o wZhm:@sddlZddlmZmZddlZddlmZmZddlmZ m Z ddl m Z ddl mZddlmZgd ZGd d d eZGd d d eZeeeeefZGdddeZGdddeZGdddeZdS)N)OptionalUnion)SizeTensor) functionalinit) Parameter) CrossMapLRN2d)Module)LocalResponseNormr LayerNorm GroupNormRMSNormc s~eZdZUdZgdZeed<eed<eed<eed< ddedededed d f fd d Zde d e fddZ ddZ Z S)r aApplies 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) )sizealphabetakrrrr-C6???returnNc&t||_||_||_||_dSNsuper__init__rrrrselfrrrr __class__M/var/www/auris/lib/python3.10/site-packages/torch/nn/modules/normalization.pyr5  zLocalResponseNorm.__init__inputcCt||j|j|j|jSr)FZlocal_response_normrrrrrr$r!r!r"forward>zLocalResponseNorm.forwardcCdjdi|jSNz){size}, alpha={alpha}, beta={beta}, k={k}r!format__dict__rr!r!r" extra_reprAzLocalResponseNorm.extra_repr)rrr) __name__ __module__ __qualname____doc__ __constants__int__annotations__floatrrr(r0 __classcell__r!r!rr"r s*  r c sxeZdZUeed<eed<eed<eed< ddededededd f fd d Zd edefd dZde fddZ Z S)r rrrrrrr rNcrrrrrr!r"rKr#zCrossMapLRN2d.__init__r$cCr%r)_cross_map_lrn2dapplyrrrrr'r!r!r"r(Tr)zCrossMapLRN2d.forwardcCr*r+r,r/r!r!r"r0Wr1zCrossMapLRN2d.extra_repr)rrr ) r2r3r4r7r8r9rrr(strr0r:r!r!rr"r Es&  r c seZdZUdZgdZeedfed<eed<e ed<    dde dede d e d d f fd d Z dddZ de d e fddZd efddZZS)r a 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 variance 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.r?r@rAh㈵>TNbiasrcs||d}tt|tjr|f}t||_||_||_|jrEt t j |jfi||_ |r>t t j |jfi||_ n|ddn |dd|dd|dS)NdevicedtyperCweight)rr isinstancenumbersIntegraltupler?r@rArtorchemptyrGrCregister_parameterreset_parameters)rr?r@rArCrErFfactory_kwargsrr!r"rs&      zLayerNorm.__init__cCs4|jrt|j|jdurt|jdSdSdSr)rArones_rGrCzeros_r/r!r!r"rOs   zLayerNorm.reset_parametersr$cCr%r)r&Z layer_normr?rGrCr@r'r!r!r"r(szLayerNorm.forwardcCr*)NF{normalized_shape}, eps={eps}, elementwise_affine={elementwise_affine}r!r,r/r!r!r"r0s zLayerNorm.extra_repr)rBTTNNrN)r2r3r4r5r6rKr7r8r9bool_shape_trrOrr(r=r0r:r!r!rr"r ^s2 M !r c seZdZUdZgdZeed<eed<eed<eed<   ddedededed d f fd d Z dd dZ de d e fddZ d e fddZZS)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 variance 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@affinerWrXr@rYrBTNrcs||d}t||dkrtd||_||_||_||_|jr`__ .. math:: y_i = \frac{x_i}{\mathrm{RMS}(x)} * \gamma_i, \quad \text{where} \quad \text{RMS}(x) = \sqrt{\epsilon + \frac{1}{n} \sum_{i=1}^{n} x_i^2} The RMS 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 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). 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@rANTrcsv||d}tt|tjr|f}t||_||_||_|jr/t t j |jfi||_ n| dd|dS)NrDrG)rrrHrIrJrKr?r@rArrLrMrGrNrO)rr?r@rArErFrPrr!r"rns      zRMSNorm.__init__cCs|jr t|jdSdS)zS Resets parameters based on their initialization used in __init__. N)rArrQrGr/r!r!r"rOszRMSNorm.reset_parametersxcCst||j|j|jS)z$ Runs forward pass. )r&Zrms_normr?rGr@)rr[r!r!r"r(szRMSNorm.forwardcCsdjdi|jS)z5 Extra information about the module. rSNr!r,r/r!r!r"r0s zRMSNorm.extra_repr)NTNNrT)r2r3r4r5r6rKr7r8rr9rUrVrrOrLrr(r=r0r:r!r!rr"rAs, '  r)rItypingrrrLrrZtorch.nnrr&rZtorch.nn.parameterrZ _functionsr r;moduler __all__r r7listrVr rrr!r!r!r"s    4]