Repository navigation
docs: route vLLM engine, admission and KV knowledge - #138
Conversation
Signed-off-by: GUOGUO <xwdong1998@163.com> Agent-assisted-by: Codex / gpt-6.1-sol [effort=xhigh] (author)
Signed-off-by: GUOGUO <xwdong1998@163.com> Agent-assisted-by: Codex / gpt-6.1-sol [effort=xhigh] (author)
Signed-off-by: GUOGUO <xwdong1998@163.com> Agent-assisted-by: Codex / gpt-6.1-sol [effort=xhigh] (author)
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
Dong1017
left a comment
Dong1017
left a comment
There was a problem hiding this comment.
Scoped LGTM at 4f3b79694e69e8e7896c4cd285327fc3a9a54d1b, base 65147a03e465dc838770474736a620e04d72337e.
Reviewed six navigation documents and the generated projection against #91. Engine and focused scheduler/KV pins retain distinct inspection dates, evidence ownership and candidate/not-run boundaries. All 13 added relative link occurrences and anchors resolve. Of 59 projection records, 57 change only source revision/URL; the two focused records also refresh only root-document hashes, independently checked against the combined source documents. No runtime, registry, dependency or evidence promotion change found.
All four exact-head Ready jobs pass on Python 3.10/3.11; per-head declaration status is success. The separate repository-wide sweep fails on other PRs, but its logs report this PR compliant on all three attempts. This is not an all-workflows-green claim.
Read-only source/projection/link and CI inspection; no repository code, generator, package APIs, native backend or model execution rerun, and no full source-hash recomputation. No actionable findings within navigation scope. CI does not establish cross-pin compatibility, quality, performance or agent efficacy. This review is not #91 acceptance or merge authorization.
Dong1017
added
the
area:knowledge
ChatGPT Codex Connector
Preserve every current-main canonical source and the original 70 navigation lines. Retain original source pins, inspection dates, licenses and candidate limits. Agent-assisted-by: Codex / gpt-6.1-sol [effort=xhigh] (author) Agent-assisted-by: Codex / gpt-6.1-sol [effort=max] (author) Signed-off-by: GUOGUO <xwdong1998@163.com>
Build shared records from the real combined source and verify exact resource and context reads. Preserve all retained source metadata and candidate validation states. Agent-assisted-by: Codex / gpt-6.1-sol [effort=xhigh] (author) Agent-assisted-by: Codex / gpt-6.1-sol [effort=max] (author) Signed-off-by: GUOGUO <xwdong1998@163.com>
Dong1017
left a comment
Dong1017
left a comment
There was a problem hiding this comment.
Scoped LGTM for the integration refresh at 943d9ad489d81e2a1a2f2d7a52be319307952293, base 5b939a95d7478b1d8bff013c477972903b7de9ba, compared with previously inspected 4f3b79694e69e8e7896c4cd285327fc3a9a54d1b. No new actionable integration finding.
All 70 original added navigation lines across six Markdown files remain identical. The old-to-new increment changes no vLLM owner files. Against current main the contribution remains those navigation changes plus generated projection; 58 of 60 records change source URL/commit only, and the two focused root bindings retain the previously reviewed hashes. Combined source 487593c merges old head with current main; source-to-head is generated-only. All 13 added relative links/anchors resolve. Existing source pins, scope and candidate/not_run boundaries remain intact.
Exact-head Ready https://github.com/InferMatrixLab/InferRSI/actions/runs/40272691341 has four successful full-check/installed-package jobs; lint is green. Per-head policy reports success via 40274952625. Aggregate declarations failure/cancellation is separate, not claimed green. Existing comments/reviews were read; head/base and mergeable=true were rechecked before posting. Recheck live gates after any preceding merge changes main.
Read-only incremental source/blob/projection/link and remote-CI inspection; no local generator/tests, runtime or model/hardware execution. This does not qualify a combined runtime stack or grant #91 goal acceptance. No merge or issue closure.
Reviewer: dot / OpenAI assistant; exact runtime model identifier is not exposed.
Dong1017
left a comment
Dong1017
left a comment
There was a problem hiding this comment.
LGTM within this PR's accepted source/software scope. Meaningful CI-blocker clearance on unchanged head 943d9ad489d81e2a1a2f2d7a52be319307952293: Ready CI and declarations rerun are successful; exact-head knowledge-model-policy is success. This updates the prior scoped review rather than reopening resolved findings.
