Question

我遇到以下错误，这是回溯：

Traceback (most recent call last):
  File "train.py", line 136, in <module>
    train(epoch)
  File "train.py", line 112, in train
    output = model(data, hc)             # Get outputs of LSTM
  File "/home/ama1128/.conda/envs/matrix/lib/python3.6/site-packages/torch/nn/modules/module.py", line 489, in __call__
    result = self.forward(*input, **kwargs)
  File "/scratch/ama1128/convLSTM/model.py", line 118, in forward
    hc = self.cell_list[t](input=x[0], prev_state=[h_0, c_0])
  File "/home/ama1128/.conda/envs/matrix/lib/python3.6/site-packages/torch/nn/modules/module.py", line 489, in __call__
    result = self.forward(*input, **kwargs)
  File "/scratch/ama1128/convLSTM/model.py", line 50, in forward
    combined = torch.cat((input, h_prev), dim=1) # concatenate along channel axis
RuntimeError: Expected object of backend CUDA but got backend CPU for sequence element 1 in sequence argument at position #1 'tensors'

训练循环：

def train(epoch):
    model.train()
    hc = model.init_hidden(batch_size=1)    
    for batch_idx, (data, target) in enumerate(train_loader):
        optimizer.zero_grad()               
        target = data[1:]                   # Set target, images 2 to 18
        if gpu:                             
            data = data.cuda()
            target = target.cuda()
        output = model(data, hc)            # outputs of LSTM
        loss = criterion(output, target)    
        loss.backward()                     
        optimizer.step()

convLSTMCell转发：

def forward(self, input, prev_state):
    h_prev, c_prev = prev_state
    combined = torch.cat((input, h_prev), dim=1) # concatenate along channel axis

    combined_conv = self.conv(combined)
    cc_i, cc_f, cc_o, cc_g = torch.split(combined_conv, self.hidden_dim, dim=1)

    i = torch.sigmoid(cc_i)
    f = torch.sigmoid(cc_f)
    o = torch.sigmoid(cc_o)
    g = torch.tanh(cc_g)

    c_cur = f * c_prev + i * g
    h_cur = o * torch.tanh(c_cur)

    return h_cur, c_cur

convLSTM类别：初始化状态：

def init_hidden(self, batch_size):
        hidden = torch.zeros(batch_size, self.hidden_dim[0], self.height, self.width)
        cell = torch.zeros(batch_size, self.hidden_dim[0], self.height, self.width)
        if gpu:
            hidden.cuda()
            cell.cuda()
        return hidden, cell

convLSTM转发：

def forward(self, x, hc):
    outputs = []
    states = []
    x = torch.unsqueeze(x, 1)   # change shape from (18,3,128,128) to (18,1,3,128,128)
    h_0, c_0 = hc

    for t in range(0, self.num_layers):
        if t==0:
            hc = self.cell_list[t](input=x[0], prev_state=[h_0, c_0])
                states.append(hc)
            hidden, cell = hc
            outputs.append(hidden.view(3,128,128))
        else:
            h, c = states[t-1]                              # unpack previous states
            if gpu:
                h.cuda()
                c.cuda()
            hc = self.cell_list[t](input=x[t], prev_state=[h,c]) # current states
            states.append(hc)                               # store current states for next cell
            hidden, cell = hc                               # unpack current states
            outputs.append(hidden.view(3,128,128))

使用torch.stack重新格式化

输出，然后将其传递给损失函数。我已尽力将隐藏状态转换为gpu，但错误仍然存在。发生了什么事？

将隐藏状态传递给convLSTM时发生CUDA错误

0 个答案: