[fix] fix bug in mixed precision

[CuddleSabe]

the transformer before accelerator.prepare should not convert to fp16 in mix precision; it will broken the grad precision and result in worse training result

What does this PR do?

the transformer before accelerator.prepare should not convert to fp16 in mix precision;
it will broken the grad precision and result in worse training result

Fixes # (issue)

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline?
• Did you read our philosophy doc (important for complex PRs)?
• Was this discussed/approved via a GitHub issue or the forum? Please add a link to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.