247203f29b7e841204c76d922c1ea5b2680c3663,reagent/models/seq2slate.py,DecoderPyTorch,forward,#DecoderPyTorch#Any#Any#Any#Any#,213

Before Change

        self.num_heads = num_heads

    def forward(self, tgt, memory, tgt_src_mask, tgt_tgt_mask):
        tgt = tgt.transpose(0, 1)
        memory = memory.transpose(0, 1)
        // Pytorch assumes:
        // (1) mask is bool
        // (2) True -> item should be ignored in attention

After Change

        probs_for_placeholders = torch.zeros(
            batch_size, tgt_seq_len, 2, device=output.device
        )
        probs = torch.cat((probs_for_placeholders, output), dim=2)
        return probs


class MultiHeadedAttention(nn.Module):

Instances