[Intel GPU] Ray Compiled Graph avoid NCCL for Intel GPU

[ratnampa]

Essential Elements of an Effective PR Description Checklist

• The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
• The test plan, such as providing test command.
• The test results, such as pasting the results comparison before and after, or e2e results
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.

Purpose

Ray compiled DAG causes runtime failure for PP+TP setup with XPU (Intel GPU). This is due to with_tensor_transport set to auto and further ray resolves this to NCCL as XPU device is a GPU, but inside ray compiled dag GPU-to-GPU transfers are only yet supported for NCCL NV GPUs. Setting to "shm" resolves the issue.

Test Plan

Test Result

(Optional) Documentation Update

[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.

🚀

[gemini-code-assist]

Code Review

This pull request correctly identifies and fixes a runtime failure for Intel GPUs when using Ray's compiled DAG by forcing the communication channel to 'shm'. I've suggested extending this fix to also include ROCm platforms, as they would likely suffer from the same issue, making the solution more robust.

[gemini-code-assist]

Thanks for this fix! Your reasoning that Ray's compiled DAG defaults to NCCL for any GPU is spot on.

Following this logic, it seems that ROCm (AMD GPUs) would also face the same runtime failure. To make this fix more comprehensive and prevent issues on ROCm, I suggest including it in the condition. This would also align with how XPU and ROCm are often handled together elsewhere in the codebase.

I've also updated the comment to reflect this.

Suggested change

	# For TPU or XPU, avoid compiling NVIDIA's NCCL
	if current_platform.is_tpu() or current_platform.is_xpu():
	# For TPU, XPU, or ROCm, avoid compiling NVIDIA's NCCL
	if current_platform.is_tpu() or current_platform.is_xpu() or current_platform.is_rocm():

[ratnampa]

I am not working with the AMD GPUs, so won't be able to test and make the requested changes.

[ratnampa]

@DarkLight1337 can I please get a review for this?