PE.py

# code from https://github.com/facebookresearch/fairseq/blob/main/fairseq/modules/sinusoidal_positional_embedding.py (MIT License) - some modifications

# Copyright (c) Facebook, Inc. and its affiliates.
#
# This source code is licensed under the MIT license found in the
# LICENSE file in the root directory of this source tree.

import torch
import torch.nn as nn
import math

def make_positions(tensor, padding_idx):
    """Replace non-padding symbols with their position numbers.
    Position numbers begin at padding_idx+1. Padding symbols are ignored.
    """
    # The series of casts and type-conversions here are carefully
    # balanced to both work with ONNX export and XLA. In particular XLA
    # prefers ints, cumsum defaults to output longs, and ONNX doesn't know
    # how to handle the dtype kwarg in cumsum.
    mask = tensor.ne(padding_idx).int()
    return (torch.cumsum(mask, dim=1).type_as(mask) * mask).long() + padding_idx

class SinusoidalPositionalEmbedding(nn.Module):
    """This module produces sinusoidal positional embeddings of any length.
    Padding symbols are ignored.
    """
    def __init__(self, embedding_dim, padding_idx, init_size=1568):
        super().__init__()
        self.embedding_dim = embedding_dim
        self.padding_idx = padding_idx
        self.weights = SinusoidalPositionalEmbedding.get_embedding(
            init_size,
            embedding_dim,
            padding_idx,
        )
        self.register_buffer('_float_tensor', torch.FloatTensor(1))

    @staticmethod
    def get_embedding(num_embeddings, embedding_dim, padding_idx=None):
        """Build sinusoidal embeddings.
        This matches the implementation in tensor2tensor, but differs slightly
        from the description in Section 3.5 of "Attention Is All You Need".
        """
        half_dim = embedding_dim // 2
        emb = math.log(10000) / (half_dim - 1)
        emb = torch.exp(torch.arange(half_dim, dtype=torch.float) * -emb)
        emb = torch.arange(num_embeddings, dtype=torch.float).unsqueeze(1) * emb.unsqueeze(0)
        emb = torch.cat([torch.sin(emb), torch.cos(emb)], dim=1).view(num_embeddings, -1)
        if embedding_dim % 2 == 1:
            # zero pad
            emb = torch.cat([emb, torch.zeros(num_embeddings, 1)], dim=1)
        if padding_idx is not None:
            emb[padding_idx, :] = 0
        return emb

    def forward(self, input):
        """Input is expected to be of size [bsz x seqlen]."""
        bsz, seq_len = input.size()
        max_pos = self.padding_idx + 1 + seq_len
        if max_pos > self.weights.size(0):
            # recompute/expand embeddings if needed
            self.weights = SinusoidalPositionalEmbedding.get_embedding(
                max_pos,
                self.embedding_dim,
                self.padding_idx,
            )
        self.weights = self.weights.to(self._float_tensor)

        positions = make_positions(input, self.padding_idx)
        return self.weights.index_select(0, positions.view(-1)).view(bsz, seq_len, -1).detach()

    def max_positions(self):
        """Maximum number of supported positions."""
        return int(1e5)  # an arbitrary large number