Remove torch_dtype in to() to end deprecation

[sayakpaul]

What does this PR do?

So that part of #6885 is fixed.

[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.

[DN6]

LGTM 👍🏽

[DN6]

Additionally we should probably update all the docs where we pass in torch_dtype

[sayakpaul]

Additionally we should probably update all the docs where we pass in torch_dtype

We cannot do that yet, because there's no deprecation behaviour in from_pretrained() for torch_dtype (both pipeline_utils.py and modeling_utils.py). So, that needs to be done first (should reside in a separate PR).

Let me know if that makes sense.

[DN6]

So this PR is just changing all references to torch_dtype then?

[sayakpaul]

Please review changes more carefully.

It’s deprecating torch_dtype inside to() along with torch_device.

[DN6]

Correct. But it would mean we should change all references to using to(torch_dtype= so that the slow tests will correctly set their dtypes right?

[sayakpaul]

That is taken care of :-)

[yiyixuxu]

thank you:)