Python modules.ConvTBC方法代码示例

# 需要导入模块: from fairseq import modules [as 别名]
# 或者: from fairseq.modules import ConvTBC [as 别名]
def test_convtbc(self):
        # ksz, in_channels, out_channels
        conv_tbc = ConvTBC(4, 5, kernel_size=3, padding=1)
        # out_channels, in_channels, ksz
        conv1d = nn.Conv1d(4, 5, kernel_size=3, padding=1)

        conv_tbc.weight.data.copy_(conv1d.weight.data.transpose(0, 2))
        conv_tbc.bias.data.copy_(conv1d.bias.data)

        input_tbc = Variable(torch.randn(7, 2, 4), requires_grad=True)
        input1d = Variable(input_tbc.data.transpose(0, 1).transpose(1, 2), requires_grad=True)

        output_tbc = conv_tbc(input_tbc)
        output1d = conv1d(input1d)

        self.assertAlmostEqual(output_tbc.data.transpose(0, 1).transpose(1, 2), output1d.data)

        grad_tbc = torch.randn(output_tbc.size())
        grad1d = grad_tbc.transpose(0, 1).transpose(1, 2).contiguous()

        output_tbc.backward(grad_tbc)
        output1d.backward(grad1d)

        self.assertAlmostEqual(conv_tbc.weight.grad.data.transpose(0, 2), conv1d.weight.grad.data)
        self.assertAlmostEqual(conv_tbc.bias.grad.data, conv1d.bias.grad.data)
        self.assertAlmostEqual(input_tbc.grad.data.transpose(0, 1).transpose(1, 2), input1d.grad.data) 

# 需要导入模块: from fairseq import modules [as 别名]
# 或者: from fairseq.modules import ConvTBC [as 别名]
def ConvTBC(in_channels, out_channels, kernel_size, dropout=0, **kwargs):
    """Weight-normalized Conv1d layer"""
    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)
    return nn.utils.weight_norm(m, dim=2) 

# 需要导入模块: from fairseq import modules [as 别名]
# 或者: from fairseq.modules import ConvTBC [as 别名]
def ConvTBC(in_channels, out_channels, kernel_size, dropout=0, **kwargs):
    """Weight-normalized Conv1d layer"""
    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))
    m.weight.data.normal_(mean=0, std=std)
    m.bias.data.zero_()
    return m 

# 需要导入模块: from fairseq import modules [as 别名]
# 或者: from fairseq.modules import ConvTBC [as 别名]
def test_convtbc(self):
        # ksz, in_channels, out_channels
        conv_tbc = ConvTBC(4, 5, kernel_size=3, padding=1)
        # out_channels, in_channels, ksz
        conv1d = nn.Conv1d(4, 5, kernel_size=3, padding=1)

        conv_tbc.weight.data.copy_(conv1d.weight.data.transpose(0, 2))
        conv_tbc.bias.data.copy_(conv1d.bias.data)

        input_tbc = torch.randn(7, 2, 4, requires_grad=True)
        input1d = input_tbc.data.transpose(0, 1).transpose(1, 2)
        input1d.requires_grad = True

        output_tbc = conv_tbc(input_tbc)
        output1d = conv1d(input1d)

        self.assertAlmostEqual(output_tbc.data.transpose(0, 1).transpose(1, 2), output1d.data)

        grad_tbc = torch.randn(output_tbc.size())
        grad1d = grad_tbc.transpose(0, 1).transpose(1, 2).contiguous()

        output_tbc.backward(grad_tbc)
        output1d.backward(grad1d)

        self.assertAlmostEqual(conv_tbc.weight.grad.data.transpose(0, 2), conv1d.weight.grad.data)
        self.assertAlmostEqual(conv_tbc.bias.grad.data, conv1d.bias.grad.data)
        self.assertAlmostEqual(input_tbc.grad.data.transpose(0, 1).transpose(1, 2), input1d.grad.data) 

