[AMD] Qwen3.5-FP8 MI355X SGLang disagg baseline: image bump to v0.5.16, TP4P+TP8D / Qwen3.5-FP8 MI355X SGLang disagg 基线优化：镜像升级至v0.5.16，TP4P+TP8D拓扑优化 - #2606
Conversation
…TP4P+TP8D - Bump qwen3.5-fp8-mi355x-sglang-disagg image from v0.5.14 to v0.5.16 (3-5% throughput improvement). - Switch prefill from TP8 to TP4: saves 4 GPUs (12 vs 16 total) while delivering identical decode interactivity and 24-31% better throughput/GPU. - Infra: add docker_sg.sh wrapper for nodes where Slurm doesn't activate the docker supplementary group; update job.slurm with pre-pull and kill -9 fallback for stuck containers; update submit.sh node excludes. Co-authored-by: Cursor <cursoragent@cursor.com>
|
Thanks for the contribution! Please reach out to respective companies' CODEOWNER to fill in the latest PR_REVIEW_CHECKLIST.md before pinging core maintainer on Slack for review. In order for the signoff PR check bot to trigger, you must follow the PR_REVIEW_CHECKLIST.md template correctly, including the phrase For PR verification, add the 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. See GitHub's docs on re-running failed jobs 感谢你的贡献！请联系相应公司的 CODEOWNER 填写最新的 PR_REVIEW_CHECKLIST.md，然后再在 Slack 上联系核心维护者进行审阅。为了触发 signoff PR 检查机器人，你必须正确遵循 PR_REVIEW_CHECKLIST.md 模板，包括保留英文语句 如需进行 PR 验证，请为此 PR 添加 PR 作者有责任确保合并后所有 GitHub Action 任务完全通过。 很多时候失败只是偶发抖动（flake），重新运行失败的任务即可解决。参见 GitHub 关于重新运行失败任务的文档 |
2 similar comments
|
Thanks for the contribution! Please reach out to respective companies' CODEOWNER to fill in the latest PR_REVIEW_CHECKLIST.md before pinging core maintainer on Slack for review. In order for the signoff PR check bot to trigger, you must follow the PR_REVIEW_CHECKLIST.md template correctly, including the phrase For PR verification, add the 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. See GitHub's docs on re-running failed jobs 感谢你的贡献！请联系相应公司的 CODEOWNER 填写最新的 PR_REVIEW_CHECKLIST.md，然后再在 Slack 上联系核心维护者进行审阅。为了触发 signoff PR 检查机器人，你必须正确遵循 PR_REVIEW_CHECKLIST.md 模板，包括保留英文语句 如需进行 PR 验证，请为此 PR 添加 PR 作者有责任确保合并后所有 GitHub Action 任务完全通过。 很多时候失败只是偶发抖动（flake），重新运行失败的任务即可解决。参见 GitHub 关于重新运行失败任务的文档 |
|
Thanks for the contribution! Please reach out to respective companies' CODEOWNER to fill in the latest PR_REVIEW_CHECKLIST.md before pinging core maintainer on Slack for review. In order for the signoff PR check bot to trigger, you must follow the PR_REVIEW_CHECKLIST.md template correctly, including the phrase For PR verification, add the 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. See GitHub's docs on re-running failed jobs 感谢你的贡献！请联系相应公司的 CODEOWNER 填写最新的 PR_REVIEW_CHECKLIST.md，然后再在 Slack 上联系核心维护者进行审阅。为了触发 signoff PR 检查机器人，你必须正确遵循 PR_REVIEW_CHECKLIST.md 模板，包括保留英文语句 如需进行 PR 验证，请为此 PR 添加 PR 作者有责任确保合并后所有 GitHub Action 任务完全通过。 很多时候失败只是偶发抖动（flake），重新运行失败的任务即可解决。参见 GitHub 关于重新运行失败任务的文档 |
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
There was a problem hiding this comment.
Beyond the two inline findings, a few other candidate issues were checked and ruled out: docker_sg.sh's base64 round-trip through sg docker -c (the ${__A[@]} expansion is correctly scoped to the inner bash -c, not the outer shell); the pre-clean fallback's sudo kill -9 on docker ps -aq output (scoped to this job's own containers, not arbitrary host PIDs); and whether TP4 prefill actually shrinks the Slurm reservation (it doesn't — Slurm allocates whole 8-GPU nodes regardless of TP size, so the 12-vs-16-GPU savings is a utilization claim about the workload, not the job's node/GPU request).
