Skip to content

[AMD][MI35X] 0514 GLM-5-fp8 mtp - #1377

Closed
1am9trash wants to merge 2 commits into
mainfrom
glm5fp8-mtp-mi355-sgl-0514
Closed

1am9trash wants to merge 2 commits into
mainfrom
glm5fp8-mtp-mi355-sgl-0514

Conversation

@1am9trash

Copy link
Copy Markdown
Collaborator

No description provided.

@github-actions

Copy link
Copy Markdown
Contributor

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

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

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.

1 similar comment
@github-actions

Copy link
Copy Markdown
Contributor

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

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

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 claude Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM — routine image bump, concurrency expansion, and removal of the now-unneeded transformers pip install since the new sglang image provides glm_moe_dsa support.

Extended reasoning...

Overview

This PR updates the GLM-5-FP8 MTP benchmark on AMD MI355X:

  • Bumps the sglang-rocm image from v0.5.10rc0-rocm720-mi35x-20260415 to v0.5.11-rocm720-mi35x-20260513 for glm5-fp8-mi355x-sglang-mtp in amd-master.yaml.
  • Expands the tp=4 search-space conc-end from 128 to 256 (for both isl=1024 and isl=8192).
  • In benchmarks/single_node/glm5_fp8_mi355x.sh, removes the runtime pip install of a specific transformers SHA (previously needed because the older image lacked glm_moe_dsa support) and adds --max-running-requests ${CONC} and --cuda-graph-max-bs ${CONC} to the sglang server launch.

Security risks

None. This is an isolated benchmark recipe / config change — no auth, crypto, permissions, or production-serving code path is touched. The image bump pulls a tagged sglang-rocm image from lmsysorg, consistent with neighboring entries.

Level of scrutiny

Low. This is a benchmark recipe update for a single AMD config, following the exact pattern of many similar entries in amd-master.yaml. The pip-install removal is well-motivated by the comment that previously explained the workaround. Adding --max-running-requests and --cuda-graph-max-bs tied to $CONC is a sensible tuning knob, and CONC is already validated via check_env_vars at the top of the script.

Other factors

The bug-hunting system found no issues. The diff is small (~10 lines net), mechanical, and self-contained to AMD MI355X benchmarking — failures would surface immediately in the AMD CI job and not affect other recipes.

@1am9trash 1am9trash added AMD full-sweep-enabled Full sweep with canary gate; matrix jobs run to completion despite failures labels May 14, 2026
@github-actions

Copy link
Copy Markdown
Contributor

2 similar comments
@github-actions

Copy link
Copy Markdown
Contributor

@github-actions

Copy link
Copy Markdown
Contributor

@1am9trash
1am9trash marked this pull request as draft May 14, 2026 12:12
@functionstackx

Copy link
Copy Markdown
Collaborator

stale PR

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

AMD full-sweep-enabled Full sweep with canary gate; matrix jobs run to completion despite failures

Projects

Status: Done

Development

Successfully merging this pull request may close these issues.

2 participants