make sure to disable gradients for integer tensor

[winglian]

trafficstars

What does this PR do?

When doing 8bit lora's with deepspeed zero3, the Int8Params should have requires_grad=False otherwise the following error occurs:

[rank0]:   File "/root/miniconda3/envs/py3.11/lib/python3.11/site-packages/bitsandbytes/nn/modules.py", line 574, in __new__
[rank0]:     obj = torch.Tensor._make_subclass(cls, data, requires_grad)
[rank0]:           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank0]: RuntimeError: Only Tensors of floating point and complex dtype can require gradients

this used to work several months ago

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]

cc @ArthurZucker for final :)

[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.

[winglian]

@ArthurZucker @muellerzr I rebased this against latest transformers and tested on my end