Why my deep learning model is taking even more time in colab?

1

I am using google colab to run my deep learning model. I dont know why is it taking more time in colab/
I am using this snippet for device mounting in training the model. Plz suggest something , its taking 30 min for 1 epoch.
‘’’
if args.use_gpu:

use_cuda = torch.cuda.is_available()

device = torch.device("cuda" if use_cuda else "cpu")

else:

device='cpu'

‘’’
@ptrblck plz help sir

2

Is the notebook setup to use a GPU? Because device will still hold "cpu" if the settings aren’t changed.

1 Like
3

yes it is.
But no change in training duration…

4

The difference between Colab and your local setup would come from all used processing units, i.e.:

  • used CPU
  • data storage and bandwidth to access it
  • used GPU
  • different libraries

etc.
You could profile the code to check, which part of the training takes most of the time.

5

Thanks for reply sir,
But I am not getting last line, How to profile the code?
Actually its taking more time then my local machine.

6

You could add timers to the code and check e.g. how long the data loading needs as done in the ImageNet example. Also, the torch.autograd.profiler might be useful.