Fix B200 Dynamo vLLM recipe concurrencies

[Oseltamivir]

Summary

• Sync the PR1303 B200 Dynamo vLLM recipe benchmark concurrencies with .github/configs/nvidia-master.yaml.
• Update the low-latency, high-throughput, and max-throughput recipe files so srt-slurm actually runs every configured concurrency point.
• Append a perf-changelog.yaml entry for dsv4-fp4-b200-dynamo-vllm so the PR sweep targets this config.

Root Cause

• nvidia-master.yaml exported CONC_LIST, but the B200 Slurm launcher applies the recipe file directly. The recipe-local benchmark.concurrencies values therefore controlled the actual benchmark run and only covered a subset of the configured concurrencies.

Validation

• python utils/matrix_logic/generate_sweep_configs.py full-sweep --config-files .github/configs/nvidia-master.yaml --framework dynamo-vllm --model-prefix dsv4 --runner-type b200-multinode --multi-node
• python utils/matrix_logic/generate_sweep_configs.py test-config --config-files .github/configs/nvidia-master.yaml --config-keys dsv4-fp4-b200-dynamo-vllm
• python utils/process_changelog.py --base-ref origin/main --head-ref HEAD --changelog-file perf-changelog.yaml
• git diff --check origin/main..HEAD

[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

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.

[claude]

Claude finished @Oseltamivir's task in 47s —— View job

---

Reviewing PR #1305

• Gather context and read changed files
• Validate concurrencies against nvidia-master.yaml
• Validate perf-changelog.yaml entry position
• Post review

---

LGTM — no blocking issues found.

All three recipe concurrency values now match the nvidia-master.yaml search space exactly:

Recipe	Master conc-list	Recipe concurrencies
disagg-b200-low-latency.yaml	[1, 64, 128]	"1x64x128"
disagg-b200-high-tpt-megamoe.yaml	[1024, 2048, 4096, 8192]	"1024x2048x4096x8192"
disagg-b200-max-tpt-megamoe.yaml	[8192]	"8192"

The perf-changelog.yaml entry is correctly appended at the end of the file.

[claude]

LGTM — straightforward concurrency sync between recipe YAMLs and the master config.

Extended reasoning...

Overview

This PR updates the benchmark.concurrencies field in three B200 Dynamo vLLM recipe YAMLs (low-latency, high-tpt-megamoe, max-tpt-megamoe) under benchmarks/multi_node/srt-slurm-recipes/vllm/deepseek-v4/8k1k/ so they match the search space declared in .github/configs/nvidia-master.yaml, and appends a corresponding entry to perf-changelog.yaml.

Security risks

None. These are benchmark configuration values (numeric concurrency points) and a changelog text entry — no executable code, no auth/crypto/permissions surface.

Level of scrutiny

Low. The change is purely a values update to YAML files used by the benchmarking sweep system, with a clearly stated root cause (recipe-local concurrencies override the master config because the Slurm launcher applies the recipe directly). The diff is mechanical and small (3 single-line value changes plus a 7-line changelog append).

Other factors

The author validated via the documented sweep config generators and changelog processor. No prior reviewer feedback is outstanding, and no bugs were flagged by the bug hunting system.

[github-actions]

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=25608571677
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=25608571677

[functionstackx]

@claude rebase and resolve conflicts

[Klaud-Cold]

Claude finished @functionstackx's task in 1m 49s —— View job

---

Rebase and resolve conflicts

• Fetch latest origin/main
• Check for conflicts between this branch and main
• Rebase onto main and resolve conflicts
• Push updated branch

---

Rebased onto origin/main (which had 1 new commit: 2fb1c3b5 — PR #1304). Resolved a conflict in perf-changelog.yaml by keeping both the #1304 and #1305 entries in chronological order. Force-pushed the rebased branch.
|

[github-actions]

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=25616590246
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=25616590246

[github-actions]

see unofficial run visualizer at https://inferencex.semianalysis.com/inference?unofficialRun=25617236455
see unofficial run visualizer at https://inferencex.semianalysis.com/evaluation?unofficialRun=25617236455