U Mh9@sddlZddlZddlmZddlmZddlmZddl m Z ddl m Z dd lm Z mZdd lmZmZmZmZd d d ddgZGdd d eZGdd d eZeeeeefZGdd d eZGdddeZGdddeZdS)N) Parameter)Module) CrossMapLRN2d) functional)init)TensorSize)UnionListOptionalTupleLocalResponseNormr LayerNorm GroupNormRMSNormcsveZdZUdZddddgZeed<eed<eed<eed<deeeed d fd d Ze e d ddZ ddZ Z S)raApplies local response normalization over an input signal. The input signal is composed of several input planes, where channels occupy the second dimension. Applies normalization across channels. .. math:: b_{c} = a_{c}\left(k + \frac{\alpha}{n} \sum_{c'=\max(0, c-n/2)}^{\min(N-1,c+n/2)}a_{c'}^2\right)^{-\beta} Args: size: amount of neighbouring channels used for normalization alpha: multiplicative factor. Default: 0.0001 beta: exponent. Default: 0.75 k: additive factor. Default: 1 Shape: - Input: :math:`(N, C, *)` - Output: :math:`(N, C, *)` (same shape as input) Examples:: >>> lrn = nn.LocalResponseNorm(2) >>> signal_2d = torch.randn(32, 5, 24, 24) >>> signal_4d = torch.randn(16, 5, 7, 7, 7, 7) >>> output_2d = lrn(signal_2d) >>> output_4d = lrn(signal_4d) sizealphabetak-C6???Nrrrrreturncs&t||_||_||_||_dSNsuper__init__rrrrselfrrrr __class__P/var/www/html/venv/lib/python3.8/site-packages/torch/nn/modules/normalization.pyr3s  zLocalResponseNorm.__init__inputrcCst||j|j|j|jSr)FZlocal_response_normrrrrr!r'r$r$r%forward:szLocalResponseNorm.forwardcCsdjf|jSNz){size}, alpha={alpha}, beta={beta}, k={k}format__dict__r!r$r$r% extra_repr>szLocalResponseNorm.extra_repr)rrr) __name__ __module__ __qualname____doc__ __constants__int__annotations__floatrr r*r0 __classcell__r$r$r"r%rs  csleZdZUeed<eed<eed<eed<deeeedd fd d Zeed d dZe dddZ Z S)rrrrrrrrNrcs&t||_||_||_||_dSrrr r"r$r%rHs  zCrossMapLRN2d.__init__r&cCst||j|j|j|jSr)_cross_map_lrn2dapplyrrrrr)r$r$r%r*OszCrossMapLRN2d.forwardrcCsdjf|jSr+r,r/r$r$r%r0SszCrossMapLRN2d.extra_repr)rrr) r1r2r3r6r7r8rr r*strr0r9r$r$r"r%rBs cseZdZUdZdddgZeedfed<eed<e ed<de ee e dd fd d Z dd d dZ e e dddZed ddZZS)ra Applies Layer Normalization over a mini-batch of inputs. This layer implements the operation as described in the paper `Layer Normalization `__ .. math:: y = \frac{x - \mathrm{E}[x]}{ \sqrt{\mathrm{Var}[x] + \epsilon}} * \gamma + \beta The mean and standard-deviation are calculated over the last `D` dimensions, where `D` is the dimension of :attr:`normalized_shape`. For example, if :attr:`normalized_shape` is ``(3, 5)`` (a 2-dimensional shape), the mean and standard-deviation are computed over the last 2 dimensions of the input (i.e. ``input.mean((-2, -1))``). :math:`\gamma` and :math:`\beta` are learnable affine transform parameters of :attr:`normalized_shape` if :attr:`elementwise_affine` is ``True``. The standard-deviation is calculated via the biased estimator, equivalent to `torch.var(input, unbiased=False)`. .. note:: Unlike Batch Normalization and Instance Normalization, which applies scalar scale and bias for each entire channel/plane with the :attr:`affine` option, Layer Normalization applies per-element scale and bias with :attr:`elementwise_affine`. This layer uses statistics computed from input data in both training and evaluation modes. Args: normalized_shape (int or list or torch.Size): input shape from an expected input of size .. math:: [* \times \text{normalized\_shape}[0] \times \text{normalized\_shape}[1] \times \ldots \times \text{normalized\_shape}[-1]] If a single integer is used, it is treated as a singleton list, and this module will normalize over the last dimension which is expected to be of that specific size. eps: a value added to the denominator for numerical stability. Default: 1e-5 elementwise_affine: a boolean value that when set to ``True``, this module has learnable per-element affine parameters initialized to ones (for weights) and zeros (for biases). Default: ``True``. bias: If set to ``False``, the layer will not learn an additive bias (only relevant if :attr:`elementwise_affine` is ``True``). Default: ``True``. Attributes: weight: the learnable weights of the module of shape :math:`\text{normalized\_shape}` when :attr:`elementwise_affine` is set to ``True``. The values are initialized to 1. bias: the learnable bias of the module of shape :math:`\text{normalized\_shape}` when :attr:`elementwise_affine` is set to ``True``. The values are initialized to 0. Shape: - Input: :math:`(N, *)` - Output: :math:`(N, *)` (same shape as input) Examples:: >>> # NLP Example >>> batch, sentence_length, embedding_dim = 20, 5, 10 >>> embedding = torch.randn(batch, sentence_length, embedding_dim) >>> layer_norm = nn.LayerNorm(embedding_dim) >>> # Activate module >>> layer_norm(embedding) >>> >>> # Image Example >>> N, C, H, W = 20, 5, 10, 10 >>> input = torch.randn(N, C, H, W) >>> # Normalize over the last three dimensions (i.e. the channel and spatial dimensions) >>> # as shown in the image below >>> layer_norm = nn.LayerNorm([C, H, W]) >>> output = layer_norm(input) .. image:: ../_static/img/nn/layer_norm.jpg :scale: 50 % normalized_shapeepselementwise_affine.h㈵>TN)r>r?r@biasrcs||d}tt|tjr&|f}t||_||_||_|jrt t j |jf||_ |rtt t j |jf||_ q|ddn|dd|dd|dS)NdevicedtyperBweight)rr isinstancenumbersIntegraltupler>r?r@rtorchemptyrFrBregister_parameterreset_parameters)r!r>r?r@rBrDrEfactory_kwargsr"r$r%rs      zLayerNorm.__init__r<cCs,|jr(t|j|jdk r(t|jdSr)r@rones_rFrBzeros_r/r$r$r%rNs  zLayerNorm.reset_parametersr&cCst||j|j|j|jSr)r(Z layer_normr>rFrBr?r)r$r$r%r*szLayerNorm.forwardcCsdjf|jS)NF{normalized_shape}, eps={eps}, elementwise_affine={elementwise_affine}r,r/r$r$r%r0szLayerNorm.extra_repr)rATTNN)r1r2r3r4r5rr6r7r8bool_shape_trrNr r*r=r0r9r$r$r"r%rZs M cseZdZUdZddddgZeed<eed<eed<eed<deeeedd fd d Z dd d dZ e e dddZ e d ddZZS)raApplies Group Normalization over a mini-batch of inputs. This layer implements the operation as described in the paper `Group Normalization `__ .. math:: y = \frac{x - \mathrm{E}[x]}{ \sqrt{\mathrm{Var}[x] + \epsilon}} * \gamma + \beta The input channels are separated into :attr:`num_groups` groups, each containing ``num_channels / num_groups`` channels. :attr:`num_channels` must be divisible by :attr:`num_groups`. The mean and standard-deviation are calculated separately over the each group. :math:`\gamma` and :math:`\beta` are learnable per-channel affine transform parameter vectors of size :attr:`num_channels` if :attr:`affine` is ``True``. The standard-deviation is calculated via the biased estimator, equivalent to `torch.var(input, unbiased=False)`. This layer uses statistics computed from input data in both training and evaluation modes. Args: num_groups (int): number of groups to separate the channels into num_channels (int): number of channels expected in input eps: a value added to