Buildkite hardware ci xpu test

[pi314ever]

PLEASE FILL IN THE PR DESCRIPTION HERE ENSURING ALL CHECKLIST ITEMS (AT THE BOTTOM) HAVE BEEN CONSIDERED.

Purpose

Adds hardware CI for Intel XPU via buildkite. Relies on #1162 for XPU docker.

Test Plan

BUILDKITE_COMMIT=test VLLM_VERSION=v0.17.0 ./.buildkite/scripts/hardware_ci/run-xpu-test.sh

Test Result

All tests pass on 8xB60.

---

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. Please run mkdocs serve to sync the documentation editions to ./docs.
• (Optional) Release notes update. If your change is user facing, please update the release notes draft.

BEFORE SUBMITTING, PLEASE READ https://github.com/vllm-project/vllm-omni/blob/main/CONTRIBUTING.md (anything written below this line will be removed by GitHub Actions)

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: b0713083a9

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[lishunyang12]

Straightforward CI script addition -- a few nits below, mostly around hardening and a stale comment.

[hsliuustc0106]

@vllm-omni-reviewer

[lishunyang12]

Looks good now, thanks for the quick turnaround on the fixes.

[pi314ever]

Recent changes are for more feature-based testing (testing for features that have passed for XPU on B60 nodes) as opposed to sweeping all available tests. For future PRs that enable certain features, tests should be added as they are enabled on XPU.

[xuechendi]

Thanks, @pi314ever , looks good to me.
@hsliuustc0106 , @gcanlin , may you take a look, thanks so much

[gcanlin]

Thanks, @pi314ever , looks good to me. @hsliuustc0106 , @gcanlin , may you take a look, thanks so much

We're upgrading to v0.17.0. Please wait #1639 merged, then rebase this PR based on v0.17.0. Feel free to ping me again when the rebasing is ready :)

[pi314ever]

@gcanlin Rebased!

[gcanlin]

LGTM. Just one non-blocking thought: I notice that we're introducing many config dispatch for platform in tests. Not sure whether we have the way to clean up them.

[xuechendi]

LGTM. Just one non-blocking thought: I notice that we're introducing many config dispatch for platform in tests. Not sure whether we have the way to clean up them.

Thanks for the review. Let's put this as an AR. Will follow up with a different PR.

I am think we can at least reuse existing one in platforms/xpu/stage_configs? So all these config can get tested as well?

Also, will also need to provide multiple folders under platforms/${HW}/stage_configs/ for different products which are with different device_memory_size, so config will also be different

[pi314ever]

I have fixed all of the conflicts. Here are the latest test results

core_model and xpu and B60:

advanced_model and xpu and B60:

Docker build time: 15s with cache, ~8 min without (vLLM base docker build included)
Total test time: 10 min (7 min core_model, 2 min advanced_model, 1 min aggregate overhead)

[xuechendi]

@congw729 @gcanlin @hsliuustc0106
We have validated the script in the prepared CI node, it takes less <10min to complete entire tests.
Please take a look and if it is OK

[hsliuustc0106]

currently, do we have access to the hardware to run the pipeline?