Hello
On the colab machine, even when in fallback mode, is extremely slow.
I solved the assignment in a few minutes and have been stuck with running it for hours now.
My computer has a strong GPU, but my cuda version is too new for tensorflow 2.8.0. Is there a service I can pay and get the correct system environment to get a model that the grader will accept? even if I finish this model with Colab, I don’t want to do the rest of the coursework there.
Thank you.
Hi Ofer. We just updated the notebook to avoid the fallback runtime. Please reload the notebook from your classroom. Upgrading the graders is still in the pipeline but hopefully this revision will avoid issues with the fallback. Unfortunately, I can’t be certain if all will run as expected on your local machine. Training per epoch is about 40 seconds in this lab. The next labs should be faster to train.
Hello @Ofer
for grader to accept the model would not totally depend on the fallback runtime, but also depend on how your model algorithm has been created. So in case your assignment is stil extremely slow with fallback runtime, then kindly go back to the assignment and check if you are getting the desired/expected accuracy. Also
Another solution I could give you with my personal experience when I was doing the last two courses of this specialisation was that I used to try running a model the next time on my GPU for better and faster result, as for a day I would already tried 2-3 times running my model to get the desired output. This is personal suggestion.
Let us know, if you need any extra help or still facing the issue even after following Chris’s and my suggestion.
Regards
DP
1 Like