Disable functorch modes in testing's freeze_rng_state(), part 2 (#81109)
I forgot to update one line in
https://github.com/pytorch/pytorch/pull/81006. torch.get_rng_state()
returns a Tensor that can also be affected by modes so it also needs a
no_functorch() context manager.
Test Plan:
- tested with functorch tests on CUDA (that's how I discovered this
problem)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/81109
Approved by: https://github.com/samdow