Integration inspected against actual main db82ec8346364667b90f4fa97fdb190150141cf0, separately from the PR API's base snapshot 5b939a95d7478b1d8bff013c477972903b7de9ba. The intervening #154/#118 changes (18 paths) do not overlap this contribution. New evaluation-planner mapped canonical directories are not changed by this PR; its source delivery is not thereby covered by a real efficacy evaluation. No new actionable integration finding or unresolved review blocker was identified.
Verification limits: incremental source/integration/prior-thread and CI inspection, no new local tests or model/hardware execution. Existing source-fidelity/packaging conclusions retain their original limits; this is not goal/campaign acceptance or issue closure. Recheck exact head, applicable checks and integration immediately before any merge, especially after another knowledge PR changes shared matrices/projections. Historical red runs remain historical, not rewritten evidence.
Preserve both canonical knowledge histories before rebuilding generated records. Agent-assisted-by: Codex / gpt-6.1-sol [effort=max] (author) Signed-off-by: GUOGUO <xwdong1998@163.com>
Build the native projection from canonical source 905ecd9. Agent-assisted-by: Codex / gpt-6.1-sol [effort=max] (author) Signed-off-by: GUOGUO <xwdong1998@163.com>
Dong1017
left a comment
Dong1017
left a comment
There was a problem hiding this comment.
LGTM for repaired source/package integration at 6b0c46cce8beee09085bfd6e90a7bbc92001212c, compared only with previously inspected 943d9ad489d81e2a1a2f2d7a52be319307952293 plus affected projections/links.
All 70 navigation insertions across six Markdown files match the prior review byte-for-byte; no current-main canonical deletions, no changes to the vLLM owner resources. All 13 relative link occurrences/anchors resolve. Source 905ecd9a… → head is generated-only; 58 of 60 records change provenance only, and both changed root hashes were independently recomputed against their projections. Main #134/#142 additions are preserved, clearing the prior conflict blocker (mergeable=true).
All four Ready jobs and current per-head policy succeed. Separate Skill Eval publication fails HTTP 403; execution skipped, and the maintained evaluation map does not cover this contribution. This remains an operational limitation, not an all-checks-green or efficacy claim. No source correctness blocker found.
Inspected actual main 0c78a7e257dfaa3573b77581a39539e8a3a5da4c; API base a4a0fbd62defe2bc53ab090e1fa50bd7139663a6 differs only in four disjoint Flare paths. Static/delta/link/hash and remote-CI verification only; no generator/test rerun, model/device experiment, merge or issue acceptance.
Dong1017
left a comment
Dong1017
left a comment
There was a problem hiding this comment.
Integration-status update only at unchanged inspected head 6b0c46cce8beee09085bfd6e90a7bbc92001212c (PR API base snapshot a4a0fbd62defe2bc53ab090e1fa50bd7139663a6). Actual main is 38d5883b8e72e352865e214194cb22dc99253dbd; GitHub now reports mergeable=false / dirty. The prior source-scope review remains valid historical evidence, but is not current integration clearance.
Before merge, reconcile this branch with current main, preserve both accepted contributions, regenerate affected projections only where source-binding rules require it, and verify packaging plus applicable checks on the resulting head. No conflict-file attribution is made without a merge reproduction; no new content/feature requirement is introduced. This is not a new HTTP403 finding: shared publisher recovery was verified separately. No code execution, branch change, workflow dispatch or merge was performed by this review.
Preserve all original navigation additions and accepted main source content, including distinct source pins, inspection dates and scope boundaries. Regenerate discovery outputs separately from the committed combined source. Agent-assisted-by: Codex / gpt-6.1-sol [effort=max] (author) Signed-off-by: GUOGUO <xwdong1998@163.com>
Build the unchanged native projection from committed source 1e8982e. Agent-assisted-by: Codex / gpt-6.1-sol [effort=max] (author) Signed-off-by: GUOGUO <xwdong1998@163.com>
Skill evaluationAutomatic execution / cumulative PR budget: Advisory: obtain missing evidenceRetained evidence availability: planned / unexecuted. No approval or merge authorization is issued.
PLANNED: execution is not authorized by knowing this hash. Selected comparisons: 0 · Estimated task calls: 0. Applicable only while these revisions and inputs match the current PR. Plan bindings, budget, frozen exam and advisory detailsUnchanged-correct is valid neutral evidence, not incremental benefit. Per-case outcomes (discovery excluded; regressions and gaps first): []Showing at most 20 of 0 pairs; the evidence link contains all outcomes, including omitted regressions/gaps. Plan SHA-256: Model/settings:
Provider/settings: Uncovered / unsupported paths (81)
|
Dong1017
left a comment
Dong1017
left a comment
There was a problem hiding this comment.
