Commit e89329d6 authored by Myle Ott's avatar Myle Ott
Browse files

Updates for latest PyTorch

parent ff68a9ef

fairseq/criterions/adaptive_loss.py

+1 −1
Changes for fairseq/criterions/adaptive_loss.py: 1 added line, 1 removed line.
Original line number Diff line number Diff line
@@ -26,7 +26,7 @@ class AdaptiveLoss(FairseqCriterion): 
        """Compute the loss for the given sample.



        Returns a tuple with three elements:

        1) the loss, as a Variable

        1) the loss

        2) the sample size, which is used as the denominator for the gradient

        3) logging outputs to display while training

        """

fairseq/criterions/cross_entropy.py

+1 −1
Changes for fairseq/criterions/cross_entropy.py: 1 added line, 1 removed line.
Original line number Diff line number Diff line
@@ -23,7 +23,7 @@ class CrossEntropyCriterion(FairseqCriterion): 
        """Compute the loss for the given sample.



        Returns a tuple with three elements:

        1) the loss, as a Variable

        1) the loss

        2) the sample size, which is used as the denominator for the gradient

        3) logging outputs to display while training

        """

fairseq/criterions/fairseq_criterion.py

+1 −1
Changes for fairseq/criterions/fairseq_criterion.py: 1 added line, 1 removed line.
Original line number Diff line number Diff line
@@ -24,7 +24,7 @@ class FairseqCriterion(_Loss): 
        """Compute the loss for the given sample.



        Returns a tuple with three elements:

        1) the loss, as a Variable

        1) the loss

        2) the sample size, which is used as the denominator for the gradient

        3) logging outputs to display while training

        """

fairseq/criterions/label_smoothed_cross_entropy.py

+1 −1
Changes for fairseq/criterions/label_smoothed_cross_entropy.py: 1 added line, 1 removed line.
Original line number Diff line number Diff line
@@ -29,7 +29,7 @@ class LabelSmoothedCrossEntropyCriterion(FairseqCriterion): 
        """Compute the loss for the given sample.



        Returns a tuple with three elements:

        1) the loss, as a Variable

        1) the loss

        2) the sample size, which is used as the denominator for the gradient

        3) logging outputs to display while training

        """

fairseq/models/fconv.py

+10 −10
Changes for fairseq/models/fconv.py: 10 added lines, 10 removed lines.
Original line number Diff line number Diff line
@@ -565,23 +565,23 @@ def extend_conv_spec(convolutions): 


def Embedding(num_embeddings, embedding_dim, padding_idx):

    m = nn.Embedding(num_embeddings, embedding_dim, padding_idx=padding_idx)

    nn.init.normal(m.weight, 0, 0.1)

    nn.init.constant(m.weight[padding_idx], 0)

    nn.init.normal_(m.weight, 0, 0.1)

    nn.init.constant_(m.weight[padding_idx], 0)

    return m





def PositionalEmbedding(num_embeddings, embedding_dim, padding_idx, left_pad):

    m = LearnedPositionalEmbedding(num_embeddings, embedding_dim, padding_idx, left_pad)

    nn.init.normal(m.weight, 0, 0.1)

    nn.init.constant(m.weight[padding_idx], 0)

    nn.init.normal_(m.weight, 0, 0.1)

    nn.init.constant_(m.weight[padding_idx], 0)

    return m





def Linear(in_features, out_features, dropout=0):

    """Weight-normalized Linear layer (input: N x T x C)"""

    m = nn.Linear(in_features, out_features)

    nn.init.normal(m.weight, mean=0, std=math.sqrt((1 - dropout) / in_features))

    nn.init.constant(m.bias, 0)

    nn.init.normal_(m.weight, mean=0, std=math.sqrt((1 - dropout) / in_features))

    nn.init.constant_(m.bias, 0)

    return nn.utils.weight_norm(m)
@@ -589,8 +589,8 @@ def LinearizedConv1d(in_channels, out_channels, kernel_size, dropout=0, **kwargs 
    """Weight-normalized Conv1d layer optimized for decoding"""

    m = LinearizedConvolution(in_channels, out_channels, kernel_size, **kwargs)

    std = math.sqrt((4 * (1.0 - dropout)) / (m.kernel_size[0] * in_channels))

    nn.init.normal(m.weight, mean=0, std=std)

    nn.init.constant(m.bias, 0)

    nn.init.normal_(m.weight, mean=0, std=std)

    nn.init.constant_(m.bias, 0)

    return nn.utils.weight_norm(m, dim=2)
@@ -599,8 +599,8 @@ def ConvTBC(in_channels, out_channels, kernel_size, dropout=0, **kwargs): 
    from fairseq.modules import ConvTBC

    m = ConvTBC(in_channels, out_channels, kernel_size, **kwargs)

    std = math.sqrt((4 * (1.0 - dropout)) / (m.kernel_size[0] * in_channels))

    nn.init.normal(m.weight, mean=0, std=std)

    nn.init.constant(m.bias, 0)

    nn.init.normal_(m.weight, mean=0, std=std)

    nn.init.constant_(m.bias, 0)

    return nn.utils.weight_norm(m, dim=2)