Model.cuda() is not working

sevdaasyn · January 2, 2020, 4:50pm

I want to train my I3D network with GPU but got following error.

Traceback (most recent call last):
  File "train_sample.py", line 329, in <module>
    model_conv = train_model(rgb_i3d, criterion, optimizer,exp_lr_scheduler, num_epochs=5)
  File "train_sample.py", line 287, in train_model
    rbg_score, rgb_logits = rgb_i3d(inputs)
  File "/home/caglar/.local/lib/python3.6/site-packages/torch/nn/modules/module.py", line 541, in __call__
    result = self.forward(*input, **kwargs)
  File "/home/caglar/train-i3d/I3D_Pytorch.py", line 313, in forward
    logits = self.features(x)
  File "/home/caglar/.local/lib/python3.6/site-packages/torch/nn/modules/module.py", line 541, in __call__
    result = self.forward(*input, **kwargs)
  File "/home/caglar/.local/lib/python3.6/site-packages/torch/nn/modules/container.py", line 92, in forward
    input = module(input)
  File "/home/caglar/.local/lib/python3.6/site-packages/torch/nn/modules/module.py", line 541, in __call__
    result = self.forward(*input, **kwargs)
  File "/home/caglar/.local/lib/python3.6/site-packages/torch/nn/modules/conv.py", line 480, in forward
    self.padding, self.dilation, self.groups)
RuntimeError: Input type (torch.cuda.FloatTensor) and weight type (torch.FloatTensor) should be the same

This is the network that I used for train.

github.com

rimchang/kinetics-i3d-Pytorch/blob/master/model/I3D_Pytorch.py

import torch.nn as nn
import torch

class BasicConv3d(nn.Module):
    def __init__(self, in_planes, out_planes, kernel_size, stride, padding=0):
        super(BasicConv3d, self).__init__()
        self.conv = nn.Conv3d(in_planes, out_planes, kernel_size=kernel_size, stride=stride, padding=padding,
                              bias=False)  # verify bias false

        # verify defalt value in sonnet
        self.bn = nn.BatchNorm3d(out_planes, eps=1e-3, momentum=0.001, affine=True)
        self.relu = nn.ReLU(inplace=True)

    def forward(self, x):
        x = self.conv(x)
        x = self.bn(x)
        x = self.relu(x)
        return x

This file has been truncated. show original

And this is related part of my codes

device = torch.device( "cuda:0" if torch.cuda.is_available() else "cpu")  
#torch.cuda.is_available() return True in my compter

rgb_i3d = I3D(num_classes=NUM_CLASSES , input_channel=3)
rgb_i3d = rgb_i3d.to(device)
state_dict = torch.load(_CHECKPOINT_PATHS)
rgb_i3d.load_state_dict(state_dict)


for inputs, labels in dataloader:
     inputs = inputs.to(device)
     rbg_score, rgb_logits = rgb_i3d(inputs) #there is where my error occur

What is the problem? How can I fix it?

ptrblck · January 2, 2020, 5:17pm

Could you check the .device attribute of any parameter in your model before and after loading the state_dict, please?

sevdaasyn · January 2, 2020, 5:35pm

Before and after loading the state_dict , all device attributes are cuda:0.

for param in rgb_i3d.parameters():
    print (param.device)

ptrblck · January 2, 2020, 5:53pm

The error message points to a conv parameter, which is apparently not on the device.
Are you changing the model somehow or are you using another model in the script additionally to the one you’ve mentioned?

sevdaasyn · January 2, 2020, 6:03pm

Oh thank you I found my problem. I have invoked model.cuda() before finetune. When I moved cuda after it and problem solved.