[codex] Update MiniMax M3 B300 FlashInfer image

[cquil11]

Summary

• update minimaxm3-fp8-b300-vllm to vllm/vllm-openai:minimax-m3-0618-x86_64-cu130
• enable FlashInfer TRT-LLM attention with FP8 indexer KV and KV cache
• use the image's default FlashInfer TRT-LLM MoE backend instead of explicitly forcing Marlin
• apply a guarded runtime patch for the image's MiniMax M3 MSA top-k contiguity bug

Validation

• benchmark script passes bash -n
• MiniMax M3 B300 vLLM sweep generation succeeds
• matrix logic suite passes (156 tests)
• perf changelog validation passes

---

Note

Medium Risk
Changes how a large-model benchmark server is launched (attention/MoE/KV dtypes) and mutates installed vLLM source at runtime until the image ships a fix; benchmark correctness and sweep stability are the main concern, not security.

Overview
Moves minimaxm3-fp8-b300-vllm to vllm/vllm-openai:minimax-m3-0618-x86_64-cu130 in NVIDIA master config and documents the same image name in comments.

The B300 fixed-seq-len benchmark script now starts vLLM with FlashInfer TRT-LLM attention, FP8 indexer KV, and FP8 KV cache, and stops forcing --moe-backend marlin for TP-only runs so MoE follows the new image default. Before serve, it applies an idempotent in-container patch to sparse_attention_msa.py (.contiguous() on the MSA prefill top-k slice) to avoid warmup failures when data-parallel attention leaves multiple local KV heads.

KLAUD_DEBUG.md adds a troubleshooting section for that MSA contiguity error; perf-changelog.yaml records the config and benchmark behavior change (PR #1834).

^{Reviewed by Cursor Bugbot for commit 530f2de. 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=27792967867
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=27792967867

[github-actions]

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=27796124175
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=27796124175

[github-actions]

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=27797429264
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=27797429264

[Oseltamivir]

/reuse-sweep-run

[Oseltamivir]

lgtm