A script to download artifacts and perform CI error statistics

[ydshieh]

What does this PR do?

This script is helpful for the past CI project. It downloads all artifacts from a workflow run, and get the error statistics + the corresponding failing tests.

errors.json: the places where error occur + what those errors are

failed_tests.json: which test methods failed (can be used to determine which models will be supported in a specific backend version)

** We might adjust this script a bit once we start to perform the automation of the past CI project. **

Currently, it prints something as the following (but we save the full information in 2 json files)

('RuntimeError: Integer division of tensors using div or / is no longer supported, and in a future release div will perform true division as in Python 3. Use true_divide or floor_divide (// in Python) instead.', 66)
("AttributeError: module 'torch.jit' has no attribute '_state'", 51)
("AttributeError: module 'torch' has no attribute 'minimum'", 45)
("AttributeError: module 'torch' has no attribute 'multiply'", 25)
('AttributeError: Caught AttributeError in replica 0 on device 0.', 3)
('RuntimeError: "normal_kernel_cpu" not implemented for \'BFloat16\'', 3)
("AssertionError: Couldn't trace module.", 3)
("AttributeError: module 'torch' has no attribute 'isneginf'", 2)
("TypeError: where(): argument 'input' (position 2) must be Tensor, not int", 2)
("AttributeError: 'Tensor' object has no attribute 'nansum'", 1)
('RuntimeError: Caught RuntimeError in replica 0 on device 0.', 1)

[ydshieh]

I don't like this place much for the security concern. But at this moment, I might be the only person.

[HuggingFaceDocBuilderDev]

The documentation is not available anymore as the PR was closed or merged.

[LysandreJik]

Looks good to me, thank you @ydshieh