Skip to content
Closed
Show file tree
Hide file tree
Changes from 2 commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion .github/configs/amd-master.yaml
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
dsr1-fp4-mi355x-sglang:
image: lmsysorg/sglang:v0.5.9-rocm700-mi35x
image: lmsysorg/sglang:v0.5.10.post1-rocm720-mi35x
model: amd/DeepSeek-R1-0528-MXFP4-Preview
model-prefix: dsr1
runner: mi355x
Expand Down
6 changes: 6 additions & 0 deletions perf-changelog.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -2031,3 +2031,9 @@
- "Topologies mirror the dsv4-fp4-gb300-dynamo-vllm sibling: low-conc 1p1d-dep8-tep8 (4 nodes), mid 1p1d-dep8-dep16 (6 nodes), high 3p1d-dep8-dep16 (10 nodes). 4096 overlap between mid and high gives a topology-crossover A/B"
- "No upstream GB300 DSV4 sglang disagg recipe exists. Per-worker sglang_config (env vars + flashinfer_mxfp4 + chunked-prefill-size 4096 + disable-flashinfer-autotune + mem-fraction-static 0.82) is mirrored from NVIDIA/srt-slurm PR #69 (recipes/gb300-fp4/1k1k-dsv4/agg-2n-low-latency.yaml — GB300 DSV4 SGLang aggregated). Disagg flag set (nixl transfer backend, enable-dp-attention + moe-a2a-backend deepep) cross-checked against PR #75 (recipes/gb300-fp4/1k1k-dsv4/disagg-1p1d-tp4-mxfp4.yaml — GB300 DSV4 SGLang disagg) and the SGLang DeepSeek-V4 cookbook. Stored under benchmarks/multi_node/srt-slurm-recipes/sglang/deepseek-v4/ and overlaid onto the upstream srt-slurm checkout at runtime"
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1157

- config-keys:
- dsr1-fp4-mi355x-sglang
description:
- "Updating MI355x Deepseek-R1 FP4 SGLang Image to upstream v0.5.10.post1"
pr-link: https://github.com/InferenceMAX/InferenceMAX/pull/1237

Check warning on line 2039 in perf-changelog.yaml

View check run for this annotation

Claude / Claude Code Review

Wrong pr-link URL: InferenceMAX/InferenceMAX vs SemiAnalysisAI/InferenceX

The new `pr-link` at perf-changelog.yaml:2039 points to `https://github.com/InferenceMAX/InferenceMAX/pull/1237`, but this PR is in `SemiAnalysisAI/InferenceX`. Every other entry in the file (231 of them, including the immediately preceding entry for PR #1157 on line 2033) uses `https://github.com/SemiAnalysisAI/InferenceX/pull/<num>`. Should be `https://github.com/SemiAnalysisAI/InferenceX/pull/1237` — looks like a copy-paste typo.
Comment thread
ppalanga marked this conversation as resolved.
Outdated
Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

🟡 The new pr-link at perf-changelog.yaml:2039 points to https://github.com/InferenceMAX/InferenceMAX/pull/1237, but this PR is in SemiAnalysisAI/InferenceX. Every other entry in the file (231 of them, including the immediately preceding entry for PR #1157 on line 2033) uses https://github.com/SemiAnalysisAI/InferenceX/pull/<num>. Should be https://github.com/SemiAnalysisAI/InferenceX/pull/1237 — looks like a copy-paste typo.

Extended reasoning...

What the bug is

The new perf-changelog.yaml entry added by this PR (lines 2034-2039) sets:

  pr-link: https://github.com/InferenceMAX/InferenceMAX/pull/1237

but this PR (#1237) lives in SemiAnalysisAI/InferenceX, not InferenceMAX/InferenceMAX. The repo URL is wrong.

Why this is clearly a typo

  • Every single one of the other ~231 pr-link entries in perf-changelog.yaml uses https://github.com/SemiAnalysisAI/InferenceX/pull/<num>. This new entry is the only outlier with a different host repo.
  • The immediately preceding entry on line 2033 (for PR Day 0 DeepSeek V4 Pro FP4 GB300 disaggregated SGLang benchmarks #1157) uses the correct https://github.com/SemiAnalysisAI/InferenceX/pull/1157.
  • Per the PR metadata, this PR is [AMD] Update MI355x Deepseek-R1 FP4 SGLang Image to v0.5.10 #1237 in the SemiAnalysisAI/InferenceX repo (the same repo where perf-changelog.yaml lives). The README does mention the project was 'formerly InferenceMAX', which likely explains how this slipped in — the maintainer probably had the old name in mental cache when copy-pasting.

Proof / step-by-step

  1. Open perf-changelog.yaml at line 2039: pr-link: https://github.com/InferenceMAX/InferenceMAX/pull/1237.
  2. Grep all pr-link: lines in the file → every other one uses the SemiAnalysisAI/InferenceX host.
  3. Look at the prior entry (line 2033, PR Day 0 DeepSeek V4 Pro FP4 GB300 disaggregated SGLang benchmarks #1157): pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1157 — the convention.
  4. Check the PR metadata for this PR: it is PR [AMD] Update MI355x Deepseek-R1 FP4 SGLang Image to v0.5.10 #1237 of SemiAnalysisAI/InferenceX. The link as written either 404s or, if that other repo exists, points at an unrelated PR.

Impact

Low — pr-link is a documentation/audit field; it doesn't affect benchmark execution. But the link is broken or misleading, so the changelog cross-reference is unusable for anyone clicking through to find the originating PR.

Fix

Change line 2039 to:

  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1237

Re bug_003 dedup objection

bug_001 and bug_003 are the same finding (same file, same line, same fix); the synthesis agent has correctly merged them into a single report here.

Loading