# 需要导入模块: from fairseq import modules [as 别名]
# 或者: from fairseq.modules import ConvTBC [as 别名]
def __init__(self, dictionary, embed_dim=512, max_positions=1024,
                 convolutions=((512, 3),) * 20, dropout=0.1):
        super().__init__(dictionary)
        self.dropout = dropout
        self.num_attention_layers = None

        num_embeddings = len(dictionary)
        padding_idx = dictionary.pad()
        self.embed_tokens = Embedding(num_embeddings, embed_dim, padding_idx)
        self.embed_positions = PositionalEmbedding(
            max_positions,
            embed_dim,
            padding_idx,
            left_pad=LanguagePairDataset.LEFT_PAD_SOURCE,
        )

        in_channels = convolutions[0][0]
        # Shashi
        self.fc1 = Linear(embed_dim+embed_dim, in_channels, dropout=dropout)
        self.projections = nn.ModuleList()
        self.convolutions = nn.ModuleList()
        for (out_channels, kernel_size) in convolutions:
            self.projections.append(Linear(in_channels, out_channels)
                                    if in_channels != out_channels else None)
            self.convolutions.append(
                ConvTBC(in_channels, out_channels * 2, kernel_size,
                        dropout=dropout)
            )
            in_channels = out_channels
        self.fc2 = Linear(in_channels, embed_dim+embed_dim) 

# 需要导入模块: from fairseq import modules [as 别名]
# 或者: from fairseq.modules import ConvTBC [as 别名]
def ConvTBC(in_channels, out_channels, kernel_size, dropout=0, **kwargs):
    """Weight-normalized Conv1d layer"""
    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))
    m.weight.data.normal_(mean=0, std=std)
    m.bias.data.zero_()
    return nn.utils.weight_norm(m, dim=2) 

# 需要导入模块: from fairseq import modules [as 别名]
# 或者: from fairseq.modules import ConvTBC [as 别名]
def __init__(self, dictionary, embed_dim=512, max_positions=1024,
                 convolutions=((512, 3),) * 20, dropout=0.1):
        super().__init__(dictionary)
        self.dropout = dropout
        self.num_attention_layers = None

        num_embeddings = len(dictionary)
        padding_idx = dictionary.pad()
        self.embed_tokens = Embedding(num_embeddings, embed_dim, padding_idx)
        self.embed_positions = PositionalEmbedding(
            max_positions,
            embed_dim,
            padding_idx,
            left_pad=LanguagePairDataset.LEFT_PAD_SOURCE,
        )

        in_channels = convolutions[0][0]
        self.fc1 = Linear(embed_dim, in_channels, dropout=dropout)
        self.projections = nn.ModuleList()
        self.convolutions = nn.ModuleList()
        for (out_channels, kernel_size) in convolutions:
            self.projections.append(Linear(in_channels, out_channels)
                                    if in_channels != out_channels else None)
            self.convolutions.append(
                ConvTBC(in_channels, out_channels * 2, kernel_size,
                        dropout=dropout)
            )
            in_channels = out_channels
        self.fc2 = Linear(in_channels, embed_dim) 

# 需要导入模块: from fairseq import modules [as 别名]
# 或者: from fairseq.modules import ConvTBC [as 别名]
def __init__(
        self, dictionary, embed_dim=512, embed_dict=None, max_positions=1024,
        convolutions=((512, 3),) * 20, dropout=0.1, normalization_constant=0.5,
        left_pad=True,
    ):
        super().__init__(dictionary)
        self.dropout = dropout
        self.normalization_constant = normalization_constant
        self.left_pad = left_pad
        self.num_attention_layers = None

        num_embeddings = len(dictionary)
        self.padding_idx = dictionary.pad()
        self.embed_tokens = Embedding(num_embeddings, embed_dim, self.padding_idx)
        if embed_dict:
            self.embed_tokens = utils.load_embedding(embed_dict, self.dictionary, self.embed_tokens)

        self.embed_positions = PositionalEmbedding(
            max_positions,
            embed_dim,
            self.padding_idx,
            left_pad=self.left_pad,
        )

        convolutions = extend_conv_spec(convolutions)
        in_channels = convolutions[0][0]
        self.fc1 = Linear(embed_dim, in_channels, dropout=dropout)
        self.projections = nn.ModuleList()
        self.convolutions = nn.ModuleList()
        self.residuals = []

