Skip to content

Natural Language Processing

tensoris.backend.components.layers.nlp

Natural Language Processing Positional Encoding Layers.

Classes

PositionalEncoding

Bases: Module

Sinusoidal Positional Encoding Layer for Sequence Transformers.

References

Vaswani, A. et al. (2017). Attention Is All You Need. NIPS 2017. arXiv: https://arxiv.org/abs/1706.03762

Source code in src/tensoris/backend/components/layers/nlp.py 
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
class PositionalEncoding(nn.Module):
    """Sinusoidal Positional Encoding Layer for Sequence Transformers.

    References:
        Vaswani, A. et al. (2017). Attention Is All You Need. NIPS 2017.
        arXiv: https://arxiv.org/abs/1706.03762
    """

    def __init__(self, d_model: int, max_len: int = 5000) -> None:
        """Initialize PositionalEncoding layer.

        Args:
            d_model: Token embedding vector dimension.
            max_len: Maximum sequence length.
        """
        super().__init__()
        pe = torch.zeros(max_len, d_model)
        position = torch.arange(0, max_len, dtype=torch.float).unsqueeze(1)
        div_term = torch.exp(
            torch.arange(0, d_model, 2).float() * (-math.log(10000.0) / d_model)
        )

        pe[:, 0::2] = torch.sin(position * div_term)
        pe[:, 1::2] = torch.cos(position * div_term)
        self.register_buffer("pe", pe.unsqueeze(0))

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        """Add positional encodings to token embeddings."""
        return x + self.pe[:, : x.size(1)]
Methods:
__init__ 
__init__(d_model, max_len=5000)

Initialize PositionalEncoding layer.

Parameters:

Name Type Description Default
d_model int

Token embedding vector dimension.

 required
max_len int

Maximum sequence length.

 5000
Source code in src/tensoris/backend/components/layers/nlp.py 
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
def __init__(self, d_model: int, max_len: int = 5000) -> None:
    """Initialize PositionalEncoding layer.

    Args:
        d_model: Token embedding vector dimension.
        max_len: Maximum sequence length.
    """
    super().__init__()
    pe = torch.zeros(max_len, d_model)
    position = torch.arange(0, max_len, dtype=torch.float).unsqueeze(1)
    div_term = torch.exp(
        torch.arange(0, d_model, 2).float() * (-math.log(10000.0) / d_model)
    )

    pe[:, 0::2] = torch.sin(position * div_term)
    pe[:, 1::2] = torch.cos(position * div_term)
    self.register_buffer("pe", pe.unsqueeze(0))
forward 
forward(x)

Add positional encodings to token embeddings.

Source code in src/tensoris/backend/components/layers/nlp.py 
35
36
37
def forward(self, x: torch.Tensor) -> torch.Tensor:
    """Add positional encodings to token embeddings."""
    return x + self.pe[:, : x.size(1)]