torch.nn.DataParallel error

[gentlebreeze1]

trian error 👍 RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:1 and cuda:0!
how to solve this problem? i use to(device),but it do not work.error in @szq0214

[szq0214]

@gentlebreeze1 I did not meet this error before, are you running with our default codes? I guess your GPUs for loading the model and running the model are different, so you encounter this error.

[gentlebreeze1]

@gentlebreeze1 I did not meet this error before, are you running with our default codes? I guess your GPUs for loading the model and running the model are different, so you encounter this error.

i train my own model,use
ckpt={'epoch':epoch+1, 'state_dict':net.module.state_dict(), 'optimizer':optimizer.param_groups[0]['lr']} torch.save(ckpt, weights_path)
and load
model = _create_model(model_name, teacher=False, pretrain=True) checkpoint = torch.load(state_file) if 'state_dict' in checkpoint: model.load_state_dict(checkpoint['state_dict']) else: model.load_state_dict(checkpoint)
I don’t think it’s a problem with model loading，the erorr occurred on Discriminator.conv1

[gentlebreeze1]

i change as follows, it works.Will it affect the final result?@szq0214
class discriminatorLoss(nn.Module):
def init(self, outputs_size,K=2, loss=nn.BCEWithLogitsLoss()):
super(discriminatorLoss, self).init()
self.conv1 = nn.Conv2d(in_channels=outputs_size, out_channels=outputs_size // K, kernel_size=1, stride=1,
bias=True)
outputs_size = outputs_size // K
self.conv2 = nn.Conv2d(in_channels=outputs_size, out_channels=outputs_size // K, kernel_size=1, stride=1,
bias=True)
outputs_size = outputs_size // K
self.conv3 = nn.Conv2d(in_channels=outputs_size, out_channels=2, kernel_size=1, stride=1, bias=True)
self.loss = loss

def forward(self, outputs, targets):
    inputs = [torch.cat((i, j), 0) for i, j in zip(outputs, targets)]
    inputs = torch.cat(inputs, 1)
    batch_size = inputs.size(0)
    target = torch.FloatTensor([[1, 0] for _ in range(batch_size // 2)] + [[0, 1] for _ in range(batch_size // 2)])
    target = target.to(inputs[0].device)
    inputs=inputs[:,:,None,None]
    out = F.relu(self.conv1(inputs))
    out = F.relu(self.conv2(out))
    out = F.relu(self.conv3(out))
    output = out.view(out.size(0), -1)
    res = self.loss(output, target)
    return res

[gentlebreeze1]

my pre model size is 59.7M ,why save model size is 109M? after distillation ,model will grow bigger?@szq0214

[szq0214]

Hi @gentlebreeze1 The model size will not grow after distillation, I guess you saved other parameters in your checkpoint.