Fix item call in radam

[mlazos]

Torch.compile doesn't support data dependent control flow, in some cases this can be supported with a torch.where. To avoid a graph break we check anis_compiling condition to avoid forcing eager to all an extra kernel for torch.where.

Stack from ghstack (oldest at bottom):

• Enable dynamo traced test_param_group_with_lrscheduler_goes_right_direction #124544
• Fix capturable enablement conditions #125826
• Tighten fallback conditions for compiled optim #125825
• Stack tensors for adamax #125833
• -> Fix item call in radam #125824
• Fix call to item in asgd #125823

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/125824

• 📄 Preview Python docs built from this PR
• 📄 Preview C++ docs built from this PR
• ❓ Need help or want to give feedback on the CI? Visit the bot commands wiki or our office hours

Note: Links to docs will display an error until the docs builds have been completed.

✅ You can merge normally! (1 Unrelated Failure)

As of commit 325ad14 with merge base c59a236 ():

BROKEN TRUNK - The following job failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

• pull / linux-focal-cuda11.8-py3.10-gcc9 / test (distributed, 3, 3, linux.8xlarge.nvidia.gpu) (gh) (trunk failure)
  !!!!!!!!!!!!!!!!!!!!!!!!!!!!!! KeyboardInterrupt !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[vadimkantorov]

not related to this PR, but maybe _dispatch_sqrt (in this file and elsewhere) can be replaced by simply ** 0.5 which is supposed to work both for python scalars and for tensors (this is already used in most occurences in this file as well, the only dispatch_sqrt usage remaining)