[codex] Update MiniMax M3 B300 EAGLE3 FlashInfer image

[cquil11]

Summary

• update minimaxm3-fp8-b300-vllm-mtp to vllm/vllm-openai:minimax-m3-0618
• enable FlashInfer TRT-LLM attention with FP8 indexer KV and KV cache
• use TRITON_ATTN for the EAGLE3 drafter and the image's default FlashInfer TRT-LLM MoE backend

Validation

• benchmark script passes bash -n
• MiniMax M3 B300 vLLM sweep generation succeeds

---

Note

Medium Risk
Changes inference stack (attention, KV dtypes, MoE, speculative drafter) and mutates installed vLLM at runtime; scope is benchmark recipes only, not production serving paths.

Overview
Updates minimaxm3-fp8-b300-vllm-mtp to vllm/vllm-openai:minimax-m3-0618-x86_64-cu130 in nvidia-master.yaml and aligns the B300 EAGLE3 benchmark script with that image.

The serve command now turns on FlashInfer TRT-LLM attention (use_trtllm_attention), FP8 indexer KV and FP8 KV cache, and sets the EAGLE3 draft to TRITON_ATTN instead of FLASH_ATTN. For TP-only runs, --moe-backend marlin is removed so MoE uses the image default FlashInfer TRT-LLM path.

Before launch, the script applies a temporary in-container patch to sparse_attention_msa.py (.contiguous() on sliced prefill top-k) so MSA CSR build works with CUDA graphs when TP leaves multiple local KV heads.

perf-changelog.yaml records the image, attention/KV, MoE, and patch changes for this config key.

^{Reviewed by Cursor Bugbot for commit d17d410. Bugbot is set up for automated code reviews on this repo. Configure here.}

[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.

[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.

[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.

[github-actions]

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=27792967799
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=27792967799

[github-actions]

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=27796130302
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=27796130302

[github-actions]

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=27805367644
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=27805367644

[Oseltamivir]

/reuse-sweep-run

[Oseltamivir]

lgtm