a h,@sddlZddlmZddlZddlmZddlmZmZddl m Z m Z ddl m Z ddlmZgd ZGd d d eZGd d d eZGdddeZGdddeZGddde eZdS)N)Any)Tensor) functionalinit) ParameterUninitializedParameter)LazyModuleMixin)Module)BilinearIdentity LazyLinearLinearcs:eZdZdZeeddfdd ZeedddZZS) r aA placeholder identity operator that is argument-insensitive. Args: args: any argument (unused) kwargs: any keyword argument (unused) Shape: - Input: :math:`(*)`, where :math:`*` means any number of dimensions. - Output: :math:`(*)`, same shape as the input. Examples:: >>> m = nn.Identity(54, unused_argument1=0.1, unused_argument2=False) >>> input = torch.randn(128, 20) >>> output = m(input) >>> print(output.size()) torch.Size([128, 20]) N)argskwargsreturncstdSNsuper__init__)selfrr __class__E/var/www/auris/lib/python3.9/site-packages/torch/nn/modules/linear.pyr+szIdentity.__init__inputrcCs|Srrrrrrrforward.szIdentity.forward) __name__ __module__ __qualname____doc__rrrr __classcell__rrrrr sr cs|eZdZUdZddgZeed<eed<eed<deeeddfdd Z dd d d Z eed ddZ e d ddZ ZS)raCApplies an affine linear transformation to the incoming data: :math:`y = xA^T + b`. This module supports :ref:`TensorFloat32`. On certain ROCm devices, when using float16 inputs this module will use :ref:`different precision` for backward. Args: in_features: size of each input sample out_features: size of each output sample bias: If set to ``False``, the layer will not learn an additive bias. Default: ``True`` Shape: - Input: :math:`(*, H_\text{in})` where :math:`*` means any number of dimensions including none and :math:`H_\text{in} = \text{in\_features}`. - Output: :math:`(*, H_\text{out})` where all but the last dimension are the same shape as the input and :math:`H_\text{out} = \text{out\_features}`. Attributes: weight: the learnable weights of the module of shape :math:`(\text{out\_features}, \text{in\_features})`. The values are initialized from :math:`\mathcal{U}(-\sqrt{k}, \sqrt{k})`, where :math:`k = \frac{1}{\text{in\_features}}` bias: the learnable bias of the module of shape :math:`(\text{out\_features})`. If :attr:`bias` is ``True``, the values are initialized from :math:`\mathcal{U}(-\sqrt{k}, \sqrt{k})` where :math:`k = \frac{1}{\text{in\_features}}` Examples:: >>> m = nn.Linear(20, 30) >>> input = torch.randn(128, 20) >>> output = m(input) >>> print(output.size()) torch.Size([128, 30]) in_features out_featuresweightTNr$r%biasrcsr||d}t||_||_ttj||ffi||_|rZttj|fi||_n | dd| dS)Ndevicedtyper() rrr$r%rtorchemptyr&r(register_parameterreset_parameters)rr$r%r(r*r+factory_kwargsrrrr]s   zLinear.__init__rcCs`tj|jtdd|jdur\t|j\}}|dkrFdt|nd}t|j| |dS)N)arr)rZkaiming_uniform_r&mathsqrtr(Z_calculate_fan_in_and_fan_outuniform_)rZfan_in_boundrrrr/rs  zLinear.reset_parametersrcCst||j|jSr)FZlinearr&r(rrrrr|szLinear.forwardcCs d|jd|jd|jduS)Nz in_features=, out_features=, bias=)r$r%r(rrrr extra_reprszLinear.extra_repr)TNNrr r!r"Z __constants__int__annotations__rboolrr/rstrr=r#rrrrr2s" % rcs*eZdZdeeeddfdd ZZS)NonDynamicallyQuantizableLinearTNr'cstj|||||ddS)N)r(r*r+r)rr$r%r(r*r+rrrrs z(NonDynamicallyQuantizableLinear.