[AMD][AgentX] MINIMAX-M3 FP4 MI355X agentX vLLM by ajith-sirra-amd · Pull Request #2118 · SemiAnalysisAI/InferenceX

ajith-sirra-amd · 2026-07-08T11:36:51Z

Summary

Add MiniMax-M3 FP4 single-node agentic benchmark support on MI355X using vLLM
New script: benchmarks/single_node/agentic/minimaxm3_fp4_mi355x.sh with three KV offload backends: none , native (vLLM OffloadingConnector), lmcache (LMCache MP server + LMCacheMPConnector)
New master config entry minimaxm3-fp4-mi355x-vllm-agentic with TP4 search space across none, native, and lmcache backends
Image: vllm/vllm-openai-rocm:nightly-69715823df89b11ee684b84066390cbb9092d5c1
Model: amd/MiniMax-M3-MXFP4

Signed-off-by: ajith-sirra-amd <ajith.sirra@amd.com>

github-actions · 2026-07-08T11:37:00Z

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 As a PR reviewer and CODEOWNER, I have reviewed this and have.

For PR verification, add the full-sweep-fail-fast label (strongly recommended) to this PR — the benchmark sweep only runs on labeled PRs. Use full-sweep-enabled only if you need matrix jobs to keep running past a failure.

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 As a PR reviewer and CODEOWNER, I have reviewed this and have。

如需进行 PR 验证，请为此 PR 添加 full-sweep-fail-fast 标签（强烈推荐）— 基准测试 sweep 仅在带有标签的 PR 上运行。仅当需要矩阵任务在失败后继续运行时才使用 full-sweep-enabled。

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

github-actions · 2026-07-08T11:37:00Z

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 As a PR reviewer and CODEOWNER, I have reviewed this and have.

For PR verification, add the full-sweep-fail-fast label (strongly recommended) to this PR — the benchmark sweep only runs on labeled PRs. Use full-sweep-enabled only if you need matrix jobs to keep running past a failure.

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 As a PR reviewer and CODEOWNER, I have reviewed this and have。

如需进行 PR 验证，请为此 PR 添加 full-sweep-fail-fast 标签（强烈推荐）— 基准测试 sweep 仅在带有标签的 PR 上运行。仅当需要矩阵任务在失败后继续运行时才使用 full-sweep-enabled。

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

github-actions · 2026-07-08T11:37:01Z

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 As a PR reviewer and CODEOWNER, I have reviewed this and have.

For PR verification, add the full-sweep-fail-fast label (strongly recommended) to this PR — the benchmark sweep only runs on labeled PRs. Use full-sweep-enabled only if you need matrix jobs to keep running past a failure.

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 As a PR reviewer and CODEOWNER, I have reviewed this and have。

如需进行 PR 验证，请为此 PR 添加 full-sweep-fail-fast 标签（强烈推荐）— 基准测试 sweep 仅在带有标签的 PR 上运行。仅当需要矩阵任务在失败后继续运行时才使用 full-sweep-enabled。

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

Signed-off-by: ajith-sirra-amd <ajith.sirra@amd.com>

…& Fixing MODEL_PATH Arg in Serve Command. Signed-off-by: ajith-sirra-amd <ajith.sirra@amd.com>

Signed-off-by: ajith-sirra-amd <ajith.sirra@amd.com>

github-actions · 2026-07-08T16:50:03Z

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=28942489112
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=28942489112

…hmark Drop lmcache case branch, helper functions, and config entry; keep native-only offload path. Add OFFLOAD_ARGS fallback, enable prefix caching and hybrid KV cache manager, bump max-num-seqs to 2x CONC. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

github-actions · 2026-07-09T03:34:46Z

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=28987547078
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=28987547078

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Signed-off-by: ajith-sirra-amd <ajith.sirra@amd.com>

…nalysisAI/InferenceX into amd/agentx-minimax-m3-vllm

github-actions · 2026-07-09T15:17:24Z

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=29016569168
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=29016569168

github-actions · 2026-07-09T15:54:31Z

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=29016569168
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=29016569168

github-actions · 2026-07-09T20:07:36Z

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=29031361208
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=29031361208

github-actions · 2026-07-10T00:06:51Z

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=29031361208
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=29031361208

github-actions · 2026-07-10T02:22:08Z

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=29031361208
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=29031361208

seungrokj

LGTM

cquil11 · 2026-07-10T16:41:47Z

Needs another run.

