fix gradient accumulation steps without using deepspeed

[baoyf4244]

What does this PR do?

when training without deepspeed, before loss backward, the loss is not scaling with gradient_accumulation_steps, this is caused by the parameter gradient_accumulation_steps not passed to Accelerator when creating this object.

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,
  Pull Request section?
• 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.

[muellerzr]

Please try using main again, #35651 has fixed this.

In the future, the trainer explicitly does not rely on the Accelerator’s highest GA interface because it doesn’t account (easily) for the token nuances needed that Unsloth found. (We use the second/mid level API instead)

[baoyf4244]

Ok, I'm sorry I have not find that issue a few days ago.