[DNM][AMD] agentx-v0.4

[seungrokj]

Summary

• Add minimaxm2.5-fp4-mi355x-vllm-agentic-lmcache and kimik2.5-fp4-mi355x-vllm-agentic-lmcache entries in amd-master.yaml
• Add minimaxm2.5_fp4_mi355x.sh agentic benchmark script with LMCache support
• Refactor kimik2.5_fp4_mi355x.sh: simplify env vars, build LMCache from source (ROCm HIP), tune LMCACHE_L1_SIZE_GB/TTL/chunk size
• Refactor qwen3.5_fp8_mi355x.sh: add HiCache offloading support, add 256k trace corpus cap via WEKA_LOADER_OVERRIDE
• Set LMCACHE_CHUNK_SIZE default to 32 for MiniMax agentic script

🤖 Generated with Claude Code

---

Note

Medium Risk
Large, experimental benchmark-only changes (runtime LMCache git builds, removed ROCm patches, hardcoded DRAM/TODOs) can make sweeps flaky or non-reproducible until tuned, but no production serving paths are touched.

Overview
Expands MI355X agentic-coding coverage in amd-master.yaml with new matrix targets that compare GPU-only vs HiCache (SGLang) or LMCache (vLLM) KV offload for Qwen3.5 FP4, GLM5.1 FP4, Kimi K2.5 FP4, and MiniMax M2.5 FP4/FP8—including a Kimi LMCache variant pinned to a 1500 GB DRAM experiment. The existing Qwen3.5 FP8 HiCache target is retuned (new SGLang image, TP=4, wider concurrency grid).

Agentic launch scripts are reworked to match: SGLang paths gain none/hicache offload wiring, larger host DRAM budgets (~3 TB), 060226 (256k-capped) trace overrides, and server-flag tweaks; vLLM paths drop the prior ROCm LMCache monkey-patches in favor of cloning and HIP-building LMCache at job time, with retuned L1 pool sizes, 7200s read TTL, and chunk size 32 on MiniMax. Kimi’s script is slimmed to that model; MiniMax FP4 gets a dedicated script; kimik2.5_fp4_mi355x_org.sh preserves an alternate Kimi LMCache sizing/trace setup. GLM5.1’s script is extended for HiCache while keeping GLM-specific SGLang flags.

Slurm launcher excludes bad node mia1-p01-g37 alongside existing MI355X blocklist entries.

^{Reviewed by Cursor Bugbot for commit 244de16. 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.

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 0607cc7. Configure here.}