cquil11 · 2026-07-10T17:33:41Z

@claude fix the formatting in the yaml on this PR

Klaud-Cold · 2026-07-10T17:34:07Z

Claude finished @cquil11's task in 1m 23s —— View job

Fixed YAML formatting

Inspect YAML changes in this PR
Fix formatting issues
Validate YAML parses
Commit and push

What was wrong: This branch had accidentally added one extra space of indentation to the additional-settings block of the unrelated dsv4-fp4-mi355x-sglang-disagg-agentic-hicache entry (the decode section, configs/amd-master.yaml:3170-3172). That made the entire configs/amd-master.yaml fail to parse (yaml.scanner.ScannerError: mapping values are not allowed here, line 3170), which would have broken every sweep driven by that config file.

Fix: Restored the original 10-space indentation so the block matches main exactly. Pushed as e8ff5ba. Both configs/amd-master.yaml and perf-changelog.yaml now parse cleanly, and the PR diff vs main is back to purely additive (the new minimaxm3-fp4-mi355x-vllm-agentic entry, the new script, and the changelog entry).
· Branch

The extra leading space broke YAML parsing of amd-master.yaml. Co-authored-by: Cameron Quilici <60715037+cquil11@users.noreply.github.com> Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

github-actions · 2026-07-10T17:36:57Z

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=29111506959
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=29111506959

functionstackx

thanks for the contribution @ajith-sirra-amd @seungrokj

can u please update recipes.vllm.ai with ur new configs?

vllm recipes doesn't currently support offloading tho the inferact guys said they are open to contributions to support mooncake store, lmcache, vllm native offloading support in vllm recipes

as in sglang cookbook recipes , they also have support for cpu offloading recipes

github-actions · 2026-07-10T22:32:20Z

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=29111547629
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=29111547629

[AMD] MINIMAX-M3 FP4 vLLM Agentic Support

a007c03

Signed-off-by: ajith-sirra-amd <ajith.sirra@amd.com>

ajith-sirra-amd requested a review from a team July 8, 2026 11:36

ajith-sirra-amd requested a review from billishyahao as a code owner July 8, 2026 11:36

github-project-automation Bot added this to InferenceMAX Board Jul 8, 2026

ajith-sirra-amd requested review from 1am9trash, chunfangamd, seungrokj and yctseng0211 as code owners July 8, 2026 11:36

[AMD] MINIMAX-M3 FP4 vLLM Agentic Support. Added PR Link

d6be9de

Signed-off-by: ajith-sirra-amd <ajith.sirra@amd.com>

seungrokj changed the title ~~[AMD] MINIMAX-M3 FP4 vLLM Agentic Support~~ [AMD] MINIMAX-M3 FP4 MI355X agentX vLLM Jul 8, 2026

seungrokj added the AMD label Jul 8, 2026

[AMD] MINIMAX-M3 FP4 vLLM Agentic Support. Search Space Modification.

cdac231

Signed-off-by: ajith-sirra-amd <ajith.sirra@amd.com>

claude Bot reviewed Jul 8, 2026

View reviewed changes

Comment thread benchmarks/single_node/agentic/minimaxm3_fp4_mi355x.sh

Comment thread benchmarks/single_node/agentic/minimaxm3_fp4_mi355x.sh

Comment thread benchmarks/single_node/agentic/minimaxm3_fp4_mi355x.sh Outdated

seungrokj reviewed Jul 8, 2026

View reviewed changes

Comment thread configs/amd-master.yaml Outdated

ajith-sirra-amd added 5 commits July 8, 2026 17:18

[AMD] MINIMAX-M3 FP4 vLLM Agentic Support. Search Space Modification.

89daabe

Signed-off-by: ajith-sirra-amd <ajith.sirra@amd.com>

[AMD] MINIMAX-M3 FP4 vLLM Agentic Support. Search Space Modification.

c54a190

Signed-off-by: ajith-sirra-amd <ajith.sirra@amd.com>

[AMD] MINIMAX-M3 FP4 vLLM Agentic Support. Search Space Modification …

bf6d169

…& Fixing MODEL_PATH Arg in Serve Command. Signed-off-by: ajith-sirra-amd <ajith.sirra@amd.com>

[AMD] MINIMAX-M3 FP4 vLLM Agentic Support. Search Space Modification …

8bf107f

…& Fixing MODEL_PATH Arg in Serve Command. Signed-off-by: ajith-sirra-amd <ajith.sirra@amd.com>

[AMD] MINIMAX-M3 FP4 vLLM Agentic Support. Search Space Modification.

