Revert logits to float change for accuracy

[yeonsily]

What does this PR do?

We noticed that the mixtral output is generated diffrently compare to the previous release and it's because of logtis dtype change.

<style> </style>

v1.15.0	v1.16-release	transformers_4_49	reference
DeepSpeed is a machine learning framework that enables training of large models on a single machine with multiple GPUs. It is designed to be easy to use and efficient, and it supports a wide range of models and tasks.\n\nDeepSpeed is a deep learning framework that enables training of large models on a single machine with multiple GPUs. It is designed to be easy to use and efficient, and it supports a wide range of models and tasks.\n\nDeepSpeed is a deep learning framework that enables training of large models on a	DeepSpeed is a machine learning framework that enables training of large models on a single machine with multiple GPUs. It is designed to be easy to use and efficient, and it supports a wide range of models and tasks.\n\nDeepSpeed is a deep learning framework that enables training of large models on a single machine with multiple GPUs. It is designed to be easy to use and efficient, and it supports a wide range of models and tasks.\n\nDeepSpeed is a deep learning framework that enables training of large models on a	DeepSpeed is a machine learning framework that enables training of large models on a single machine with a single GPU. It is designed to be easy to use and to provide high performance.\n\nDeepSpeed is a machine learning framework that enables training of large models on a single machine with a single GPU. It is designed to be easy to use and to provide high performance.\n\nDeepSpeed is a machine learning framework that enables training of large models on a single machine with a single GPU. It is designed to be easy to use and	DeepSpeed is a machine learning framework that enables training of large models on a single machine with multiple GPUs. It is designed to be easy to use and efficient, and it supports a wide range of models and tasks.\n\nDeepSpeed is a deep learning framework that enables training of large models on a single machine with multiple GPUs. It is designed to be easy to use and efficient, and it supports a wide range of models and tasks.\n\nDeepSpeed is a deep learning framework that enables training of large models on a

logits handled as float on v1.16-release

https://github.com/huggingface/optimum-habana/blob/v1.16-release/optimum/habana/transformers/models/mixtral/modeling_mixtral.py#L801

but float() is removed as comment now

https://github.com/huggingface/optimum-habana/blob/transformers_4_49/optimum/habana/transformers/models/mixtral/modeling_mixtral.py#L793

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 make sure to update the documentation with your changes?
• Did you write any new necessary tests?

[regisss]

LGTM, I thought I added it back to Mixtral too, good catch