Fix nvidia-master.yaml metadata to align with recipe files

[nlevin-ui]

Audit found 34 metadata mismatches across 412 entries in nvidia-master.yaml where declared TP/EP/DP-attn values didn't match the actual recipe files.

Fixes applied:

• B200 FP8 TRT 8K/1K: 15 prefill_ep (8->1), 1 prefill_dp_attn (true->false)
• H200 FP8 TRT: 17 prefill_dp_attn mismatches
  • 1K/1K: false->true (9 entries)
  • 8K/1K: true->false (8 entries)
• GB200 FP4 TRT 8K/1K: 2 decode_dp_attn (true->false)

All changes are metadata-only in nvidia-master.yaml to reflect actual recipe values. No recipe files were modified.

[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

[claude]

Claude finished @nlevin-ui's task in 43s —— View job

---

PR Review: Fix nvidia-master.yaml metadata alignment

• Gather context and review diff
• Validate changes against PR description
• Check perf-changelog.yaml requirement
• Post review

---

Review Summary: The metadata corrections look correct and consistent with the PR description. I verified all 35 changes:

Category	Change	Count
B200 FP8 TRT 8K/1K	prefill_ep: 8 → 1	15
B200 FP8 TRT 8K/1K (STP)	prefill_dp_attn: true → false	1
H200 FP8 TRT 1K/1K	dp-attn: false → true	9
H200 FP8 TRT 8K/1K	dp-attn: true → false	8
GB200 FP4 TRT 8K/1K	decode_dp_attn: true → false	2
Total		35

All changes are metadata-only corrections aligning declared values with actual recipe files. No issues with model prefixes or image accessibility.

However, there is one blocking issue:

🔴 BLOCKING: Master config file .github/configs/nvidia-master.yaml was modified but perf-changelog.yaml was not updated. When changing .github/configs/nvidia-master.yaml, you must add a corresponding entry to perf-changelog.yaml documenting the changes and including the PR link.

[functionstackx]

thanks for the fix

[claude]

LGTM — straightforward metadata-only corrections aligning declared ep/dp-attn values with actual recipe files.

Extended reasoning...

Overview

This PR modifies a single file (.github/configs/nvidia-master.yaml) to fix 34 metadata mismatches where declared ep and dp-attn values did not match the actual recipe YAML files they reference. The changes span B200 FP8 TRT (15 ep corrections 8→1, 1 dp-attn correction), H200 FP8 TRT (9 dp-attn false→true for 1K/1K, 8 dp-attn true→false for 8K/1K), and GB200 FP4 TRT (2 decode dp-attn true→false).

Security risks

None. This is a config/metadata file with no executable code, authentication, or permission changes.

Level of scrutiny

Low scrutiny is appropriate. The changes are mechanical one-field corrections in a declarative YAML config. Each change is a single value flip (ep: 8 → ep: 1, or dp-attn: true ↔ dp-attn: false) and the PR description clearly documents the audit methodology and categorizes every correction.

Other factors

The recipe files referenced in the comments are hosted in an external repository, so I could not locally verify the claimed mismatches. However, the changes are self-consistent, well-documented in the PR description, and the pattern of corrections (e.g., all B200 8K/1K prefill entries consistently changing ep from 8 to 1) is coherent. No outstanding reviewer comments exist, and no prior reviews from me. The only timeline entry is an automated bot reminder.