Extended reasoning...
This is an informational note, not a full review. This PR is not being approved because it modifies configs/amd-master.yaml, which is CODEOWNER-protected (per .github/CODEOWNERS, that file requires sign-off from the listed AMD owners), and because the bug hunting system already found two nit-level issues that are posted as inline comments. Beyond those, I checked three additional candidate concerns raised by finder agents and confirmed they don't hold up: the docker_sg.sh base64/sg-docker argv relay is correctly scoped so ${__A[@]} expands inside the inner bash -c rather than the outer shell; the stuck-container SIGKILL fallback in job.slurm operates on docker ps -aq output which is already scoped to containers on the Docker daemon reachable via $DOCKER_CMD (this job's own containers), not arbitrary host PIDs; and the TP4-prefill GPU savings described in the PR is a compute-utilization claim, not a change to the Slurm node/GPU reservation (which is fixed per-node at --gres=gpu:8 regardless of TP size) -- expected behavior, not a bug.
ChangLiu0709
changed the title
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=31821883369 |
|
Claude finished @chunfangamd's task in 2m 5s —— View job Review: PR #2606
LGTM - no blocking issues found. The PR is now cleanly scoped to just the two files (the earlier
Note: since |
There was a problem hiding this comment.
LGTM — approving this baseline config change.
What was reviewed: the SGLang disagg TP topology change (TP8P+TP8D → TP4P+TP8D) and image bump in configs/amd-master.yaml, plus the new perf-changelog.yaml entry. The two nits I raised on the prior revision are resolved: the pr-link now correctly points to #2606 (was #2557), and the job.slurm/docker_sg.sh infra changes I flagged for duplication are no longer part of this PR's diff (only the two config files remain). The GPU-savings claim for TP4 prefill under exclusive SLURM allocation was also checked and holds.
Extended reasoning...
Overview
This revision of the PR is now scoped to exactly two files: configs/amd-master.yaml (SGLang image bump v0.5.14→v0.5.16, and prefill TP8→TP4 for the qwen3.5-fp8-mi355x-sglang-disagg baseline) and perf-changelog.yaml (one new append-only changelog entry documenting the change). The previously-reviewed job.slurm/docker_sg.sh/submit.sh infra changes are no longer present in this diff — they appear to have been split out, so my earlier duplication nit on job.slurm no longer applies to this PR.
Security risks
None. This is a YAML config change to benchmark topology parameters and a changelog entry; no code execution paths, credentials, or user input are touched.
Level of scrutiny
Low. This is a mechanical, low-blast-radius change: a pinned image tag bump and a numeric TP parameter change in a benchmark config, backed by the author's own local benchmark table in the PR description. It doesn't touch shared infra, security-sensitive code, or CODEOWNER-restricted paths beyond the standard config review process already noted by the recipe-reminder bot.
Other factors
Both nits I raised on the earlier revision are now resolved: the pr-link field was corrected from pull/2557 to pull/2606 (matching this PR's number, consistent with sibling changelog entries), and the job.slurm duplication concern is moot since that file isn't part of this PR's diff anymore. The bug-hunting system found no bugs this run, and separately ruled out a candidate concern about whether the TP4 prefill GPU savings claim holds under exclusive whole-node SLURM allocation.
|
see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=32082978745 |
|
/reuse-sweep-run |
chunfangamd
left a comment
chunfangamd
left a comment
There was a problem hiding this comment.
As a PR reviewer and CODEOWNER, I have reviewed this and have:
- Verified that as of the moment of typing this, this is the latest version of PR_REVIEW_CHECKLIST.md
- Verified that the general code quality meets the InferenceX standard and does not make the code quality any worse.
- Verified that this PR has passed PR validation. Please link to GitHub Action workflow that shows this. Link: https://github.com/SemiAnalysisAI/InferenceX/actions/runs/32082978745
- Verified that this PR passes evals. Please link to GitHub Action workflow that shows this. Value: 97.35%; Link: https://github.com/SemiAnalysisAI/InferenceX/actions/runs/32082978745
- Verified that speculative decoding PRs uses chat templates to align the AL distribution to real world
- For agentic workloads: verified that speculative-decoding configs (EAGLE / MTP / draft models) run with simulated synthetic acceptance, with the acceptance-length value taken from the committed golden AL curve in golden_al_distribution/ for that model, thinking mode, and draft length. A submission may choose any supported draft length, but it may not substitute a different acceptance target.