__init__)TNN)rr r!r?rArr#rrrrrCsrCcseZdZUdZgdZeed<eed<eed<eed<deeeedd fd d Z dd d dZ eeedddZ e d ddZ ZS)r aApplies a bilinear transformation to the incoming data: :math:`y = x_1^T A x_2 + b`. Args: in1_features: size of each first input sample, must be > 0 in2_features: size of each second input sample, must be > 0 out_features: size of each output sample, must be > 0 bias: If set to ``False``, the layer will not learn an additive bias. Default: ``True`` Shape: - Input1: :math:`(*, H_\text{in1})` where :math:`H_\text{in1}=\text{in1\_features}` and :math:`*` means any number of additional dimensions including none. All but the last dimension of the inputs should be the same. - Input2: :math:`(*, H_\text{in2})` where :math:`H_\text{in2}=\text{in2\_features}`. - Output: :math:`(*, H_\text{out})` where :math:`H_\text{out}=\text{out\_features}` and all but the last dimension are the same shape as the input. Attributes: weight: the learnable weights of the module of shape :math:`(\text{out\_features}, \text{in1\_features}, \text{in2\_features})`. The values are initialized from :math:`\mathcal{U}(-\sqrt{k}, \sqrt{k})`, where :math:`k = \frac{1}{\text{in1\_features}}` bias: the learnable bias of the module of shape :math:`(\text{out\_features})`. If :attr:`bias` is ``True``, the values are initialized from :math:`\mathcal{U}(-\sqrt{k}, \sqrt{k})`, where :math:`k = \frac{1}{\text{in1\_features}}` Examples:: >>> m = nn.Bilinear(20, 30, 40) >>> input1 = torch.randn(128, 20) >>> input2 = torch.randn(128, 30) >>> output = m(input1, input2) >>> print(output.size()) torch.Size([128, 40]) ) in1_features in2_featuresr%rDrEr%r&TN)rDrEr%r(rcs||d}t|dkr*td|||_||_||_ttj|||ffi||_ |rxttj|fi||_ n | dd| dS)Nr)rz"in1_features must be > 0, but got r() rr ValueErrorrDrEr%rr,r-r&r(r.r/)rrDrEr%r(r*r+r0rrrrs   zBilinear.__init__r1cCsHdt|jd}t|j| ||jdurDt|j| |dS)Nr)r4r5r&sizerr6r()rr8rrrr/s zBilinear.reset_parameters)input1input2rcCst|||j|jSr)r9Zbilinearr&r()rrHrIrrrrszBilinear.forwardc Cs(d|jd|jd|jd|jduS)Nz in1_features=z, in2_features=r:r;)rDrEr%r(r<rrrr=s zBilinear.extra_repr)TNNr>rrrrr s& %r csbeZdZUdZeZeed<eed<dee ddfdd Z dd fd d Z dd d d Z Z S)r aA :class:`torch.nn.Linear` module where `in_features` is inferred. In this module, the `weight` and `bias` are of :class:`torch.nn.UninitializedParameter` class. They will be initialized after the first call to ``forward`` is done and the module will become a regular :class:`torch.nn.Linear` module. The ``in_features`` argument of the :class:`Linear` is inferred from the ``input.shape[-1]``. Check the :class:`torch.nn.modules.lazy.LazyModuleMixin` for further documentation on lazy modules and their limitations. Args: out_features: size of each output sample bias: If set to ``False``, the layer will not learn an additive bias. Default: ``True`` Attributes: weight: the learnable weights of the module of shape :math:`(\text{out\_features}, \text{in\_features})`. The values are initialized from :math:`\mathcal{U}(-\sqrt{k}, \sqrt{k})`, where :math:`k = \frac{1}{\text{in\_features}}` bias: the learnable bias of the module of shape :math:`(\text{out\_features})`. If :attr:`bias` is ``True``, the values are initialized from :math:`\mathcal{U}(-\sqrt{k}, \sqrt{k})` where :math:`k = \frac{1}{\text{in\_features}}` r&r(TN)r%r(rcsH||d}tdddtfi||_||_|rDtfi||_dS)Nr)rF)rrrr&r%r()rr%r(r*r+r0rrrr s  zLazyLinear.__init__r1cs |s|jdkrtdS)Nr)has_uninitialized_paramsr$rr/r<rrrr/szLazyLinear.reset_parameterscCs|rrtR|jd|_|j|j|jf|jdurL|j|jf| Wdn1sh0Y|jdkr|jd|jjdksJd|jdd|jjd|jd|_dS)Nrz%The in_features inferred from input: z/ is not equal to in_features from self.weight: ) rJr,Zno_gradshaper$r&Z materializer%r(r/rrrrinitialize_parameterss   &  z LazyLinear.initialize_parameters)TNN)rr r!r"rZ cls_to_becomerr@r?rArr/rMr#rrrrr s  r )r4typingrr,rZtorch.nnrr9rZtorch.nn.parameterrrZlazyr moduler __all__r rrCr r rrrrs    VV