problem in notebook 12c_ulmfit

[AFanaei]

when running the 12c_ulmfit notebook for training a classifier i got the following error:

course-v3\nbs\dl2\exp\nb_09b.py in one_batch(self, i, xb, yb)
     41             if not self.in_train: return
---> 42             self.loss.backward();                           self('after_backward')
     43             self.opt.step();                                self('after_step')

course-v3\.env\lib\site-packages\torch\tensor.py in backward(self, gradient, retain_graph, create_graph)
    194         """
--> 195         torch.autograd.backward(self, gradient, retain_graph, create_graph)
    196 

course-v3\.env\lib\site-packages\torch\autograd\__init__.py in backward(tensors, grad_tensors, retain_graph, create_graph, grad_variables)
     92 
---> 93     grad_tensors = _make_grads(tensors, grad_tensors)
     94     if retain_graph is None:

course-v3\.env\lib\site-packages\torch\autograd\__init__.py in _make_grads(outputs, grads)
     34                     raise RuntimeError("grad can be implicitly created only for scalar outputs")
---> 35                 new_grads.append(torch.ones_like(out, memory_format=torch.preserve_format))
     36             else:

RuntimeError: CUDA error: device-side assert triggered

it will fail no matter using pretraining or not.

[AFanaei]

i managed to find the problem.
it was because the output size of pooling layer was not matching the number of categories in data set. (i was running the note book with my own dataset)
after changing the number of output the problem was fixed .

emb_sz, nh, nl = 100, 100, 1
dps = tensor([0.4, 0.3, 0.4, 0.05, 0.5]) * 0.25
tok_pad = vocab.index(PAD)
n_out = len(proc_cat.vocab)
model = get_text_classifier(len(vocab), emb_sz, nh, nl, n_out , tok_pad, bptt, *dps)