)NZDynamicQuantizedConv3dr,r.r,r,r-r"szConv3d._get_namer/cCsNt|jdkrtd|jdkrs  zConv3d.forward)rrrrTrNN)T)r?r@rArBrCr rDrErFr&r"rrHr>rIr,r,r*r-r sc sDeZdZdZejZdfdd Zd d Zde e e d d d Z Z S)r a_A dynamically quantized transposed convolution module with floating point tensors as inputs and outputs. For details on input arguments, parameters, and implementation see :class:`~torch.nn.ConvTranspose1d`. For special notes, please, see :class:`~torch.nn.quantized.dynamic.Conv1d` Attributes: weight (Tensor): packed tensor derived from the learnable weight parameter. scale (Tensor): scalar for the output scale zero_point (Tensor): scalar for the output zero point See :class:`~torch.nn.ConvTranspose1d` for other attributes. Examples:: >>> # With square kernels and equal stride >>> # xdoctest: +SKIP >>> m = nndq.ConvTranspose1d(16, 33, 3, stride=2) >>> # non-square kernels and unequal stride and with padding >>> m = nndq.ConvTranspose1d(16, 33, (3, 5), stride=(2, 1), padding=(4, 2)) >>> output = m(input) >>> # exact output size can be also specified as an argument >>> downsample = nndq.Conv1d(16, 16, 3, stride=2, padding=1) >>> upsample = nndq.ConvTranspose1d(16, 16, 3, stride=2, padding=1) >>> h = downsample(input) >>> h.size() torch.Size([1, 16, 6]) >>> output = upsample(h, output_size=input.size()) >>> output.size() torch.Size([1, 16, 12]) rrTrNc  sHtd|| | d} tt|j||||||||| | f | dSr)rr r!r"r%r r&r'rrrrrZoutput_paddingrrrrrrr)r*r,r-r&s(  zConvTranspose1d.__init__cCsdS)NZDynamicQuantizedConvTranpose1dr,r.r,r,r-r" szConvTranspose1d._get_namer/cCs*t|jdkrtdtjj||j|S)Nr2r3)r6r7r8torchrr;Zconv_transpose1d_dynamicr<r'r0r(r,r,r-r>szConvTranspose1d.forward) rrrrTrrNN)T) r?r@rArBrCr rDr&r"rrHr>rIr,r,r*r-r s! c sDeZdZdZejZdfdd Zd d Zde e e d d d Z Z S)r abA dynamically quantized transposed convolution module with floating point tensors as inputs and outputs. For details on input arguments, parameters, and implementation see :class:`~torch.nn.ConvTranspose2d`. For special notes, please, see :class:`~torch.nn.quantized.dynamic.Conv2d` Attributes: weight (Tensor): packed tensor derived from the learnable weight parameter. scale (Tensor): scalar for the output scale zero_point (Tensor): scalar for the output zero point See :class:`~torch.nn.ConvTranspose2d` for other attributes. Examples:: >>> # With square kernels and equal stride >>> m = nnq.ConvTranspose2d(16, 33, 3, stride=2) >>> # non-square kernels and unequal stride and with padding >>> m = nnq.ConvTranspose2d(16, 33, (3, 5), stride=(2, 1), padding=(4, 2)) >>> # xdoctest: +SKIP >>> output = m(input) >>> # exact output size can be also specified as an argument >>> downsample = nnq.Conv2d(16, 16, 3, stride=2, padding=1) >>> upsample = nnq.ConvTranspose2d(16, 16, 3, stride=2, padding=1) >>> h = downsample(input) >>> h.size() torch.Size([1, 16, 6, 6]) >>> output = upsample(h, output_size=input.size()) >>> output.size() torch.Size([1, 16, 12, 12]) rrTrNc  sHtd|| | d} tt|j||||||||| | f | dSr)rr r!r"r%r r&rPr*r,r-r&<s(  zConvTranspose2d.__init__cCsdS)NZDynamicQuantizedConvTranpose2dr,r.r,r,r-r"IszConvTranspose2d._get_namer/cCs(t|jdkrtdtj||j|S)NrKrL)r6r7r8rr;Zconv_transpose2d_dynamicr<rRr,r,r-r>LszConvTranspose2d.forward) rrrrTrrNN)T) r?r@rArBrCr rDr&r"rrHr>rIr,r,r*r-r s! c sDeZdZdZejZdfdd Zd d Zde e e d d d Z Z S)rapA dynamically quantized transposed convolution module with floating point tensors as inputs and outputs. For details on input arguments, parameters, and implementation see :class:`~torch.nn.ConvTranspose3d`. For special notes, please, see :class:`~torch.nn.quantized.dynamic.Conv3d` Attributes: weight (Tensor): packed tensor derived from the learnable weight parameter. scale (Tensor): scalar for the output scale zero_point (Tensor): scalar for the output zero point See :class:`~torch.nn.ConvTranspose3d` for other attributes. Examples:: >>> # With cubic kernels and equal stride >>> m = nnq.ConvTranspose3d(16, 33, 3, stride=2) >>> # non-cubic kernels and unequal stride and with padding >>> m = nnq.ConvTranspose3d(16, 33, (3, 3, 5), stride=(2, 1, 1), padding=(4, 2, 2)) >>> # xdoctest: +SKIP >>> output = m(input) >>> # exact output size can be also specified as an argument >>> downsample = nnq.Conv3d(16, 16, 3, stride=2, padding=1) >>> upsample = nnq.ConvTranspose3d(16, 16, 3, stride=2, padding=1) >>> h = downsample(input) >>> h.size() torch.Size([1, 16, 6, 6, 6]) >>> output = upsample(h, output_size=input.size()) >>> output.size() torch.Size([1, 16, 12, 12, 12]) rrTrNc  sHtd|| | d} tt|j||||||||| | f | dSr)rr r!r"r%rr&rPr*r,r-r&ys(  zConvTranspose3d.__init__cCsdS)NZDynamicQuantizedConvTranpose3dr,r.r,r,r-r"szConvTranspose3d._get_namer/cCs(t|jdkrtdtj||j|S)NrOz&Input shape must be `(N, C, T, H, W)`!)r6r7r8rr;Zconv_transpose3d_dynamicr<rRr,r,r-r>szConvTranspose3d.forward) rrrrTrrNN)T) r?r@rArBrCrrDr&r"rrHr>rIr,r,r*r-rUs! )rBrQZtorch.nnrCZtorch.nn.functionalZ functionalr9rZ torch._opsrZtorch.nn.common_typesrZtorch.nn.modules.utilsrrrZ"torch.ao.nn.quantized.modules.convrZtorch.ao.nn.quantizedZaor;Znnqr__all__r r r r r rr,r,r,r-s"     GAA==