Fix get_device_name for cuda platforms that return bytes

[mgoin]

A more general fix for the proposed changes in #12565 and #12635

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.
Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can do one of these:

• Add ready label to the PR
• Enable auto-merge.

🚀

[hmellor]

Thanks for the more general fix!

[DarkLight1337]

@youkaichao can you take a look at this? As per #12635 (comment), I think it's better to tell users to upgrade their pynvml version instead of silently casting the type.

[hmellor]

We could do both? We fix it because we can, and tell the user that they should upgrade pynvml to prevent the warning.

[DarkLight1337]

I'm not sure whether there are other implications besides just the type issue when outdated pynvml is used. @youkaichao should know more.

[youkaichao]

the outdated pynvml package in general does not work, and I think we can be more aggressive to turn

vllm/vllm/platforms/cuda.py

Line 31 in 4f4d427

if pynvml.__file__.endswith("__init__.py"):

into an error.

[youkaichao]

a better solution is to make sure we are always using nvidia-ml-py, rather than pynvml, if we can.

I don't know if we can find some advanced python import usage to achieve it.