Revert "[AMD][MI30X]Update Qwen3.5 perf (#986)" [skip-sweep]

[cquil11]

Reverts #986

MI355X script changes are preserved to retain PR #1036's subsequent fixes. Only the non-MI355X portion of #986 is reverted.

A re-open PR with the same contents will follow.

[github-actions]

Thanks for the contribution! For vLLM & SGLang, please ensure that your recipes is similar to the official vLLM recipes and/or the SGLang cookbook

• https://cookbook.sglang.io/
• https://docs.vllm.ai/projects/recipes/en/latest/

If it is not, please create a PR first before we can merge your PR into the master branch. Let's ensure that the documentation is first class such that the entire ML community can benefit from your hard work! Thank you

• https://github.com/vllm-project/recipes
• https://github.com/sgl-project/sgl-cookbook

PR authors are responsible for ensuring that after merging, all GitHub Action jobs fully pass. A lot of the time, failures are just flakes and simply re-running the failed jobs will fix it. If re-running failed jobs is attempted, PR authors are responsible for ensuring it passes. See GitHub's docs on re-running failed jobs: https://docs.github.com/en/actions/how-tos/manage-workflow-runs/re-run-workflows-and-jobs#re-running-failed-jobs-in-a-workflow

If additional help is needed, PR authors can reach out to core maintainers over Slack.

[github-actions]

Thanks for the contribution! For vLLM & SGLang, please ensure that your recipes is similar to the official vLLM recipes and/or the SGLang cookbook

• https://cookbook.sglang.io/
• https://docs.vllm.ai/projects/recipes/en/latest/

If it is not, please create a PR first before we can merge your PR into the master branch. Let's ensure that the documentation is first class such that the entire ML community can benefit from your hard work! Thank you

• https://github.com/vllm-project/recipes
• https://github.com/sgl-project/sgl-cookbook

PR authors are responsible for ensuring that after merging, all GitHub Action jobs fully pass. A lot of the time, failures are just flakes and simply re-running the failed jobs will fix it. If re-running failed jobs is attempted, PR authors are responsible for ensuring it passes. See GitHub's docs on re-running failed jobs: https://docs.github.com/en/actions/how-tos/manage-workflow-runs/re-run-workflows-and-jobs#re-running-failed-jobs-in-a-workflow

If additional help is needed, PR authors can reach out to core maintainers over Slack.

[claude]

LGTM — clean partial revert of PR #986 config/script changes for MI300X and MI325X, correctly preserving the MI355X changes from PR #1036.

Extended reasoning...

Overview

This PR reverts the non-MI355X portion of PR #986, touching 4 benchmark shell scripts (qwen3.5 bf16/fp8 for mi300x and mi325x), the amd-master.yaml config (downgrading sglang images from v0.5.10 back to v0.5.9 for those 4 configs), and the corresponding perf-changelog.yaml entry. MI355X configs are intentionally left alone, preserving fixes from PR #1036.

Security risks

None. Changes are limited to benchmark config/image versions and server CLI arguments (attention backend, memory fraction, radix cache flag). No auth, crypto, or permission-sensitive code is touched.

Level of scrutiny

Low. This is a mechanical revert of configuration and shell script parameters, scoped to a subset of the original PR. The intent matches the diff exactly: image versions roll back to v0.5.9, CLI args revert to the simpler triton/0.8 baseline, and the changelog entry is removed.

Other factors

No bugs were flagged by the automated system. No prior reviews exist on this PR. The PR description clearly explains the scope (partial revert, MI355X preserved), and a follow-up PR is planned.