06_multicat codebook different from lecture

Part 1 (2020)
1

The lecture on Lesson 06 @55:30.

shows the following code

x,y = dls.train.one_batch()
activs = learn.model(x)
activs.shape

however, the codebook provides a slightly different method and introduces to_cpu method with no explanation.

x,y = to_cpu(dls.train.one_batch())
activs = learn.model(x)
activs.shape

So I tried running the former code illustrated in the video and received the following error.

---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
/tmp/ipykernel_81/2222099996.py in <module>
      2 #learn.model is the thing that contains the model itself (i.e., cnn) you can treat it as a function
      3 x,y = dls.train.one_batch()
----> 4 activs = learn.model(x)
      5 activs.shape

/opt/conda/lib/python3.7/site-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
   1100         if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
   1101                 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1102             return forward_call(*input, **kwargs)
   1103         # Do not call functions when jit is used
   1104         full_backward_hooks, non_full_backward_hooks = [], []

/opt/conda/lib/python3.7/site-packages/torch/nn/modules/container.py in forward(self, input)
    139     def forward(self, input):
    140         for module in self:
--> 141             input = module(input)
    142         return input
    143 

/opt/conda/lib/python3.7/site-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
   1100         if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
   1101                 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1102             return forward_call(*input, **kwargs)
   1103         # Do not call functions when jit is used
   1104         full_backward_hooks, non_full_backward_hooks = [], []

/opt/conda/lib/python3.7/site-packages/torch/nn/modules/container.py in forward(self, input)
    139     def forward(self, input):
    140         for module in self:
--> 141             input = module(input)
    142         return input
    143 

/opt/conda/lib/python3.7/site-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
   1100         if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
   1101                 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1102             return forward_call(*input, **kwargs)
   1103         # Do not call functions when jit is used
   1104         full_backward_hooks, non_full_backward_hooks = [], []

/opt/conda/lib/python3.7/site-packages/torch/nn/modules/conv.py in forward(self, input)
    444 
    445     def forward(self, input: Tensor) -> Tensor:
--> 446         return self._conv_forward(input, self.weight, self.bias)
    447 
    448 class Conv3d(_ConvNd):

/opt/conda/lib/python3.7/site-packages/torch/nn/modules/conv.py in _conv_forward(self, input, weight, bias)
    441                             _pair(0), self.dilation, self.groups)
    442         return F.conv2d(input, weight, bias, self.stride,
--> 443                         self.padding, self.dilation, self.groups)
    444 
    445     def forward(self, input: Tensor) -> Tensor:

/opt/conda/lib/python3.7/site-packages/fastai/torch_core.py in __torch_function__(self, func, types, args, kwargs)
    339         convert=False
    340         if _torch_handled(args, self._opt, func): convert,types = type(self),(torch.Tensor,)
--> 341         res = super().__torch_function__(func, types, args=args, kwargs=kwargs)
    342         if convert: res = convert(res)
    343         if isinstance(res, TensorBase): res.set_meta(self, as_copy=True)

/opt/conda/lib/python3.7/site-packages/torch/_tensor.py in __torch_function__(cls, func, types, args, kwargs)
   1049 
   1050         with _C.DisableTorchFunction():
-> 1051             ret = func(*args, **kwargs)
   1052             if func in get_default_nowrap_functions():
   1053                 return ret

RuntimeError: Input type (torch.cuda.FloatTensor) and weight type (torch.FloatTensor) should be the same

Can anyone shed some light on this?

2

The problem is the input is on the GPU (cuda in torch.cuda.FloatTensor), and the model weights are on the CPU (torch.FloatTensor). If you want to do a calculation, you need both on either GPU or CPU. You can’t mix and match.

The “to_cpu” function moves the “one_batch” from GPU to CPU.
x,y = to_cpu(dls.train.one_batch())
So, you would have the x and y inputs on CPU, and the weights on CPU.

When you reverted back to
“x,y = dls.train.one_batch()”
The x and y inputs are on the GPU. The fastai dataloaders move the inputs into the GPU (check the source code). Regular Pytorch dataloaders keep the inputs in the CPU.