Fix a shape annotation and typos in mamba slow forward

[vasqu]

What does this PR do?

It only addresses typos and a wrong shape annotation in the comments of mamba's slow forward call. There's no change in the logic or anything.

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.

@ArthurZucker

[ArthurZucker]

Ok good nit! 🤗 (does jamba also has this typo?)

[vasqu]

Yup, same typo. The shape annotation is correct tho.

Another thing I've noticed in Jamba are these line;

transformers/src/transformers/models/jamba/modeling_jamba.py

Lines 916 to 920 in 5962d62

	if self.training:
	# In training mode, we don't want to perform in-place operations on ssm_state so we can compute the backwards pass
	ssm_state = cache_params.ssm_states[self.layer_idx].clone()
	else:
	ssm_state = cache_params.ssm_states[self.layer_idx]

If you remember my issue from the past ( #29526 ), I've added something similar but not differentiating between training and eval. Might be worth to change in base mamba as well.

[ArthurZucker]

Down to change, but with a bench / something that shows it does produce improvements!