多头自注意力机制

import torch
import torch.nn.functional as F


class SelfAttention(torch.nn.Module):
    def __init__(self, embed_size, heads):
        super(SelfAttention, self).__init__()
        self.embed_size = embed_size
        self.heads = heads
        self.head_dim = embed_size // heads

        assert (
                self.head_dim * heads == embed_size
        ), "Embedding size needs to be divisible by heads"

        self.values = torch.nn.Linear(self.head_dim, self.head_dim, bias=False)
        self.keys = torch.nn.Linear(self.head_dim, self.head_dim, bias=False)
        self.queries = torch.nn.Linear(self.head_dim, self.head_dim, bias=False)
        self.fc_out = torch.nn.Linear(heads * self.head_dim, embed_size)

    def forward(self, values, keys, query, mask):
        N = query.shape[0]
        value_len, key_len, query_len = values.shape[1], keys.shape[1], query.shape[1]

        # Split the embedding into self.heads pieces
        values = values.reshape(N, value_len, self.heads, self.head_dim)
        keys = keys.reshape(N, key_len, self.heads, self.head_dim)
        query = query.reshape(N, query_len, self.heads, self.head_dim)

        values = self.values(values)
        keys = self.keys(keys)
        queries = self.queries(query)

        energy = torch.einsum("nqhd,nkhd->nhqk", [queries, keys])

        if mask is not None:
            energy = energy.masked_fill(mask == 0, float("-1e20"))

        attention = F.softmax(energy / (self.embed_size ** (1 / 2)), dim=3)

        out = torch.einsum("nhql,nlhd->nqhd", [attention, values]).reshape(N, query_len, self.heads * self.head_dim)

        out = self.fc_out(out)

        return out