08a31864a8e7a633546790d8fed54455b914d96b,python/baseline/pytorch/transformer.py,TransformerDecoder,__init__,#TransformerDecoder#Any#Any#Any#Any#Any#Any#,178

Before Change

        self.feed_forward = FFN(d_model, pdrop, d_ff=d_ff, activation_type=activation_type)
        self.ln1 = LayerNorm(d_model)
        self.ln2 = LayerNorm(d_model)
        self.ln3 = LayerNorm(d_model)
        self.dropout = nn.Dropout(pdrop)

    def forward(self, x, memory, src_mask, tgt_mask):

After Change

    def __init__(self, num_heads, d_model, pdrop, scale=True, activation_type="relu", d_ff=None):
        super(TransformerDecoder, self).__init__()
        self.d_model = d_model
        self.d_ff = d_ff if d_ff is not None else 4 * d_model
        self.self_attn = MultiHeadedAttention(num_heads, self.d_model, pdrop, scale=scale)
        self.src_attn = MultiHeadedAttention(num_heads, self.d_model, pdrop, scale=scale)
        self.ffn = nn.Sequential(pytorch_linear(self.d_model, self.d_ff),
                                 pytorch_activation(activation_type),

Instances