[rlhf] support named placement group

[youkaichao]

v1 engine will create the engine in a separate process, and controlling the process placement becomes complicated.

suggested by the ray team, we can use https://docs.ray.io/en/latest/ray-core/scheduling/placement-group.html#advanced-named-placement-group , and pass the ray placement group name by an env var.

however, I find the named-placement-group does not work right now. it hangs when looking up the placement group.

steps to reproduce the hang:

$ cd examples/offline_inference
$ VLLM_USE_V1=1 python3 rlhf.py

it hangs when we try to get the placement group, i.e. the added line:

        current_placement_group = ray.util.get_placement_group(
            envs.VLLM_RAY_PG_NAME)

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

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 either: Add ready label to the PR or enable auto-merge.

🚀

[youkaichao]

without this pr, VLLM_USE_V1=1 python3 rlhf.py will hang in

vllm/vllm/executor/ray_utils.py

Line 310 in 8ca7a71

bundles = current_placement_group.bundle_specs

underlying hang is:

https://github.com/ray-project/ray/blob/8b26f5c13b7c66fb4e19094f9870083261ed9672/python/ray/util/placement_group.py#L139

[youkaichao]

right now, before the problem is fixed, for RLHF users, they need to use VLLM_USE_V1=1 VLLM_ENABLE_V1_MULTIPROCESSING=0 python3 rlhf.py .

[youkaichao]

close as it is finished in #14705