Remove tensor option for _global_exception_occured

[irenedea]

What does this PR do?

Makes it able to pass MLFlowLogger to a child process with export PYTORCH_CUDA_ALLOC_CONF="expandable_segments:True"

Addresses the error:

RuntimeError: Tensors allocated with expandable_segments:True cannot be shared between processes. Consider using expandable_segments:False in data loading workers via torch.cuda.memory._set_allocator_settings('expandable_segments:False')

Run with fix: test-hf-checkpoint-fix-Vknjzg
Run without fix: test-hf-checkpoint-baseline-5ZGdqG

What issue(s) does this change relate to?

Before submitting

• Have you read the contributor guidelines?
• Is this change a documentation change or typo fix? If so, skip the rest of this checklist.
• Was this change discussed/approved in a GitHub issue first? It is much more likely to be merged if so.
• Did you update any related docs and document your change?
• Did you update any related tests and add any new tests related to your change? (see testing)
• Did you run the tests locally to make sure they pass?
• Did you run pre-commit on your change? (see the pre-commit section of prerequisites)

[mvpatel2000]

nit

[chenmoneygithub]

LGTM!

[mvpatel2000]

LGTM