        layer_in_channels = [in_channels]
        for i, (out_channels, kernel_size, residual) in enumerate(convolutions):
            if residual == 0:
                residual_dim = out_channels
            else:
                residual_dim = layer_in_channels[-residual]
            self.projections.append(Linear(residual_dim, out_channels)
                                    if residual_dim != out_channels else None)
            if kernel_size % 2 == 1:
                padding = kernel_size // 2
            else:
                padding = 0
            self.convolutions.append(
                ConvTBC(in_channels, out_channels * 2, kernel_size,
                        dropout=dropout, padding=padding)
            )
            self.residuals.append(residual)
            in_channels = out_channels
            layer_in_channels.append(out_channels)
        self.fc2 = Linear(in_channels, embed_dim) 

# 需要导入模块: from fairseq import modules [as 别名]
# 或者: from fairseq.modules import ConvTBC [as 别名]
def __init__(
        self, dictionary, embed_dim=512, embed_dict=None, max_positions=1024,
        convolutions=((512, 3),) * 20, dropout=0.1, normalization_constant=0.5,
        left_pad=True, token_dropout=0.0
    ):
        super().__init__(dictionary)
        self.dropout = dropout
        self.token_dropout = token_dropout
        self.normalization_constant = normalization_constant
        self.left_pad = left_pad
        self.num_attention_layers = None

        num_embeddings = len(dictionary)
        self.padding_idx = dictionary.pad()
        self.embed_tokens = Embedding(num_embeddings, embed_dim, self.padding_idx)
        if embed_dict:
            self.embed_tokens = utils.load_embedding(embed_dict, self.dictionary, self.embed_tokens)

        self.embed_positions = PositionalEmbedding(
            max_positions,
            embed_dim,
            self.padding_idx,
            left_pad=self.left_pad,
        )

        convolutions = extend_conv_spec(convolutions)
        in_channels = convolutions[0][0]
        self.fc1 = Linear(embed_dim, in_channels, dropout=dropout)
        self.projections = nn.ModuleList()
        self.convolutions = nn.ModuleList()
        self.residuals = []

        layer_in_channels = [in_channels]
        for i, (out_channels, kernel_size, residual) in enumerate(convolutions):
            if residual == 0:
                residual_dim = out_channels
            else:
                residual_dim = layer_in_channels[-residual]
            self.projections.append(Linear(residual_dim, out_channels)
                                    if residual_dim != out_channels else None)
            if kernel_size % 2 == 1:
                padding = kernel_size // 2
            else:
                padding = 0
            self.convolutions.append(
                ConvTBC(in_channels, out_channels * 2, kernel_size,
                        dropout=dropout, padding=padding)
            )
            self.residuals.append(residual)
            in_channels = out_channels
            layer_in_channels.append(out_channels)
        self.fc2 = Linear(in_channels, embed_dim) 

# 需要导入模块: from fairseq import modules [as 别名]
# 或者: from fairseq.modules import ConvTBC [as 别名]
def __init__(
        self, dictionary, embed_dim=512, embed_dict=None, max_positions=1024,
        convolutions=((512, 3),) * 20, dropout=0.1,
    ):
        super().__init__(dictionary)
        self.dropout = dropout
        self.num_attention_layers = None

        num_embeddings = len(dictionary)
        self.padding_idx = dictionary.pad()
        self.embed_tokens = Embedding(num_embeddings, embed_dim, self.padding_idx)
        if embed_dict:
            self.embed_tokens = utils.load_embedding(embed_dict, self.dictionary, self.embed_tokens)

        self.embed_positions = PositionalEmbedding(
            max_positions,
            embed_dim,
            self.padding_idx,
        )

        convolutions = extend_conv_spec(convolutions)
        in_channels = convolutions[0][0]
        self.fc1 = Linear(embed_dim, in_channels, dropout=dropout)
        self.projections = nn.ModuleList()
        self.convolutions = nn.ModuleList()
        self.residuals = []

        layer_in_channels = [in_channels]
        for _, (out_channels, kernel_size, residual) in enumerate(convolutions):
            if residual == 0:
                residual_dim = out_channels
            else:
                residual_dim = layer_in_channels[-residual]
            self.projections.append(Linear(residual_dim, out_channels)
                                    if residual_dim != out_channels else None)
            if kernel_size % 2 == 1:
                padding = kernel_size // 2
            else:
                padding = 0
            self.convolutions.append(
                ConvTBC(in_channels, out_channels * 2, kernel_size,
                        dropout=dropout, padding=padding)
            )
            self.residuals.append(residual)
            in_channels = out_channels
            layer_in_channels.append(out_channels)
        self.fc2 = Linear(in_channels, embed_dim) 

