[BugFix] Use torch.zeros for argument in torch.where

[sebimarkgraf]

Description

This reorders the function calls to make the device lookup, right before usage.
Also, this uses torch.zeros_like as argument for the torch.where call instead
of relying on autocasting.

Motivation and Context

We are observing issues to export the module into ONNX on intel devices.
This seems to be due to the tracer having issues to match the inserted constant for the 0 to constant zero node autocast.

Error:

<class 'AssertionError'>: Expected same OpOverload packet, got prim.device != aten.where
While executing %where : [num_users=1] = call_function[target=torch.ops.aten.where.ScalarSelf](args = (%expand, %scalar_tensor_default, %unsqueeze_2), kwargs = {})

Interestingly this issue only appears intermittently, we believe due to our CI landing on intel cpus with specific features activated.
Therefore, a direct reproduction is hard to detail.

Types of changes

What types of changes does your code introduce? Remove all that do not apply:

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds core functionality)
• Breaking change (fix or feature that would cause existing functionality to change)
• Documentation (update in the documentation)
• Example (update in the folder of examples)

Checklist

Go over all the following points, and put an x in all the boxes that apply.
If you are unsure about any of these, don't hesitate to ask. We are here to help!

• I have read the CONTRIBUTION guide (required)
• My change requires a change to the documentation.
• I have updated the tests accordingly (required for a bug fix or a new feature).
• I have updated the documentation accordingly.

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/rl/3239

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[vmoens]

LGTM thanks