GPU problem with grading C1W4A_Build_a_Conditional_GAN

I received this error for grading my assignment for Week4 :
Cell #UNQ_C4. Can’t compile the student’s code. Error: AssertionError(‘\nFound no NVIDIA driver on your system. Please check that you\nhave an NVIDIA GPU and installed a driver from\nOfficial Drivers | NVIDIA’,)

It is very similar to this in week1 ; C1W1 Assignment submission error

So I began by applying the recommendation of this topic. I refreshed totally the file to be sure that I changed only the code between the right lines… but I still have the same grading error… On the other hand the result of the calculation is quite nice, creating fabulous faky figures !!!
Many thanks for your help

Since you’ve already checked for the usual problems, the only other thing I can think of is - did you happen to use the global variable device in your code in #UNQ_C4? It was defined earlier as device = 'cuda', which won’t work in the grader.

Hello @Wendy ,
I checked my code I the answer is no. I have not used this device variable in then code I wrote (but It appears in then code provided in the course).

Hmm. Weird. I tried submitting my version just to make sure the grader still works with it - which it does.

If you’d like to DM me a copy of your ipynb, I’m happy to take a look and see if notice what the issue is.

A couple of comments for anyone else who might run into this issue:

  1. My advice above to avoid using the global device variable was wrong in this case. As @Philippe_Loue pointed out, the provided portion of the code in #UNQ_C4 uses device and that portion of the code of course works in the grader, so the grader must be assigning the device global to an appropriate value before running the #UNQ_C4 cell.

  2. One tip for the implementation of your portion of the code in #UNQ_C4: Don’t forget that you can use the variables defined for you earlier in the cell (one_hot_labels, etc). Using these can help you make sure to get the device assignments right so they work with the grader (which uses CPU) as well as in the Coursera environment (which uses GPU).