[BugFix] Multi-gpu temp bug fix

[horheynm]

Signed-off-by: George Ohashi [email protected]

SUMMARY:
Bug caused by transformers >4.50.0, where in multi-gpu cases, a bug in data-parallel forward pass.

...
src/llmcompressor/transformers/finetune/session_mixin.py:289: in compute_loss
    loss = super().compute_loss(
venv/lib/python3.10/site-packages/transformers/trainer.py:3783: in compute_loss
    outputs = model(**inputs)
venv/lib/python3.10/site-packages/torch/nn/modules/module.py:1736: in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
venv/lib/python3.10/site-packages/torch/nn/modules/module.py:1747: in _call_impl
    return forward_call(*args, **kwargs)
venv/lib/python3.10/site-packages/torch/nn/parallel/data_parallel.py:183: in forward
    inputs, module_kwargs = self.scatter(inputs, kwargs, self.device_ids)
venv/lib/python3.10/site-packages/torch/nn/parallel/data_parallel.py:207: in scatter
    return scatter_kwargs(inputs, kwargs, device_ids, dim=self.dim)
...

Fix by monkey patching the test to use one GPU.
The other way is to use a lower transformers version.

Failures are seen here:
https://github.com/neuralmagic/llm-compressor-testing/actions/runs/14014229250/job/39237843197

TEST PLAN:
Pass nightly tests

[github-actions]

👋 Hi! Thank you for contributing to llm-compressor. Please add the ready label when the PR is ready for review.

Note: This is required to complete the testing suite, please only add the label once the PR is code complete and local testing has been performed.

[kylesayrs]

Does this bug affect anyone else? Are there any other related issues on transformers?

[dsikka]

Converting to draft as pinning to 4.49 temporarily