Skip to content

[AMD/ROCM] Update minimaxm2.5-fp8-mi355x-atom config#1194

Merged
seungrokj merged 2 commits intomainfrom
srok/atom_minimaxm2.5_fp8
May 3, 2026
Merged

[AMD/ROCM] Update minimaxm2.5-fp8-mi355x-atom config#1194
seungrokj merged 2 commits intomainfrom
srok/atom_minimaxm2.5_fp8

Conversation

@seungrokj
Copy link
Copy Markdown
Collaborator

@seungrokj seungrokj commented Apr 27, 2026 •
edited
Loading

Hi, This is just a minor change. Will do sweep after high priority dsv4 stuffs.

Summary

  • Update Atom image to rocm7.2.2_ubuntu24.04_py3.12_pytorch_release_2.10.0_atom0.1.2.post
  • Update search-space: expand conc-end to 256, remove tp8/ep8 configs
  • Add perf-changelog entry for minimaxm2.5-fp8-mi355x-atom

Test plan

  • CI benchmark run for minimaxm2.5-fp8-mi355x-atom

🤖 Generated with Claude Code

@seungrokj seungrokj requested a review from a team April 27, 2026 12:01
@seungrokj seungrokj changed the title Update minimaxm2.5-fp8-mi355x-atom config [AMD/ROCM] Update minimaxm2.5-fp8-mi355x-atom config Apr 27, 2026
@seungrokj seungrokj added the AMD label Apr 27, 2026
claude[bot]
claude Bot reviewed Apr 27, 2026
View reviewed changes
Comment thread perf-changelog.yaml Outdated
Comment thread .github/configs/amd-master.yaml
chunfangamd
chunfangamd approved these changes Apr 30, 2026
View reviewed changes
Copy link
Copy Markdown
Collaborator

@chunfangamd chunfangamd left a comment

Choose a reason for hiding this comment

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

lgtm

functionstackx
functionstackx approved these changes Apr 30, 2026
View reviewed changes
Copy link
Copy Markdown
Contributor

@functionstackx functionstackx left a comment

Choose a reason for hiding this comment

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

lgtm as long as u implement the small change on perf-changelog.yaml (probably claude can fix it for u) & validation run fully passes

Comment thread perf-changelog.yaml Outdated
@seungrokj
Copy link
Copy Markdown
Collaborator Author

seungrokj commented May 1, 2026

e2e perf/accuracy
https://github.com/SemiAnalysisAI/InferenceX/actions/runs/25196734216

@functionstackx
Copy link
Copy Markdown
Contributor

functionstackx commented May 1, 2026

e2e perf/accuracy https://github.com/SemiAnalysisAI/InferenceX/actions/runs/25196734216

thanks! plz ping on slack when it is all green

@github-actions
Copy link
Copy Markdown
Contributor

github-actions Bot commented May 2, 2026

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=25244154712
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=25244154712

@functionstackx
Copy link
Copy Markdown
Contributor

functionstackx commented May 2, 2026

@claude rebase this PR & resolve conflicts. note the recent changes to the amd-master.yaml schema

@Klaud-Cold
Copy link
Copy Markdown
Collaborator

Klaud-Cold commented May 2, 2026 •
edited
Loading

Claude finished @functionstackx's task in 3m 35s —— View job

Rebased PR onto main & resolved conflicts

  • Gather context: understand current branch state and recent main changes
  • Understand amd-master.yaml schema changes on main
  • Reset branch to origin/main and re-apply intended changes
  • Fix perf-changelog.yaml (populated description, correct PR link [AMD/ROCM] Update minimaxm2.5-fp8-mi355x-atom config #1194, trailing newline)
  • Push rebased branch

What was done

The branch had no common merge base with origin/main (main was squash-reset), so a traditional git rebase couldn't work. Instead, I reset the branch to origin/main and re-applied the 3 intended changes as a single clean commit:

amd-master.yaml — minimaxm2.5-fp8-mi355x-atom:

  • Updated image to rocm7.2.2_ubuntu24.04_py3.12_pytorch_release_2.10.0_atom0.1.2.post
  • Fixed schema: now correctly uses scenarios: fixed-seq-len: (branch previously had the old seq-len-configs: format)
  • Expanded conc-end to 256 for tp2/tp4
  • Removed tp8/ep8 configs
  • No longer carries stale gptoss-fp4-mi355x-atom regression (the Fix GPT-OSS ATOM config schema #1261 fix is preserved)

minimaxm2.5_fp8_mi355x_atom.sh:

  • Added MEM_FRAC_STATIC=0.9 and --gpu-memory-utilization

perf-changelog.yaml:

|

@github-actions @functionstackx
Update minimaxm2.5-fp8-mi355x-atom config
22012da 
- Update Atom image to rocm7.2.2_ubuntu24.04_py3.12_pytorch_release_2.10.0_atom0.1.2.post
- Expand conc-end to 256 for tp2/tp4, remove tp8/ep8 configs
- Add gpu-memory-utilization 0.9 to benchmark script
- Add perf-changelog entry

Co-authored-by: functionstackx <functionstackx@users.noreply.github.com>
@Klaud-Cold Klaud-Cold force-pushed the srok/atom_minimaxm2.5_fp8 branch from 16441a1 to 22012da Compare May 2, 2026 19:07
@github-actions
Copy link
Copy Markdown
Contributor

github-actions Bot commented May 2, 2026

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=25259578554
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=25259578554

@github-actions
Copy link
Copy Markdown
Contributor

github-actions Bot commented May 3, 2026

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=25271940165
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=25271940165

@seungrokj seungrokj merged commit c790ce5 into main May 3, 2026
25 checks passed
@seungrokj seungrokj deleted the srok/atom_minimaxm2.5_fp8 branch May 3, 2026 15:38
xiaohuguo2023 pushed a commit to xiaohuguo2023/InferenceX that referenced this pull request May 6, 2026
@seungrokj @github-actions
@functionstackx @chunfangamd @xiaohuguo2023
Update minimaxm2.5-fp8-mi355x-atom config (SemiAnalysisAI#1194)
e19f6e7 
- Update Atom image to rocm7.2.2_ubuntu24.04_py3.12_pytorch_release_2.10.0_atom0.1.2.post
- Expand conc-end to 256 for tp2/tp4, remove tp8/ep8 configs
- Add gpu-memory-utilization 0.9 to benchmark script
- Add perf-changelog entry

Co-authored-by: claude[bot] <41898282+claude[bot]@users.noreply.github.com>
Co-authored-by: functionstackx <functionstackx@users.noreply.github.com>
Co-authored-by: Chun Fang <chun.fang@amd.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@claude claude[bot] claude[bot] left review comments

@functionstackx functionstackx functionstackx approved these changes

@chunfangamd chunfangamd chunfangamd approved these changes

@billishyahao billishyahao Awaiting requested review from billishyahao billishyahao is a code owner

@yctseng0211 yctseng0211 Awaiting requested review from yctseng0211 yctseng0211 is a code owner

@1am9trash 1am9trash Awaiting requested review from 1am9trash 1am9trash is a code owner

Assignees

No one assigned

Labels

AMD sweep-enabled

Projects

InferenceMAX Board
Status: Done

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

4 participants

@seungrokj @functionstackx @Klaud-Cold @chunfangamd