TransformerDecoder is a stack of N decoder layers Args: tgt_n_token: the size of target vocab d_model: the number of expected features in the decoder inputs (default=512). n_head: the number of heads in the multi head attention models (default=8).
| 103 | |
| 104 | |
| 105 | class TransformerDecoder(layer.Layer): |
| 106 | """TransformerDecoder is a stack of N decoder layers |
| 107 | Args: |
| 108 | tgt_n_token: the size of target vocab |
| 109 | d_model: the number of expected features in the decoder inputs (default=512). |
| 110 | n_head: the number of heads in the multi head attention models (default=8). |
| 111 | dim_feedforward: the dimension of the feedforward network model (default=2048). |
| 112 | n_layers: the number of sub-decoder-layers in the decoder (default=6). |
| 113 | """ |
| 114 | |
| 115 | def __init__(self, tgt_n_token, d_model=512, n_head=8, dim_feedforward=2048, n_layers=6): |
| 116 | super(TransformerDecoder, self).__init__() |
| 117 | self.tgt_n_token = tgt_n_token |
| 118 | self.d_model = d_model |
| 119 | self.n_head = n_head |
| 120 | self.dim_feedforward = dim_feedforward |
| 121 | self.n_layers = n_layers |
| 122 | |
| 123 | # target_emb / pos_emb / n-layers |
| 124 | self.target_emb = layer.Embedding(input_dim=tgt_n_token, output_dim=d_model) |
| 125 | self.target_pos_emb = layer.Embedding(input_dim=tgt_n_token, output_dim=d_model) |
| 126 | self.layers = [] |
| 127 | for _ in range(n_layers): |
| 128 | self.layers.append(TransformerDecoderLayer(d_model=d_model, n_head=n_head, dim_feedforward=dim_feedforward)) |
| 129 | |
| 130 | def forward(self, dec_inputs, enc_inputs, enc_outputs): |
| 131 | """ |
| 132 | Args: |
| 133 | dec_inputs: [batch_size, tgt_len] |
| 134 | enc_inputs: [batch_size, src_len] |
| 135 | enc_outputs: [batch_size, src_len, d_model] |
| 136 | |
| 137 | """ |
| 138 | |
| 139 | # [batch_size, tgt_len, d_model] |
| 140 | tgt_word_emb = self.target_emb(dec_inputs) |
| 141 | self.target_pos_emb.initialize(dec_inputs) |
| 142 | self.target_pos_emb.from_pretrained(W=TransformerDecoder._get_sinusoid_encoding_table(self.tgt_n_token, self.d_model), |
| 143 | freeze=True) |
| 144 | # [batch_size, tgt_len, d_model] |
| 145 | tgt_pos_emb = self.target_pos_emb(dec_inputs) |
| 146 | # [batch_size, tgt_len, d_model] |
| 147 | dec_outputs = autograd.add(tgt_word_emb, tgt_pos_emb) |
| 148 | |
| 149 | # dec_self_attn_pad_mask [batch_size, tgt_len, tgt_len] |
| 150 | dec_self_attn_pad_mask = TransformerDecoder._get_attn_pad_mask(dec_inputs, dec_inputs) |
| 151 | # [batch_size, tgt_len, tgt_len] |
| 152 | dec_self_attn_subsequent_mask = TransformerDecoder._get_attn_subsequence_mask(dec_inputs) |
| 153 | |
| 154 | # dec_self_attn_mask [batch_size, tgt_len, tgt_len] |
| 155 | dec_self_attn_mask = tensor.gt((dec_self_attn_pad_mask + dec_self_attn_subsequent_mask), 0) |
| 156 | |
| 157 | # dec_enc_attn_mask [batch_size, tgt_len, src_len] |
| 158 | dec_enc_attn_mask = TransformerDecoder._get_attn_pad_mask(dec_inputs, enc_inputs) |
| 159 | |
| 160 | dec_self_attns, dec_enc_attns = [], [] |
| 161 | |
| 162 | for layer in self.layers: |