[AMD] Optimize kimik2.5-int4-mi300x-vllm, dsr1-fp8-mi300x-sglang

[lishuoshuo-amd]

Automated performance optimization from Hyperloom CI (2026-05-05).

kimik2.5-int4-mi300x-vllm

Metric	Value
Baseline (tok/s/GPU)	38.65
Optimized (tok/s/GPU)	51.86
Optimization Gain	+34.2%
InferenceX (tok/s/GPU)	39.25
vs InferenceX	+32.1%

Server flag changes:

• --max-num-seqs: 256 → 128

dsr1-fp8-mi300x-sglang

Metric	Value
Baseline (tok/s/GPU)	88.0
Optimized (tok/s/GPU)	91.7
Optimization Gain	+4.2%
InferenceX (tok/s/GPU)	149.31
vs InferenceX	+4.2%

Server flag changes:

• --mem-fraction-static: 0.8 → 0.85
• --num-continuous-decode-steps: 4 → 16

Related Issue

Automated by Hyperloom CI

Type of Change

• Configuration change

Checklist

• I have tested my changes locally
• I have updated documentation if necessary
• If I changed a container image or config, I have already updated perf-changelog.yaml

[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://cookbook.sglang.io/
• https://docs.vllm.ai/projects/recipes/en/latest/

If it is not, please create a PR first before we can merge your 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/sgl-cookbook

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

If additional help is needed, PR authors can reach out to core maintainers over Slack.