Repository navigation
Conversation
Publish frozen contracts, source-bound results and rejected candidates with four-track status, reproduction limits and inventory hashes. Keep numerical, model-quality and complete-HTTP conclusions separate. Agent-assisted-by: OpenAI Codex (author) Signed-off-by: GUOGUO <xwdong1998@163.com>
Dong1017
added
the
area:workbench
ChatGPT Codex Connector
Add the main-based native qualification and distinguish it from historical full-model/HTTP campaign evidence. Agent-assisted-by: OpenAI Codex (author) Signed-off-by: GUOGUO <xwdong1998@163.com>
Dong1017
left a comment
Dong1017
left a comment
There was a problem hiding this comment.
Static evidence review at a0677a03: one historical-validator limitation should be documented before reuse. The reported single-pair 1.57% result, rejected candidates and incomplete qualification remain appropriately bounded; this is not campaign acceptance or an unconditional LGTM.
- Please preserve frozen validator bytes/receipts and document the mixed FAIL/missing-evidence precedence limitation below. A future qualified version should preserve observed FAIL while also reporting missing evidence.
- Reuse the existing cross-PR inventory finding when integrating both bundles; it remains unresolved.
- Software CI passed, but no device/model execution or independent raw-artifact replay was performed in this review. New-base full-model and HTTP qualification remain NOT_RUN.
Inspected evidence
Head: a0677a0; base: 2baebf6. Static inspection covered archived validators/scorers, frozen contracts, principal reports/projections, inventory and upstream receipt. The 28 quality projections retain 24 FAIL / 4 INCONCLUSIVE; 40 run projections agree with the report. Current-head software run: https://github.com/InferMatrixLab/InferRSI/actions/runs/41418369337 . No hardware efficacy inferred.
| same({k: reference['media'][k] for k in MEDIA_KEYS}, | ||
| {k: value['media'][k] for k in MEDIA_KEYS}, | ||
| f'{label}:sample-{index}:decoded_pixels_and_pcm') | ||
| verdict = 'INCONCLUSIVE' if missing else 'FAIL' if failures else 'PASS' |
There was a problem hiding this comment.
P2 — historical verdict precedence differs from the documented contract. An observed cleanup failure (or tensor mismatch) plus a missing rank configuration yields INCONCLUSIVE here, even though failures contains the violation; validator/README.md says observed gate violations remain FAIL. Please annotate this limitation while retaining frozen bytes and receipts, and use a separately versioned FAIL-first validator/wrapper with a mixed failure+missing regression before future qualification. This does not invalidate the reported C1/C13 PASS receipts, whose failure and missing lists are empty; no affected mixed-outcome receipt is established by this static review.
|
Review at head
|
Governing Issue and delivery slice
Refs #297
Downstream engine contribution
vLLM-Omni Draft PR #9745
publishes only the C1 BF16 prefix-attention implementation, tests and recipe
against main: base
b84bade9b9be8fdad0a1a9b3e8dd3ab4273dc4d9, head7a20cf9b987b22fe3ce00221cb30cc3c6fb90e63. Its new-base focused native suitepassed 12 tests with actual Ascend fused attention, source binding and cleanup.
Portable native record. New-base full-model 5 s/15 s quality and
clean HTTP A/B are not run. Historical C1 strict/latency results below retain
their original revision; the 50% campaign target remains unmet.
Publish the requested Draft tracking snapshot for MiniMax-H3 Ref2VA on
four logical Ascend 910C devices, covering 5 s/15 s × lossless/lossy.
Campaign verdict: INCONCLUSIVE. No accepted 50% complete-HTTP reduction.
This is the issue owner's evidence delivery slice; final goal acceptance remains
with the requester.
The structure follows PR #294:
frozen plans, source attribution, results, rejected attempts, validators,
integrity hashes and replay limits. Its eight-device 910B1 campaign and accepted
result remain separate from this four-logical-device 910C campaign.
campaign and this tracking PR. The target remains at least 50% matched-B0
complete-HTTP reduction for each track, strict bit-exact lossless checks and
frozen automatic lossy gates without human media inspection.
C0–C23 direction ledger, 28 automatic quality comparisons, 40 C12–C23 run
projections, experimental engine diffs and source identities, eight unchanged
frozen snapshots, plus inventory registration and count assertions.
final four-track acceptance, a production engine contribution or a
self-contained device replay environment.
Current results
Evidence cutoff: 2026-11-02 15:52 +08.
The clean 5 s pair used two warmups and three measured complete MP4 responses
per fresh server, with audit/profiling disabled. Source integrity, cleanup and
strict paired checks passed. It is one screening pair, not the final three-pair
AB/BA/AB protocol. The delta includes migration+C1; no isolated migration-only
clean timing arm exists. Statistical significance is not claimed.
Recent completed diagnostic 5s-00 requests:
Each passed 134 input checks and cleanup; none proceeds to 15 s or clean timing.
C23 V2 also passed 44 native tests, including three executed four-rank suites,
and reduced its static 50-call attention cycle by 66.46%. That primitive
measurement excludes FFN, changing-input quality, complete-model timing and HTTP;
the complete model's quality failure prevents promotion.
Changed artifacts and evidence
Baseline engine:
df1c1b05936381dbaa3891c1ae4c4b21ab375a5d.Migration source:
922f1e9f1cb85b9f2b66d86e1eb500e6872b2dcb.Executed InferRSI:
50c19da43e0b358e6f990c00719891ad308a8ac1.Latest rejected engine candidate C23:
6eb7bb094effb40853431208d1631effaf452a3f.The publication commit is not a new experiment runtime revision.
Checks and dependencies
python tools/validate_evidence.py: PASS, 21 bundles / 528 hash checks,zero errors. Its 16 external/unavailable records across the inventory remain
explicit; successful byte checks do not replay devices or change verdicts.
ruff check src tests tools: PASS.pytest tests/test_evidence_inventory.py tests/test_smoke.py tests/test_claim.py:24 passed, 1 skipped. The symlink-escape fixture skipped because symlink
creation is unavailable in this Windows environment.
snapshots remained exact. Relative Markdown links and the pending artifact
scan passed. Private host identifiers, paths, credentials, raw media and weights
are excluded. These checks do not substitute for independent review.
was started for this publication. Original MindIE-SD source/binaries, shared
checkouts and global SDK/driver environments were not changed.
and independent review are not claimed by the local focused tests.
Knowledge authorship declaration
Not applicable under the current protected-path policy: this changes campaign
evidence, its inventory entry and inventory count assertions only.
Agent-assisted-by: OpenAI Codex (author)
Remaining work and Issue closure
All 24 direction families are charged. The original active deadline is
2026-11-03 01:19 +08, with no extra final-confirmation budget. Further new
mechanism search requires a requester extension of the direction ceiling;
this tracking PR grants no execution budget or new device run.
Four-track final confirmation, full lossy quality coverage and independent review
remain incomplete. The observed failed configurations stay FAIL; missing samples
are not PASS. Goal acceptance belongs to the requester and is not claimed.
Keep this PR Draft for tracking and keep the acceptance Issue open.