CI: publish Omni images to a separate Docker Hub repository

[sheralskumar]

Purpose

Point vLLM-Omni AMD hardware CI at a dedicated Docker Hub repository so Omni ROCm CI images are no longer published alongside core vLLM images under rocm/vllm-ci.

• Build/push: .buildkite/test-template-amd-omni.j2: tag and push rocm/vllm-omni:$BUILDKITE_COMMIT.
• Pull: .buildkite/scripts/hardware_ci/run-amd-test.sh: pull rocm/vllm-omni:${BUILDKITE_COMMIT} so test steps use the same image as the AMD build step.

The image tag is the commit SHA only (no -rocm-omni suffix), since the repository name already scopes Omni.

Test Plan

• CI: After merge, confirm a build that runs the AMD Buildkite path: the amd-build step builds and pushes to rocm/vllm-omni:<commit>, and subsequent AMD GPU steps successfully docker pull that image and run tests.

Test Result

• Run/observe AMD CI: build push + test pull both succeed against rocm/vllm-omni. (https://buildkite.com/vllm/vllm-omni-amd-ci/builds/5417/steps/canvas and https://hub.docker.com/repository/docker/rocm/vllm-omni/tags/3688ef1a6be0cb0b30f4edbb10092970195cb113/sha256-bc25394e7ccc453324febe5429306cfce3404bd4f16d25c0147c298e847b6566)

---

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. Please provide the test scripts & test commands. Please state the reasons if your codes don't require additional test scripts. For test file guidelines, please check the test style doc
• The test results. Please paste the results comparison before and after, or the e2e results.

[chatgpt-codex-connector]

Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits.
Credits must be used to enable repository wide code reviews.

[sheralskumar]

@tjtanaa would you please review? Thank you.

[tjtanaa]

Thanks. Let me run some tests with this new changes.

[hsliuustc0106]

CI-only change, straightforward string replacement from rocm/vllm-ci to rocm/vllm-omni repository.

The test result checkbox is still pending. Before merge, please confirm:

1. The new rocm/vllm-omni repository exists and is writable by the AMD CI pipeline
2. No existing CI jobs or external workflows depend on pulling from rocm/vllm-ci:commit-rocm-omni

The TODO comment about ECR Public Gallery should remain as a reminder for future migration.

[tjtanaa]

I will revisit this again tonight to check if https://status.canonical.com/ usptream is restored.

[tjtanaa]

LGTM @hsliuustc0106 @sheralskumar , the changes are validated, this is the build result https://buildkite.com/vllm/vllm-omni-amd-ci/builds/5417/steps/canvas

[lishunyang12]

LGTM. The change is consistent across both files:

• test-template-amd-omni.j2: The build step tags and pushes to rocm/vllm-omni:$BUILDKITE_COMMIT (was rocm/vllm-ci:$BUILDKITE_COMMIT-rocm-omni).
• run-amd-test.sh: The pull step uses rocm/vllm-omni:${BUILDKITE_COMMIT} (was rocm/vllm-ci:${BUILDKITE_COMMIT}-rocm-omni).

The repository rename properly scopes Omni images away from core vLLM, and dropping the -rocm-omni tag suffix makes sense since the repo name already disambiguates. CI evidence in the PR description confirms the pipeline works end-to-end.