[Feature] Add NPU Support for Single Agent

[lowdy1]

Description

Adds support for Ascend NPUs in most single-agent sota-implementations. These changes are non-intrusive to existing code and have been thoroughly tested on Ascend NPU devices.
As the next step, we are working to extend TorchRL NPU support to multi-agent experiments and GRPO, and we plan to submit a separate PR for this work soon.

Motivation and Context

We aim to extend TorchRL to enable execution on NPUs, providing broader hardware compatibility.
And our issue is #3154

• 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)
• 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/3229

• 📄 Preview Python docs built from this PR

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

❗ 1 Active SEVs

There are 1 currently active SEVs. If your PR is affected, please view them below:

• B200 runners are down due to network issues

❌ 3 New Failures, 1 Cancelled Job, 11 Unrelated Failures

As of commit 78a77ed with merge base 1ac0fd0 ():

NEW FAILURES - The following jobs have failed:

• Continuous Benchmark (PR) / CPU Pytest benchmark (gh)
  Workflow failed! Resource not accessible by integration
• Habitat Tests on Linux / tests (3.9, 12.8) / linux-job (gh)
  RuntimeError: Command docker exec -t 2247f4d3151d4466d26b2f1ddbcae430d48390e00eb4219b066e2abdf0532670 /exec failed with exit code 1
• Libs Tests on Linux / unittests-gym (3.9, 12.8) / linux-job (gh)
  RuntimeError: Command docker exec -t 261628e2925ec93e087f636efdd729dd09f22c11facf85606bcf8249d0d64912 /exec failed with exit code 1

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

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

BROKEN TRUNK - The following jobs failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

• Continuous Benchmark / GPU Pytest benchmark (gh) (trunk failure)
  Process completed with exit code 139.
• Unit-tests on Linux / tests-cpu (3.10) / linux-job (gh) (trunk failure)
  test/test_envs.py::TestLibThreading::test_num_threads
• Unit-tests on Linux / tests-cpu (3.11) / linux-job (gh) (trunk failure)
  test/test_envs.py::TestLibThreading::test_num_threads
• Unit-tests on Linux / tests-cpu (3.12) / linux-job (gh) (trunk failure)
  test/test_envs.py::TestLibThreading::test_num_threads
• Unit-tests on Linux / tests-cpu (3.13) / linux-job (gh) (trunk failure)
  test/test_envs.py::TestLibThreading::test_num_threads
• Unit-tests on Linux / tests-cpu (3.14) / linux-job (gh) (trunk failure)
  test/test_envs.py::TestLibThreading::test_num_threads
• Unit-tests on Linux / tests-gpu (3.12, 13.0) / linux-job (gh) (trunk failure)
  test/test_envs.py::TestLibThreading::test_num_threads
• Unit-tests on Linux / tests-olddeps (3.9, 11.6) / linux-job (gh) (trunk failure)
  ##[error]The operation was canceled.
• Unit-tests on Linux / tests-optdeps (3.12, 13.0) / linux-job (gh) (trunk failure)
  test/test_envs.py::TestLibThreading::test_num_threads
• Unit-tests on Linux / tests-stable-gpu (3.12, 13.0) / linux-job (gh) (trunk failure)
  test/test_envs.py::TestLibThreading::test_num_threads
• Unit-tests on Windows / unittests-cpu (3.10, windows.4xlarge, cpu) / windows-job (gh) (trunk failure)
  test/test_envs.py::TestLibThreading::test_num_threads

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

[vmoens]

Thanks for the contribution! I made several improvements to this PR:

1. Added hasattr guard for compatibility

Wrapped all torch.npu.is_available() calls with hasattr(torch, "npu") to ensure compatibility with PyTorch versions that don't have NPU support. This matches the pattern already used in torchrl/collectors/_single.py.

2. Created centralized get_available_device() utility

Instead of duplicating the device selection logic across 22 files, I created a new utility function in torchrl/_utils.py:

def get_available_device(return_str: bool = False) -> torch.device | str:
    """Return the available accelerator device, or CPU if none is found.

    Checks for accelerator availability in the following order: CUDA, NPU, MPS.
    Returns the first available accelerator, or CPU if none are present.

    .. note::
        PyTorch generally assumes a single accelerator type per system.
        Running with multiple accelerator types (e.g., both CUDA and NPU)
        is not officially supported. This function simply returns the first
        available accelerator it finds.
    """

This also adds MPS support for consistency. All 22 sota-implementation files now use:

device = torch.device(cfg.network.device) if cfg.network.device else get_available_device()

3. Added documentation

Added get_available_device to docs/source/reference/utils.rst.