Scoped LGTM for integration head dbea6bd2410bb6ea208820a8136e5e998d5a3b32, compared with reviewed 6b0c46cce8beee09085bfd6e90a7bbc92001212c; current base 51fa1d31cc4a805490ddfe8055a36c81d70c4f08.
All 70 navigation insertions across six documents are retained byte-for-byte; the three vLLM procedure bodies are unchanged from the prior head. The 13 added relative links/anchors resolve. Main-relative changes remain those navigation additions plus regenerated projections; runtime, generator, tests and registry are unchanged. Prior integration-conflict concern is cleared.
flowchart LR
A[Preserved source navigation] --> B[Current-main reconciliation]
B --> C[Refreshed catalog projection]
C --> D[Software checks pass]
Ready 40603290430: all four Python 3.10/3.11 full-check/installed-package jobs pass. Exact-head canonical policy 40606014358 succeeds. Skill Eval plan selects zero comparisons, so provides no efficacy result.
Static delta, link, projection/test-source and remote CI inspection only; no local generator/test execution, full independent hash recomputation or model/device run. Original source pins and candidate/not-run limits remain. This conclusion does not grant #91 acceptance or perform a merge.
|
Merge-wave integration prerequisite at inspected head #153 changed the same generated catalog provenance records. Integrate current main while preserving both contributions, then regenerate via the existing generator rather than choosing ours/theirs wholesale. Re-run applicable Ready/package checks on the resulting head. This is integration repair, not a new content or efficacy requirement. Existing green checks predate this combined state. flowchart LR
A[Reviewed contribution] --> B[Integrate current main]
B --> C[Regenerate and verify]
C --> D[Merge]
|
Preserve all original navigation additions and accepted main source content, including distinct source pins, inspection dates and scope boundaries. Regenerate discovery outputs separately from the committed combined source. Agent-assisted-by: Codex / gpt-6.1-sol [effort=max] (author) Signed-off-by: GUOGUO <xwdong1998@163.com>
Generate the unchanged native projection from committed source 92f03ea. Preserve current main and original candidate semantics with actual source hashes. Agent-assisted-by: Codex / gpt-6.1-sol [effort=max] (author) Signed-off-by: GUOGUO <xwdong1998@163.com>
Dong1017
left a comment
Dong1017
left a comment
There was a problem hiding this comment.
LGTM 8028f81
Independent review against frozen integration base 25d6d5a603824e3893415841e1709521cd0eeb04. The original 70 navigation insertions across six documents and all 13 links/anchors are preserved; removing those blocks recovers the exact main bytes. Both matrix row sequences, source pins, original inspection dates and candidate boundaries remain. Metadata, registry, runtime, generators and tests are unchanged. No actionable findings remain.
Verified all 352 indexed source bindings to 92f03ea79dbd52e38abb2216f7c062ae823bbabe: 350 previously verified identical Git blobs plus two fresh Git-byte hashes. The generated-only child contains 67 index files and the coverage refresh; all 66 coverage rows are unchanged and only its 132-input digest changes. Inspected the actual single-render/readback, complete 423-resource package, three changed SKILL resource reads, three retained sub-skill catalog pointers and README engine-root route.
All four current-head Ready full-suite/clean-wheel jobs pass on Python 3.10/3.11; the inspected 3.11 log merges this head into the stated base and reports 1413 passed, 9 skipped and 150 subtests passed. Lint, fresh declarations and automatic plan/publication pass. Execute is correctly skipped; zero comparisons provide no efficacy evidence.
Subsequent main aaf748dbc7c9c62ebca338d23d70ae3c78253871 changes admission-policy code/tests/docs only, with no indexed or coverage input changes; it is not claimed as the Ready-tested base. The latest stable policy context is pending normal re-evaluation at this snapshot. The integrator must check current head/base and latest applicable gates before merging. This review adds no upstream reread, cross-pin compatibility, backend/device/model or performance qualification.
Reviewer: Codex / gpt-6.1-sol [effort=ultra], actual independent reviewer.
Dong1017
left a comment
Dong1017
left a comment
There was a problem hiding this comment.
Integration-status update at unchanged head 8028f81cf4d5389978a41856c22300da6e038162, PR API base 25d6d5a603824e3893415841e1709521cd0eeb04, against observed main 0b19c5c3b649ebac939ffecae5583d3a22f598e7.
