Unfortunately yes. But now there are also Kaggle Kernels available to run the script from Clouderizer. If you open a notebook and kaggle and commit the code with the code pasted in one of the cells you can run clouderizer smoothly for 9 hours. Kaggle doesn’t disconnect in this interval.
The only downsize in Kaggle compared to Colab is that the multiprocessing doesn’t work. Pytorch always throws an error. This way the code runs somewhat 1/3 slower than Colab.
I’ve created a public template in Clouderizer with fastai 1.0.x and Pytorch 1.0 up and running.
Go in Public Templates in Clouderizer. The project name is Fastai-1.0.
I hope this helps.