RuntimeError: Input, output and indices must be on the current device

我看了一下模型的输入，都放到了GPU上，然后我在网上查了一下，发现我在模型内部创建了新的模型，这些新的模型也要放在GPU上

        self.bert = BertModel.from_pretrained(bert_config)
        self.multi_head1 = MultiHeadedAttention(8, 768, dropout)
        self.multi_head2 = MultiHeadedAttention(8, 768, dropout)
        self.linear1 = nn.Linear(768, 1)
        self.linear2 = nn.Linear(512, 1)

        self.bert = BertModel.from_pretrained(bert_config)
        self.multi_head1 = MultiHeadedAttention(8, 768, dropout).to(conf['device'])
        self.multi_head2 = MultiHeadedAttention(8, 768, dropout).to(conf['device'])
        self.linear1 = nn.Linear(768, 1).to(conf['device'])
        self.linear2 = nn.Linear(512, 1).to(conf['device'])

上面一个是报错前的代码，下面是放到GPU上后，运行没毛病的代码。很奇怪欸，bert也没to(conf['device'])呀，为啥就能跑呢