[tests] add the missing .eval() for inference

[faaany]

No description provided.

[BenjaminBossan]

Thanks for extending the tests and making them XPU compatible. I found one small issue, the rest looks good.

Regarding the testing, did you run the bnb regression tests and the LoftQ tests successfully on XPU?

[faaany]

Thanks for extending the tests and making them XPU compatible. I found one small issue, the rest looks good.

Regarding the testing, did you run the bnb regression tests and the LoftQ tests successfully on XPU?

No, not all bnb and LoftQ tests pass. We only focus on cases that pass on CUDA but fail on XPU. If they fail on CUDA, we are not trying to fix them now.

[BenjaminBossan]

No, not all bnb and LoftQ tests pass. We only focus on cases that pass on CUDA but fail on XPU. If they fail on CUDA, we are not trying to fix them now.

Could you please share which CUDA tests are failing? Normally, that shouldn't happen.

[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.

[BenjaminBossan]

@faaany I merged your other PR. I think this one has some overlap with it, could you please rebase on/merge with the latest main branch? Then I'll do a final review.

[BenjaminBossan]

Thanks for adding the missing .eval() calls to the tests.