Disable quick init for deepspeed

[muellerzr]

What does this PR do?

As @philschmid tested, this method works for FSDP but not for DS (I think because DS needs their gathered params to not be in this way). So this PR disables it

DS vs FSDP training logs:

DS:

{'loss': 2.5297, 'grad_norm': 166.13894215336276, 'learning_rate': 1.9230769230769234e-06, 'epoch': 0.01}                                                                                 
{'loss': 2.3812, 'grad_norm': 7.978569298066882, 'learning_rate': 1.923076923076923e-05, 'epoch': 0.12}                                                                                   
{'loss': 2.1066, 'grad_norm': 3.4303861731711813, 'learning_rate': 3.846153846153846e-05, 'epoch': 0.24}                                                                                  
{'loss': 2.0788, 'grad_norm': 3.1286244034212807, 'learning_rate': 4.99613632163459e-05, 'epoch': 0.36}    

FSDP (where this is enabled):

{'loss': 2.5297, 'grad_norm': 165.95199584960938, 'learning_rate': 1.9230769230769234e-06, 'epoch': 0.01}                                                                                 
{'loss': 2.3801, 'grad_norm': 7.723988056182861, 'learning_rate': 1.923076923076923e-05, 'epoch': 0.12}                                                                                   
{'loss': 2.1089, 'grad_norm': 3.3734328746795654, 'learning_rate': 3.846153846153846e-05, 'epoch': 0.24}                                                                                  
{'loss': 2.0727, 'grad_norm': 2.6789469718933105, 'learning_rate': 4.99613632163459e-05, 'epoch': 0.36} 

FSDP prior to the init PR/without quick load:

{'loss': 2.5297, 'grad_norm': 166.12049865722656, 'learning_rate': 1.9230769230769234e-06, 'epoch': 0.01}                                                                                 
{'loss': 2.3754, 'grad_norm': 5.313587665557861, 'learning_rate': 1.923076923076923e-05, 'epoch': 0.12}                                                                                   
{'loss': 2.1038, 'grad_norm': 3.9090209007263184, 'learning_rate': 3.846153846153846e-05, 'epoch': 0.24}                                                                                  
{'loss': 2.0933, 'grad_norm': 3.003643274307251, 'learning_rate': 4.99613632163459e-05, 'epoch': 0.36} 

It's close enough that it's reasonable IMO

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,
  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.

@SunMarc @amyeroberts

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[amyeroberts]

Thanks!