从零实现PositionalEncoding与MultiHeadAttention

位置编码

#@save
class PositionalEncoding(nn.Module):
    """位置编码"""
    def __init__(self, num_hiddens, dropout, max_len=1000):
        super(PositionalEncoding, self).__init__()
        self.dropout = nn.Dropout(dropout)
        # 创建一个足够长的P
        self.P = torch.zeros((1, max_len, num_hiddens))
        X = torch.arange(max_len, dtype=torch.float32).reshape(
            -1, 1) / torch.pow(10000, torch.arange(
            0, num_hiddens, 2, dtype=torch.float32) / num_hiddens)
        self.P[:, :, 0::2] = torch.sin(X)
        self.P[:, :, 1::2] = torch.cos(X)

    def forward(self, X):
        X = X + self.P[:, :X.shape[1], :].to(X.device)
        return self.dropout(X)

掩蔽softmax操作

#@save
def masked_softmax(X, valid_lens):
    """通过在最后一个轴上掩蔽元素来执行softmax操作"""
    # X:3D张量，valid_lens:1D或2D张量
    if valid_lens is None:
        return nn.functional.softmax(X, dim=-1)
    else:
        shape = X.shape
        if valid_lens.dim() == 1:
            valid_lens = torch.repeat_interleave(valid_lens, shape[1])
        else:
            valid_lens = valid_lens.reshape(-1)
        # 最后一轴上被掩蔽的元素使用一个非常大的负值替换，从而其softmax输出为0
        X = sequence_mask(X.reshape(-1, shape[-1]), valid_lens,
                              value=-1e6)
        return nn.functional.softmax(X.reshape(shape), dim=-1)

def sequence_mask(X, valid_lens, value=-1e6):
    """
    将序列中超出有效长度的位置替换为掩码值
    
    参数:
    X - 输入张量, 形状为(batch_size * num_entries, seq_len)
    valid_lens - 每个序列的有效长度, 形状为(batch_size * num_entries, )
    value - 掩码填充值, 默认-1e6
    
    返回:
    掩码处理后的张量, 形状与X相同
    """
    # 获取序列最大长度
    max_len = X.size(1)
    
    # 创建位置索引矩阵[0, 1, ..., max_len-1]
    mask = torch.arange(max_len, device=X.device)[None, :] 
    
    # 将有效长度转换为列向量
    valid_lens = valid_lens[:, None]
    
    # 生成布尔掩码 (True表示需要掩码的位置)
    mask = mask >= valid_lens
    
    # 应用掩码, 将True位置替换为指定值
    return X.masked_fill(mask, value)

缩放点积注意力

#@save
class DotProductAttention(nn.Module):
    """缩放点积注意力"""
    def __init__(self, dropout, **kwargs):
        super(DotProductAttention, self).__init__(**kwargs)
        self.dropout = nn.Dropout(dropout)

    # queries的形状：(batch_size，查询的个数，d)
    # keys的形状：(batch_size，“键－值”对的个数，d)
    # values的形状：(batch_size，“键－值”对的个数，值的维度)
    # valid_lens的形状:(batch_size，)或者(batch_size，查询的个数)
    def forward(self, queries, keys, values, valid_lens=None):
        d = queries.shape[-1]
        # 设置transpose_b=True为了交换keys的最后两个维度
        scores = torch.bmm(queries, keys.transpose(1,2)) / math.sqrt(d)
        self.attention_weights = masked_softmax(scores, valid_lens)
        return torch.bmm(self.dropout(self.attention_weights), values)

多头注意力

多头注意力融合了来自于多个注意力汇聚的不同知识，这些知识的不同来源于相同的查询、键和值的不同的子空间表示。

#@save
def transpose_qkv(X, num_heads):
    """为了多注意力头的并行计算而变换形状"""
    # 输入X的形状:(batch_size，查询或者“键－值”对的个数，num_hiddens)
    # 输出X的形状:(batch_size，查询或者“键－值”对的个数，num_heads，
    # num_hiddens/num_heads)
    X = X.reshape(X.shape[0], X.shape[1], num_heads, -1)

    # 输出X的形状:(batch_size，num_heads，查询或者“键－值”对的个数,
    # num_hiddens/num_heads)
    X = X.permute(0, 2, 1, 3)

    # 最终输出的形状:(batch_size*num_heads,查询或者“键－值”对的个数,
    # num_hiddens/num_heads)
    return X.reshape(-1, X.shape[2], X.shape[3])

#@save
def transpose_output(X, num_heads):
    """逆转transpose_qkv函数的操作"""
    X = X.reshape(-1, num_heads, X.shape[1], X.shape[2])
    X = X.permute(0, 2, 1, 3)
    return X.reshape(X.shape[0], X.shape[1], -1)

#@save
class MultiHeadAttention(nn.Module):
    """多头注意力"""
    def __init__(self, key_size, query_size, value_size, num_hiddens,
                 num_heads, dropout, bias=False, **kwargs):
        super(MultiHeadAttention, self).__init__(**kwargs)
        self.num_heads = num_heads
        self.attention = DotProductAttention(dropout)
        self.W_q = nn.Linear(query_size, num_hiddens, bias=bias)
        self.W_k = nn.Linear(key_size, num_hiddens, bias=bias)
        self.W_v = nn.Linear(value_size, num_hiddens, bias=bias)
        self.W_o = nn.Linear(num_hiddens, num_hiddens, bias=bias)

    def forward(self, queries, keys, values, valid_lens):
        # queries，keys，values的形状:
        # (batch_size，查询或者“键－值”对的个数，num_hiddens)
        # valid_lens　的形状:
        # (batch_size，)或(batch_size，查询的个数)
        # 经过变换后，输出的queries，keys，values　的形状:
        # (batch_size*num_heads，查询或者“键－值”对的个数，
        # num_hiddens/num_heads)
        queries = transpose_qkv(self.W_q(queries), self.num_heads)
        keys = transpose_qkv(self.W_k(keys), self.num_heads)
        values = transpose_qkv(self.W_v(values), self.num_heads)

        if valid_lens is not None:
            # 在轴0，将第一项（标量或者矢量）复制num_heads次，
            # 然后如此复制第二项，然后诸如此类。
            valid_lens = torch.repeat_interleave(
                valid_lens, repeats=self.num_heads, dim=0)

        # output的形状:(batch_size*num_heads，查询的个数，
        # num_hiddens/num_heads)
        output = self.attention(queries, keys, values, valid_lens)

        # output_concat的形状:(batch_size，查询的个数，num_hiddens)
        output_concat = transpose_output(output, self.num_heads)
        return self.W_o(output_concat)