[AMD][MI35X] 0521 DSV4

[1am9trash]

Bump docker image to rocm/sgl-dev:rocm720-mi35x-8c3b5aa-20260521-DSv4.
No server args change.

Succecuss run: https://github.com/SemiAnalysisAI/InferenceX/actions/runs/26211414795
Chart: https://inferencex.semianalysis.com/inference?unofficialRun=26211414795

[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://docs.sglang.io/cookbook/intro
• https://recipes.vllm.ai/

If it is not, please create a PR first before we can merge your single node 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/sglang/tree/main/docs_new

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

As a rule of thumb, generally, PR authors should request a review & get a PR approval from the respective companies' CODEOWNERS before requesting a review from core maintainers.

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

[claude]

LGTM — straightforward image tag bump for DSv4 sglang config with minor concurrency range tuning.

Extended reasoning...

Overview

This PR makes a small, self-contained change to a single file: .github/configs/amd-master.yaml. It bumps the docker image tag for the dsv4-fp4-mi355x-sglang benchmark recipe from rocm/sgl-dev:rocm720-mi35x-b19052c-20260518-DSv4 to rocm/sgl-dev:rocm720-mi35x-8c3b5aa-20260521-DSv4, and slightly tweaks the concurrency search space (raising dp-attn: true start from 32 to 64, and lowering dp-attn: false end from 64 to 32) for both ISL=1024 and ISL=8192 scenarios.

Security risks

None. This is a YAML config change scoped to a single benchmark sweep entry in a CI sweep configuration. No code, no auth, no secrets, no infrastructure changes.

Level of scrutiny

Low. Image tag bumps for individual benchmark recipes are routine and well-understood in this repo (the comment block above the entry explicitly notes that bumping sglang is just an image tag bump). The concurrency range adjustments are mechanical tuning of the search space and only affect what configurations get swept — they cannot introduce correctness regressions outside this single recipe.

Other factors

The change is isolated to one recipe block; nothing else in the config is touched. The bug hunter agent found nothing. The PR is labeled full-sweep-enabled so the sweep itself will validate the bump end-to-end.

[github-actions]

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=26211414795
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=26211414795

[github-actions]

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=26211414795
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=26211414795

[github-actions]

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=26211414795
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=26211414795

[chunfangamd]

Thanks for the work @1am9trash!