Try again, it should be fixed now.

EDIT:
disregard my previous comment.
the question requires that you do not remove duplicates, see Daotar's comment, this solved my issue.

Yes, I've tried a previously working solution and it failed, so maybe someone changed the test suite and now it doesn't work.

I get that, but still... the timeout is erratic and you can see others complaining about that too

The 12000 miliseconds timeout is for the whole test suite, in Python there are 9 fixed tests, 100 random tests and 100 random huge tests.

There is something fishy with this exercise, or better put, with the test cases. I get it, the first couple of attempts may not be optimal but when I ran my solution on large arrays (in fact much larger than stated in the instructions e.g.:

l = np.random.randint(np.iinfo(np.int64).max - (1<<62) , np.iinfo(np.int64).max, size=1000000)

it still completes in about 5 sec rather than taking longer than 12000 milsec timeout claimed by the test platform.

This question has an unclear description/instructions with respect to corner cases, specifically regarding how duplicates are to be handled.. Clarification would improve the question