fix(_is_mslk_available): guard against missing CUDA/ROCm runtime

[space192]

Summary

_is_mslk_available() crashes with OSError when mslk is installed but no CUDA/ROCm runtime is present (e.g. CPU-only submission nodes).

The function is called at module level in torchao/quantization/quantize_/workflows/float8/float8_tensor.py, so the crash propagates through any transitive import chain that touches torchao — including transformers, vllm, etc. — even when the user never explicitly imports mslk or uses FP8.

Repro:

# On a CPU-only machine with mslk installed from the CUDA PyTorch index
python -c "import torchao"
# OSError: Could not load this library: .../mslk/mslk.so

Fix

Check torch.cuda.is_available() before attempting import mslk. This covers both CUDA and ROCm (HIP maps to the CUDA API surface). A warning is emitted so users know mslk was skipped and why.

Test plan

• Verified on a CPU-only node: import torchao succeeds with a warning instead of crashing.
• On GPU nodes: behavior is unchanged, mslk loads normally.

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/4544

• 📄 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.

[pytorch-bot]

The following ciflow label(s) have been added but CI has not been triggered yet because the workflows are awaiting approval:

• ciflow/rocm

Once a maintainer approves the workflows (scroll to the bottom of the PR page), the corresponding CI jobs will be triggered automatically. Please ping one of the reviewers if you do not have access to approve and run workflows.