[AMD][CI] Name extra-suite jobs after the ROCm stack they run

[michaelzhang-ai]

Motivation

#34204 swapped the AMD PR gate to ROCm 7.2 and, in pr-test-amd-extra.yml, made the job display names suppress the suffix for rocm720 and add -rocm700 for ROCm 7.0. That inverts the convention both callers use for their own jobs: pr-test-amd-rocm720.yml suffixes every job -rocm720, and pr-test-amd.yml leaves them bare.

Two consequences today:

• A ROCm 7.2 PR shows stage-b-test-1-gpu-small-amd-rocm720 (linux-mi300-1gpu-sglang) next to an unsuffixed extra-a-test-1-gpu-small-amd (linux-mi300-1gpu-sglang), even though both start the same ROCm 7.2 container.
• The six reusable-call job names changed stacks mid-history, so every name-keyed consumer starts a fresh series on 2026-08-12: query_job_status.py --job prefix reports (amd-ci-job-monitor.yml), ci_failures_analysis.py failure clusters, and the AMD nightly CI job matrix. The caller prefix (call-pr-test-amd-extra vs call-pr-test-amd-extra-rocm720) keeps 7.0 and 7.2 rows distinct, so nothing is conflated — but each of the six rows is orphaned from its own past.

Modifications

Suffix on the stack instead of on "is this the default": -rocm720 for ROCm 7.2, no suffix for ROCm 7.0. pull_request runs of the extra workflow get no inputs, so they resolve to the ROCm 7.2 default and are labelled -rocm720, matching the container the job actually starts.

caller	before #34204	on main today	this PR
pr-test-amd-rocm720.yml (7.2 gate)	extra-a-test-1-gpu-small-amd-rocm720 (…)	extra-a-test-1-gpu-small-amd (…)	extra-a-test-1-gpu-small-amd-rocm720 (…)
pr-test-amd.yml (7.0 shadow)	extra-a-test-1-gpu-small-amd (…)	extra-a-test-1-gpu-small-amd-rocm700 (…)	extra-a-test-1-gpu-small-amd (…)
direct pull_request (label-gated)	extra-a-test-1-gpu-small-amd (…), ROCm 7.0 container	extra-a-test-1-gpu-small-amd (…), ROCm 7.2 container	extra-a-test-1-gpu-small-amd-rocm720 (…), ROCm 7.2 container

Same for extra-a-test-1-gpu-large-amd and extra-a-test-2-gpu-large-amd. Both reusable-call rows go back to the names they carried before #34204, so their history reconnects.

Only job display names change; job ids, suite names (run_suite.py --suite extra-a-test-*-amd), the rocm_version inputs, and the containers are untouched. The one PR-visible change is the third row: the label-gated extra-a checks on a pull request gain the -rocm720 suffix. They are gated behind run-ci + run-ci-extra and so shouldn't be required checks, but worth confirming against branch protection before merge.

actionlint is clean on the file.

Accuracy Tests

N/A — CI job naming only.

Speed Tests and Profiling

N/A — CI job naming only.

Checklist

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.
• Follow the SGLang code style guidance.

  

---

CI States

Latest PR Test (Base): ✅ Run #31640177469
Latest PR Test (Extra): ❌ Run #31640177272