Tabular data not moved to gpu automatically

fastai fastai dev
1

I was creating databunch with from df, creating learner with Learner(because I wanted to use custom pytorch model)

magic_model = magic().cuda()
data = TabularDataBunch.from_df("", train_df, "target", valid_idx=valid_index, procs=procs,bs=batch_size,test_df=test_df)#.cuda()
    learn = Learner(data, magic_model, loss_func=torch.nn.BCEWithLogitsLoss(), callback_fns=[partial(EarlyStoppingCallback, monitor='val_loss', min_delta=1e-3, patience=3)])

epoch     train_loss  valid_loss  time    
Traceback (most recent call last):                                               
  File "fastai3_gpu.py", line 95, in <module>
    learn.fit_one_cycle(20, 1e-3,wd=0.05)
  File "/xd/envs/py3.6/lib/python3.6/site-packages/fastai/train.py", line 22, in fit_one_cycle
    learn.fit(cyc_len, max_lr, wd=wd, callbacks=callbacks)
  File "/xd/envs/py3.6/lib/python3.6/site-packages/fastai/basic_train.py", line 199, in fit
    fit(epochs, self, metrics=self.metrics, callbacks=self.callbacks+callbacks)
  File "/xd/envs/py3.6/lib/python3.6/site-packages/fastai/basic_train.py", line 101, in fit
    loss = loss_batch(learn.model, xb, yb, learn.loss_func, learn.opt, cb_handler)
  File "/xd/envs/py3.6/lib/python3.6/site-packages/fastai/basic_train.py", line 26, in loss_batch
    out = model(*xb)
  File "/xd/envs/py3.6/lib/python3.6/site-packages/torch/nn/modules/module.py", line 489, in __call__
    result = self.forward(*input, **kwargs)
  File "fastai3_gpu.py", line 79, in forward
    x = self.fc1(x)
  File "/xd/envs/py3.6/lib/python3.6/site-packages/torch/nn/modules/module.py", line 489, in __call__
    result = self.forward(*input, **kwargs)
  File "/xd/envs/py3.6/lib/python3.6/site-packages/torch/nn/modules/linear.py", line 67, in forward
    return F.linear(input, self.weight, self.bias)
  File "/xd/envs/py3.6/lib/python3.6/site-packages/torch/nn/functional.py", line 1352, in linear
    ret = torch.addmm(torch.jit._unwrap_optional(bias), input, weight.t())
RuntimeError: Expected object of backend CPU but got backend CUDA for argument #4 'mat1'

It seems to me that the model is gpu but data isn’t gpu. But in this Moving inference to the CPU it seems that they should all be gpu. what happened?

Also when I try to use them on gpu, same error shows up.

defaults.device = torch.device('cpu')
fastai.device = torch.device('cpu')
2

I used data = TabularDataBunch.from_df("", train_df, "target", valid_idx=valid_index, procs=procs,bs=batch_size,test_df=test_df, device=torch.device('cuda')) but resulted same ‘expected cpu but got gpu…’

upd: I used data = TabularDataBunch.from_df("", train_df, "target", valid_idx=valid_index, procs=procs,bs=batch_size,test_df=test_df, device=torch.device('cpu')) with magic_model = magic() and the whole model is working on CPU.

Now I wonder how to make it work on GPU? thx in advance.

3

Did you manage to find a solution?

4

If anyone still runs into the same problem.
In my case the issue was caused by my custom coded network. When returning the prediction I had used:

out.type(torch.DoubleTensor)

This sent the tensor to the CPU so sending the double tensor to GPU again fixed it like so:

out.type(torch.DoubleTensor).to(torch.device(‘cuda’))

Not sure if there is a better way of doing this but for now it works.

1 Like