mudler / mudler/vllm.cpp

The record-anchor baseline was lowered to zero stale without the repair that earns it

Open
#2,934 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
C++
Stars
423
Forks
53
Avg merge
20h 26m
Merged PRs (30d)
310

Description

Row: GFX1100-TG200

ae5654785 in #2894 sets scripts/record-anchor-baseline.json to
{"total": 5, "buckets": {"stale": 0, "broken": 5}}, down from main's
{"total": 33, "buckets": {"stale": 28, "broken": 5}}.

Its commit body says the repair was done:

Record-anchor ratchet: 29 stale citations across 4 matrix files pointed to
symbols that moved during upstream growth. All 28 relocated symbols updated
to their new file:line

The commit touches three files, and none of them is a matrix file:

scripts/record-anchor-baseline.json |  4 ++--
src/vt/rocm/rocm_moe_router.hip     | 19 +++++++++++++++++--
src/vt/rocm/rocm_skinny_gemm.hip    |  3 ++-

So the budget was lowered and the 28 citations were left where they were.
python3 scripts/check-agent-record.py on the merged tree reports
RECORD ANCHOR REGRESSION in bucket 'stale': 30 > baseline 0, which is the
same class of failure the file's own comment names: "It is a rot budget, never
to be raised to make a failing check pass." Lowering it without the repair has
the same effect from the other side -- it reds the gate for every later change.

The same commit body's item 3 ("Remove the dead test and its CMake entry") is
also absent from the diff. The two fixes the diff DOES carry -- the MoeSiluMul
bf16 intermediate rounding and the wvSplitK m + y < M store bound -- are real
and are not in question here.

Separately, two anchors go stale because of this stack rather than because of
main: T21 (#2804) and T25 (#2807) add lines to
src/vllm/model_executor/models/qwen3_5_gguf_weights.cpp, moving
LoadGgufSharedEmbedAndHeadBf16 1034 -> 1051 and LoadQwen3_5MTPFromGguf
1447 -> 1526. Those are the record edits the stack owes.

Found while reviewing #2894. Fixed in the same flow, on that pull request's
branch: the two moved anchors are repaired in .agents/engine-matrix.md and the
baseline goes back to main's earned 28 / 5.

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start with scripts/record-anchor-baseline.json, .agents/engine-matrix.md, and python3 scripts/check-agent-record.py. Verify the stale anchors for GFX1100-TG200, including the Qwen3.5 GGUF line moves, and confirm the baseline reflects repaired citations rather than suppressing the failure; the issue also names a missing dead-test removal and CMake entry.

Written by the indexing model from the issue text.

Assessment

Tech stack
cpp, python
Domain
testing-qa, tooling
Issue type
Bug
Difficulty
3/5
Estimated time
1-2 days
Activity status
Stale
Clarity
Clearly specified
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.