the denominator for numerical stability. Default: 1e-5 affine: a boolean value that when set to ``True``, this module has learnable per-channel affine parameters initialized to ones (for weights) and zeros (for biases). Default: ``True``. Shape: - Input: :math:`(N, C, *)` where :math:`C=\text{num\_channels}` - Output: :math:`(N, C, *)` (same shape as input) Examples:: >>> input = torch.randn(20, 6, 10, 10) >>> # Separate 6 channels into 3 groups >>> m = nn.GroupNorm(3, 6) >>> # Separate 6 channels into 6 groups (equivalent with InstanceNorm) >>> m = nn.GroupNorm(6, 6) >>> # Put all 6 channels into a single group (equivalent with LayerNorm) >>> m = nn.GroupNorm(1, 6) >>> # Activating the module >>> output = m(input) num_groups num_channelsr?affinerATN)rUrVr?rWrcs||d}t||dkr(td||_||_||_||_|jrpttj |f||_ ttj |f||_ n| dd| dd| dS)NrCrz,num_channels must be divisible by num_groupsrFrB)rr ValueErrorrUrVr?rWrrKrLrFrBrMrN)r!rUrVr?rWrDrErOr"r$r%rs     zGroupNorm.__init__r<cCs"|jrt|jt|jdSr)rWrrPrFrQrBr/r$r$r%rNs zGroupNorm.reset_parametersr&cCst||j|j|j|jSr)r(Z group_normrUrFrBr?r)r$r$r%r*szGroupNorm.forwardcCsdjf|jS)Nz8{num_groups}, {num_channels}, eps={eps}, affine={affine}r,r/r$r$r%r0#szGroupNorm.extra_repr)rATNN)r1r2r3r4r5r6r7r8rSrrNr r*r=r0r9r$r$r"r%rs -  cseZdZUdZdddgZeedfed<ee ed<e ed<de ee e ddfd d Z dd d d Z ejejdddZed ddZZS)ra}Applies Root Mean Square Layer Normalization over a mini-batch of inputs. This layer implements the operation as described in the paper `Root Mean Square Layer Normalization `__ .. math:: y = \frac{x}{\sqrt{\mathrm{RMS}[x] + \epsilon}} * \gamma The root mean squared norm is taken over the last ``D`` dimensions, where ``D`` is the dimension of :attr:`normalized_shape`. For example, if :attr:`normalized_shape` is ``(3, 5)`` (a 2-dimensional shape), the rms norm is computed over the last 2 dimensions of the input. Args: normalized_shape (int or list or torch.Size): input shape from an expected input of size .. math:: [* \times \text{normalized\_shape}[0] \times \text{normalized\_shape}[1] \times \ldots \times \text{normalized\_shape}[-1]] If a single integer is used, it is treated as a singleton list, and this module will normalize over the last dimension which is expected to be of that specific size. eps: a value added to the denominator for numerical stability. Default: :func:`torch.finfo(x.dtype).eps` elementwise_affine: a boolean value that when set to ``True``, this module has learnable per-element affine parameters initialized to ones (for weights) and zeros (for biases). Default: ``True``. Shape: - Input: :math:`(N, *)` - Output: :math:`(N, *)` (same shape as input) Examples:: >>> rms_norm = nn.RMSNorm([2, 3]) >>> input = torch.randn(2, 2, 3) >>> rms_norm(input) r>r?r@.NT)r>r?r@rcsr||d}tt|tjr&|f}t||_||_||_|jrZt t j |jf||_ n | dd|dS)NrCrF)rrrGrHrIrJr>r?r@rrKrLrFrMrN)r!r>r?r@rDrErOr"r$r%rUs     zRMSNorm.__init__r<cCs|jrt|jdS)zS Resets parameters based on their initialization used in __init__. N)r@rrPrFr/r$r$r%rNeszRMSNorm.reset_parameters)xrcCst||j|j|jS)z$ Runs forward pass. )r(Zrms_normr>rFr?)r!rYr$r$r%r*lszRMSNorm.forwardcCsdjf|jS)z5 Extra information about the module. rRr,r/r$r$r%r0rszRMSNorm.extra_repr)NTNN)r1r2r3r4r5rr6r7r r8rSrTrrNrKr r*r=r0r9r$r$r"r%r(s '   )rKrHZtorch.nn.parameterrmodulerZ _functionsrr:rr(rr r typingr r r r__all__rr6rTrrrr$r$r$r%s     3xV