# 需要导入模块: from fairseq import modules [as 别名]
# 或者: from fairseq.modules import ConvTBC [as 别名]
def __init__(
        self, dictionary, embed_dim=512, max_positions=1024,
        convolutions=((512, 3),) * 20, dropout=0.1, attention=False,
        attention_nheads=1,
    ):
        super().__init__(dictionary)
        self.dropout = dropout
        self.num_attention_layers = None

        num_embeddings = len(dictionary)
        self.padding_idx = dictionary.pad()
        self.embed_tokens = Embedding(num_embeddings, embed_dim, self.padding_idx)
        self.embed_positions = PositionalEmbedding(
            max_positions,
            embed_dim,
            self.padding_idx,
        )

        def expand_bool_array(val):
            if isinstance(val, bool):
                # expand True into [True, True, ...] and do the same with False
                return [val] * len(convolutions)
            return val

        attention = expand_bool_array(attention)

        in_channels = convolutions[0][0]
        self.fc1 = Linear(embed_dim, in_channels, dropout=dropout)
        self.projections = nn.ModuleList()
        self.convolutions = nn.ModuleList()
        self.attention = nn.ModuleList()
        self.attproj = nn.ModuleList()
        for i, (out_channels, kernel_size) in enumerate(convolutions):
            self.projections.append(
                Linear(in_channels, out_channels) if in_channels != out_channels else None
            )
            self.convolutions.append(
                ConvTBC(in_channels, out_channels * 2, kernel_size, dropout=dropout)
            )

            self.attention.append(
                SelfAttention(out_channels, embed_dim, attention_nheads) if attention[i] else None
            )
            in_channels = out_channels

        self.fc2 = Linear(in_channels, embed_dim) 

# 需要导入模块: from fairseq import modules [as 别名]
# 或者: from fairseq.modules import ConvTBC [as 别名]
def __init__(
        self, dictionary, embed_dim=512, max_positions=1024,
        convolutions=((512, 3),) * 20, dropout=0.1, attention=False,
        attention_nheads=1, left_pad=True,
    ):
        super().__init__(dictionary)
        self.dropout = dropout
        self.num_attention_layers = None
        self.left_pad = left_pad

        num_embeddings = len(dictionary)
        self.padding_idx = dictionary.pad()
        self.embed_tokens = Embedding(num_embeddings, embed_dim, self.padding_idx)
        self.embed_positions = PositionalEmbedding(
            max_positions,
            embed_dim,
            self.padding_idx,
            left_pad=self.left_pad,
        )

        def expand_bool_array(val):
            if isinstance(val, bool):
                # expand True into [True, True, ...] and do the same with False
                return [val] * len(convolutions)
            return val

        attention = expand_bool_array(attention)

        in_channels = convolutions[0][0]
        self.fc1 = Linear(embed_dim, in_channels, dropout=dropout)
        self.projections = nn.ModuleList()
        self.convolutions = nn.ModuleList()
        self.attention = nn.ModuleList()
        self.attproj = nn.ModuleList()
        for i, (out_channels, kernel_size) in enumerate(convolutions):
            self.projections.append(
                Linear(in_channels, out_channels) if in_channels != out_channels else None
            )
            self.convolutions.append(
                ConvTBC(in_channels, out_channels * 2, kernel_size, dropout=dropout)
            )

            self.attention.append(
                SelfAttention(out_channels, embed_dim, attention_nheads) if attention[i] else None
            )
            in_channels = out_channels

        self.fc2 = Linear(in_channels, embed_dim)