3bc7e38ed98f3b7a13fcee2726ec38b27e5c4e1b,onmt/Models.py,Encoder,__init__,#Encoder#Any#Any#Any#,36

Before Change

                           bidirectional=opt.brnn)


        self.multiattn = nn.ModuleList([onmt.modules.MultiHeadedAttention(8, self.hidden_size) for _ in range(self.layers)])
        self.linear_out = nn.ModuleList([BLinear(self.hidden_size, 2*self.hidden_size) for _ in range(self.layers)])
        self.linear_final = nn.ModuleList([BLinear(2*self.hidden_size, self.hidden_size) for _ in range(self.layers)])

        self.layer_norm = nn.ModuleList([BLayerNorm(self.hidden_size) for _ in range(self.layers)])

After Change

        
        if self.positional_encoding:
            self.pe = make_positional_encodings(opt.word_vec_size, 5000).cuda()
        if self.encoder_layer == "transformer":
            self.transformer = nn.ModuleList([TransformerEncoder(self.hidden_size, opt)
                                               for _ in range(opt.layers)])
        
    def load_pretrained_vectors(self, opt):
        if opt.pre_word_vecs_enc is not None:
            pretrained = torch.load(opt.pre_word_vecs_enc)
            self.word_lut.weight.data.copy_(pretrained)

Instances