[🐛BUG] DIN Model, 负采样时，interaction.repeat 处理 token_seq 类型变量的时候维度不匹配错误

[transposition]

Describe the bug

 def repeat(self, sizes):
        """Repeats each tensor along the batch dim.
        Args:
            sizes (int): repeat times.
        Example:
            >>> a = Interaction({'k': torch.zeros(4)})
            >>> a.repeat(3)
            The batch_size of interaction: 12
                k, torch.Size([12]), cpu
            >>> a = Interaction({'k': torch.zeros(4, 7)})
            >>> a.repeat(3)
            The batch_size of interaction: 12
                k, torch.Size([12, 7]), cpu
        Returns:
            a copyed Interaction object with repeated Tensors.
        """
        ret = {}
        for k in self.interaction:
            if len(self.interaction[k].shape) == 1:
                ret[k] = self.interaction[k].repeat(sizes)
            else:
                ret[k] = self.interaction[k].repeat([sizes, 1])
        new_pos_len_list = self.pos_len_list * sizes if self.pos_len_list else None
        new_user_len_list = self.user_len_list * sizes if self.user_len_list else None
        return Interaction(ret, new_pos_len_list, new_user_len_list)

对于token_seq  self.interaction[k] 会是三维（batch size, hist length, seq  size），而不是两维
 if len(self.interaction[k].shape) == 1:
                ret[k] = self.interaction[k].repeat(sizes)
            else:
                ret[k] = self.interaction[k].repeat([sizes, 1])  

Expected behavior

 if len(self.interaction[k].shape) == 1:
                ret[k] = self.interaction[k].repeat(sizes)
            else:
                ret[k] = self.interaction[k].repeat([sizes ] + [1] *( len(self.interaction[k].shape) -1 )  )

[chenyushuo]

感谢你发现了这个bug，我们在#933 中修复了它。