Using positional encoding in pytorch

Question

Is there any built-in positional encoding in pytorch? Basically, I want to be able to specify the dimension of the encoding, and then be able to get the i'th encoding for every i.

Answer 1

There isn't, as far as I'm aware.

However, you can use an implementation from PyTorch's documentation:

class PositionalEncoding(nn.Module):

    def __init__(self, d_model: int, dropout: float = 0.1, max_len: int = 5000):
        super().__init__()
        self.dropout = nn.Dropout(p=dropout)

        position = torch.arange(max_len).unsqueeze(1)
        div_term = torch.exp(torch.arange(0, d_model, 2) * (-math.log(10000.0) / d_model))
        pe = torch.zeros(max_len, 1, d_model)
        pe[:, 0, 0::2] = torch.sin(position * div_term)
        pe[:, 0, 1::2] = torch.cos(position * div_term)
        self.register_buffer('pe', pe)

    def forward(self, x: Tensor) -> Tensor:
        """
        Arguments:
            x: Tensor, shape ``[seq_len, batch_size, embedding_dim]``
        """
        x = x + self.pe[:x.size(0)]
        return self.dropout(x)

You can find it here.

Answer 2

There is now an implementation of RoPE at torchtune.modules.RotaryPositionalEmbeddings; torchtune is authored by the official pytorch team.