[Bugfix] Add MPS float64->float32 downcast

[bsprenger]

Description

Add MPS float64 -> float32 downcast in _default_dtype_and_device() to prevent crashes when moving specs to MPS device.

MPS does not support float64 tensors. This is a problem e.g. when gymnasium/MuJoCo specs (which use float64) are created on CPU and later moved to MPS via .to("mps"), the spec constructors call torch.full(..., device="mps", dtype=torch.float64) which raises an error.

The fix adds a guard in _default_dtype_and_device() that automatically downcasts float64 -> float32 with a UserWarning when the device is MPS.

Motivation and Context

Gymnasium and MuJoCo environments produce float64 observation/action spaces by default. When a user creates a collector with env_device="mps" (e.g., in the a trainer), it will try to move the specs to MPS with float64, and crashes.

The fix prevents this crash by ensuring float64 specs are always downcast to float32 on MPS.

Fixes: #3549

• I have raised an issue to propose this change (required for new features and bug fixes)

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)

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/3548

• 📄 Preview Python docs built from this PR

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

❌ 4 New Failures, 1 Cancelled Job

As of commit fadc7b8 with merge base 491ed0f ():

NEW FAILURES - The following jobs have failed:

• Continuous Benchmark (PR) / GPU Pytest benchmark (gh)
  Workflow failed! Resource not accessible by integration
• Continuous Benchmark / CPU Pytest benchmark (gh)
  TypeError: expected Tensor()
• Generate documentation / build-docs (3.12, 12.8) / linux-job (gh)
  RuntimeError: Command docker exec -t a7b69d3173917de4aaa1ec896bfcca247487c9c7a206aeafb980c9618da68d44 /exec failed with exit code 2
• Unit-tests on Linux / tests-cpu (3.12) / linux-job (gh)
  test/test_custom_envs.py::TestPendulum::test_pendulum_env[device1]

CANCELLED JOB - The following job was cancelled. Please retry:

• Continuous Benchmark (PR) / CPU Pytest benchmark (gh)
  ##[error]The operation was canceled.

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

[vmoens]

LGTM! Thanks!
Maybe let's add a test to check that there is no float64 support with mps? Like that the test will fail if it ever becomes a thing and we know we can upgrade

[bsprenger]

@vmoens Thanks for the review, I added a test for that. Thanks!! 😄