Well. Simply put, this is an issue on the official Space side:
I was able to reproduce the failure with the current grader code. The failing request is made by the Space backend—not by your browser—and it is still targeting the legacy api-inference.huggingface.co route. Current Hugging Face inference uses the Inference Providers routing system instead.
There is also a problem with the current fallback: when the model call fails, the grader compares the formatted submission and reference solution as strings. This means a functionally correct alternative solution can be reported as incorrect during an infrastructure failure.
I opened Space PR #10 to update the routing and prevent an unavailable grader from recording an incorrect answer. The same failure is also tracked in course issue #1270.
The PR has not been merged yet, so the official quiz may continue failing until the Space is updated and rebuilt. I do not think further browser, DNS, or network changes on your side will help at this point.
I have the same problem (Part 11), even for code questions like set parameter to True it returns error no address association with the hostname. Should I keep trying different formats of answers or is the server down and I should wait? Thank you very much for you answer.
