deep-learning - 如何修复 "RuntimeError: Function AddBackward0 returned an invalid gradient at index 1 - expected type torch.FloatTensor but got torch.LongTensor"

Question

我在运行机器翻译代码时遇到了这个错误。

RuntimeError Traceback (most recent call last) in 5 decoder = Decoder(len(out_vocab), embed_size, num_hiddens, num_layers, 6 attention_size, drop_prob) ----> 7 train(encoder, decoder, dataset, lr, batch_size, num_epochs)

in train(encoder, decoder, dataset, lr, batch_size, num_epochs) 13 dec_optimizer.zero_grad() 14 l = batch_loss(encoder, decoder, X, Y, loss) ---> 15 l.backward() 16 enc_optimizer.step() 17 dec_optimizer.step()

/usr/lib64/python3.6/site-packages/torch/tensor.py in backward(self, gradient, retain_graph, create_graph) 105 products. Defaults to False. 106 """ --> 107 torch.autograd.backward(self, gradient, retain_graph, create_graph) 108 109 def register_hook(self, hook):

/usr/lib64/python3.6/site-packages/torch/autograd/init.py in backward(tensors, grad_tensors, retain_graph, create_graph, grad_variables) 91 Variable._execution_engine.run_backward( 92 tensors, grad_tensors, retain_graph, create_graph, ---> 93 allow_unreachable=True) # allow_unreachable flag 94 95

RuntimeError: Function AddBackward0 returned an invalid gradient at index 1 - expected type torch.FloatTensor but got torch.LongTensor

我认为错误出在 batch_loss 函数中。但我不知道为什么，也无法修复它。

def batch_loss(encoder, decoder, X, Y, loss):
    batch_size = X.shape[0]
    enc_state = None
    enc_outputs, enc_state = encoder(X, enc_state)
    # 初始化解码器的隐藏状态
    dec_state = decoder.begin_state(enc_state)
    # 解码器在最初时间步的输入是BOS
    dec_input = torch.tensor([out_vocab.stoi[BOS]] * batch_size)
    # 我们将使用掩码变量mask来忽略掉标签为填充项PAD的损失
    mask, num_not_pad_tokens = torch.ones(batch_size), 0
    l = torch.tensor([0])
    for y in Y.t():
        dec_output, dec_state = decoder(dec_input, dec_state, enc_outputs)
        l = l + (mask * loss(dec_output, y)).sum()
        dec_input = y  # 使用强制教学
        num_not_pad_tokens += mask.sum().item()
        # 当遇到EOS时，序列后面的词将均为PAD，相应位置的掩码设成0
        mask = mask * (y != out_vocab.stoi[EOS]).float()
    return l / num_not_pad_tokens

def train(encoder, decoder, dataset, lr, batch_size, num_epochs):
    d2lt.params_init(encoder, init=nn.init.xavier_uniform_)
    d2lt.params_init(decoder, init=nn.init.xavier_uniform_)

    enc_optimizer = optim.Adam(encoder.parameters(), lr=lr)
    dec_optimizer = optim.Adam(decoder.parameters(), lr=lr)
    loss = nn.CrossEntropyLoss(reduction='none')
    data_iter = tdata.DataLoader(dataset, batch_size, shuffle=True)
    for epoch in range(num_epochs):
        l_sum = 0.0
        for X, Y in data_iter:
            enc_optimizer.zero_grad()
            dec_optimizer.zero_grad()
            l = batch_loss(encoder, decoder, X, Y, loss)
            l.backward()
            enc_optimizer.step()
            dec_optimizer.step()
            l_sum += l.item()
        if (epoch + 1) % 10 == 0:
            print("epoch %d, loss %.3f" % (epoch + 1, l_sum / len(data_iter)))

期待正面的回复。

Answer 1

感谢 Proyag。只需将 l = torch.tensor([0]) 替换为 l = torch.tensor([0], dtype=torch.float) 即可解决我的问题。