02cf9fd

Signed-off-by: ajith-sirra-amd <ajith.sirra@amd.com>

ajith-sirra-amd added the full-sweep-enabled label Jul 8, 2026

functionstackx reviewed Jul 8, 2026

View reviewed changes

Comment thread configs/amd-master.yaml Outdated

seungrokj and others added 2 commits July 9, 2026 10:26

Merge branch 'main' into amd/agentx-minimax-m3-vllm

ca7be40

seungrokj and others added 2 commits July 9, 2026 12:35

chore: revert max-num-seqs to CONC and drop hybrid KV cache flag

82466e2

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

[AMD] MINIMAX-M3 FP4 vLLM Agentic Support. Search Space Modification.

4bff323

Signed-off-by: ajith-sirra-amd <ajith.sirra@amd.com>

ajith-sirra-amd added 2 commits July 9, 2026 17:27

[AMD] MINIMAX-M3 FP4 vLLM Agentic Support. Modifying Search Space.

9f628cc

Signed-off-by: ajith-sirra-amd <ajith.sirra@amd.com>

Merge branch 'amd/agentx-minimax-m3-vllm' of https://github.com/SemiA…

7fad233

…nalysisAI/InferenceX into amd/agentx-minimax-m3-vllm

Merge branch 'main' into amd/agentx-minimax-m3-vllm

7e8ccea

cquil11 added the agentx AgentX benchmarks, recipes, and infrastructure label Jul 9, 2026 — with ChatGPT Codex Connector

Merge branch 'main' into amd/agentx-minimax-m3-vllm

fe31146

seungrokj mentioned this pull request Jul 10, 2026

[AMD][AgentX] MINIMAX-M3 FP4 MI355X agentX vLLM #2129

Draft

Merge branch 'main' into amd/agentx-minimax-m3-vllm

493a0cf

seungrokj self-requested a review July 10, 2026 12:00

seungrokj approved these changes Jul 10, 2026

View reviewed changes

cquil11 added 3 commits July 10, 2026 11:14

fix(agentic): preserve matrix CPU DRAM value

ccb54cd

fix(agentic): use explicit MI355X server port

cf4bba6

fix(agentic): remove recipe env fallbacks

060ec35

SemiAnalysisAI deleted a comment from seungrokj Jul 10, 2026

Merge main into amd/agentx-minimax-m3-vllm

f323281

SemiAnalysisAI deleted a comment from seungrokj Jul 10, 2026

cquil11 added full-sweep-enabled and removed full-sweep-enabled labels Jul 10, 2026

github-actions Bot and others added 2 commits July 10, 2026 17:35

fix(configs): restore indentation of dsv4 decode additional-settings

e8ff5ba

The extra leading space broke YAML parsing of amd-master.yaml. Co-authored-by: Cameron Quilici <60715037+cquil11@users.noreply.github.com> Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Merge branch 'main' into amd/agentx-minimax-m3-vllm

794d8be

functionstackx requested changes Jul 10, 2026

View reviewed changes

Uh oh!

Conversation

ajith-sirra-amd commented Jul 8, 2026

Summary

Uh oh!

github-actions Bot commented Jul 8, 2026

Uh oh!

github-actions Bot commented Jul 8, 2026

Uh oh!

github-actions Bot commented Jul 8, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

github-actions Bot commented Jul 8, 2026

Uh oh!

Uh oh!

github-actions Bot commented Jul 9, 2026

Uh oh!

github-actions Bot commented Jul 9, 2026

Uh oh!

github-actions Bot commented Jul 9, 2026

Uh oh!

github-actions Bot commented Jul 9, 2026

Uh oh!

github-actions Bot commented Jul 10, 2026

Uh oh!

github-actions Bot commented Jul 10, 2026

Uh oh!

seungrokj left a comment

Choose a reason for hiding this comment

Uh oh!

cquil11 commented Jul 10, 2026

Uh oh!

cquil11 commented Jul 10, 2026

Uh oh!

Klaud-Cold commented Jul 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Fixed YAML formatting

Uh oh!

github-actions Bot commented Jul 10, 2026

Uh oh!

functionstackx left a comment

Choose a reason for hiding this comment

Uh oh!

github-actions Bot commented Jul 10, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

Klaud-Cold commented Jul 10, 2026 •

edited

Loading