- Verified against the current MODELS.md that this PR does not submit a deprecated model, scenario, or model-scenario combination.
- Verified that the model architecture isn't changed with benchmark hacks like using --hf-overrides to skipping indexer for every x layers on models that don't natively support this. As a general rule, we won't accept optimizations that reduces the number of model architecture FLOPs. Anything that makes that same computation run faster is fair game; FLOPs at lower precisions is fine, given that the config passes private evals. As an general north star princple, we should only use optimizations which is used in production by customers that care about accuracy
- If an company claims that they support vLLM/SGLang as first class LLM inference engines on their hardware, I have verified that the respective vLLM submission made using upstream https://hub.docker.com/u/vllm docker repo, upstream SGLang https://hub.docker.com/u/lmsysorg docker repo. The only exceptions are for new hardware, such as MI455X UALoE72, Vera Rubin NVL72, Rubin NVL8, etc., and for new model architectures where there is an actual reason why vLLM/SGLang does not fundamentally support them yet as supported by vLLM/SGLang community maintainers
- If an company claims that they support vLLM/SGLang as first class upstream in-tree LLM inference engines on their hardware, I have have verified that the respective vLLM/SGLang submission has been made before additional frameworks (TRT-LLM, ATOM, etc.). The only exceptions are for new hardware, such as MI455X UALoE72, Vera Rubin NVL72, Rubin NVL8, etc., and for new model architectures where there is an actual reason why vLLM/SGLang does not fundamentally support them yet.
- Verified that every single-node vLLM/SGLang recipe in this PR is documented in the official vLLM recipes and/or the SGLang cookbook:
- I linked the corresponding upstream PR in the vLLM recipe repo or SGLang repo and verified that it is MERGED before this InferenceX PR merges. An opened, draft, or closed-without-merge upstream PR does not satisfy this requirement. If the matching recipe was already published, I linked the published recipe/cookbook page in the additional detail section below.
- Verified that this PR does not patch the inference engine or serving stack — the pinned image must run as shipped. This covers .patch files / git apply / patch, inline patches embedded in benchmark scripts (e.g. a python3/sed heredoc that rewrites installed engine sources before serving), in-place edits of site-packages, monkey-patching, overwriting container files, and installing forked/rebuilt engine wheels on top of the pinned image. The only exception is a patch covered by a filled-out waiver at docs/waiver/
<PR_NUMBER>.md— named after the PR that introduces the patch and filed in that same PR, stating what is patched, why the unmodified upstream image cannot run this benchmark, the upstream PR/issue link, and the removal plan — which I have linked below in the additional detail section. - If this PR uses
append-only: true, verified that it only adds generated points or recipe variants inside a selected existing config/scenario and existing same-image visual curve: every previously generated point remains present with the same recipe, no prior point is removed or rerun, and every benchmark-affecting change in the complete diff can affect only the corresponding newly appended points (never an existing point), regardless of which file contains it. - If any of the above criteria cannot reasonably be satisfied, I have provided additional reasoning below.
Additional detail section:
- insert any additional info here
Signed: @chunfangamd
✅✅✅ Verdict: PASS ✅✅✅✅ Check 0 (CODEOWNER): PASS — @chunfangamd is a listed owner of |
|
/reuse-sweep-run |
# Conflicts: # perf-changelog.yaml
4 participants
Changes
Baseline (
qwen3.5-fp8-mi355x-sglang-disagg)lmsysorg/sglang:v0.5.14-rocm720-mi35x→lmsysorg/sglang:v0.5.16-rocm720-mi35x(ROCm 7.2, SGLang v0.5.14 → v0.5.16, 3-5% throughput improvement across all concurrency levels).Infra
dockersupplementary group viasg docker -cfor Slurm job steps where group credentials are not inherited.DOCKER_CMD_DETECTto trysg dockerfallback beforesudo docker; add pre-pull on all nodes; add kill -9 fallback for stuck containers.Local benchmark results (v0.5.16, ISL=8K/OSL=1K)
Baseline (TP4P+TP8D, 12 GPU):
Split from #2557 — baseline only, no MTP/speculative decoding changes. MTP config in #2607.
Authors
@ChangLiu0709
@xiaohuguo2023
@chunfangamd
中文说明
Qwen3.5-FP8 MI355X SGLang 分离式推理基线优化：
从 #2557 拆分 — 仅包含基线配置变更，不包含 MTP 投机解码。MTP 配置见 #2607。
Made with Cursor