GitHub now reports mergeable=false. The existing scoped review remains evidence for its reviewed content; it does not clear integration with this newer main. The current-head Ready jobs passed and the queried per-head admission status is successful, but those results do not establish a successfully tested new combination.
Next owner/action: the existing PR/governing-Issue owner should reconcile current main. Preserve both sides' canonical sources and navigation, then regenerate the affected index/coverage projections with the existing generators and validate the resulting head. This is an integration prerequisite; it does not reopen already corrected source findings or add a model/hardware-validation requirement. No conflict-file attribution is made without a merge reproduction.
Verification: remote refs, existing discussion/review and pinned CI inspection only; no code execution, branch change, workflow rerun or merge. This is not a new LGTM or Issue acceptance.
Preserve every current-main canonical source and the original bounded vLLM routing additions, distinct source pins and candidate limits. Commit the canonical union before regenerating discovery projections. Agent-assisted-by: Codex / gpt-6.1-sol [effort=max] (material author) Signed-off-by: GUOGUO <xwdong1998@163.com>
Refresh the native index and coverage projection from the committed canonical union while preserving metadata, source rights and evidence states. Agent-assisted-by: Codex / gpt-6.1-sol [effort=max] (material author) Signed-off-by: GUOGUO <xwdong1998@163.com>
Dong1017
left a comment
Dong1017
left a comment
There was a problem hiding this comment.
LGTM c9af5f8
Independent source and software integration review against main 509531c681479fbdd7ddd0e8b0032239e64a44c1, with canonical source ceb3509942bee863f24eee96e3598c4e007feb20. No actionable findings remain within this bounded navigation scope.
All 70 original navigation insertions across six documents and all 13 links/anchors are preserved. Removing those blocks recovers exact main bytes; both matrix row sequences and the three original vLLM SKILL documents remain intact. The engine and focused scheduler/KV snapshots retain their separate pins, original inspection dates and evidence boundaries. Every other main path/mode/blob is preserved at the source union, including accepted Maya, Metal-engine, Flare and FastH3 owners, registry, licensing, maturity, fixtures, runtime, policy and tests.
The sole generated child changes exactly 75 paths. All 73 discovery records retain accepted main semantics: 71 change provenance only, and two also refresh independently verified canonical SKILL hashes. The 488-input set is complete. Independently recomputing the 146-input coverage digest matches the committed output; coverage changes only that digest.
All four current-head Linux Ready jobs pass. Python 3.10 and 3.11 each report 2355 passed, 11 skipped, 183 subtests passed; both fresh isolated noneditable wheel checks report 121 source-grounded records. Logs explicitly test this head merged into the stated base. Lint, declarations and the fresh exact-target policy result pass; declaration compliance is separate from content review.
SkillEval planning/publication passes on the exact base/head. Execute is inapplicable with the actual labels lacking execution opt-in. Its zero comparisons/calls and 81 uncovered paths provide no efficacy evidence; source/projection inspection is the explicit disposition for this navigation-only change.
Local Windows package validation remained incomplete after timeout. The first attempt's render-entry count is unknown; one completed recovery render/readback is evidenced, without claiming a single historical invocation. Fresh Linux checks supply the software/package evidence above. No upstream reread, model/device/paid execution, cross-pin compatibility, backend conformance, performance or agent-efficacy result is claimed. Candidate/not_run boundaries remain unchanged; this review does not accept or close #91.
Reviewer: Codex / gpt-6.1-sol [effort=ultra], independent review only, with no material edits or authorship. Shared-account COMMENT review, not a formal self-approval.
Refs #91.
Retain the bounded navigation among the existing vLLM engine guide, admission procedure and KV-lifetime procedure while integrating current main. All original 70 inserted lines across six Markdown documents and all 13 relative link occurrences remain unchanged. The engine source
2521446ac8c92937b5a32b7dc2a84ef7fa8e99c5(inspection 2026-10-14) remains distinct from the focused source7f730b4cea9a66520a9200265928ebe016b48a39(supplier inspection 2026-10-15). This integration adds no upstream reread or tested cross-pin stack.Canonical source
ceb3509942bee863f24eee96e3598c4e007feb20(tree912550b2a558be6b1b6b4f3a960374185e2e7d0b) has parents8028f81cf4d5389978a41856c22300da6e038162and actual main509531c681479fbdd7ddd0e8b0032239e64a44c1, in that order. Removing the exact original navigation blocks recovers every complete main document. Both matrix row sequences and every main blob outside those six documents are preserved at the source commit, including the accepted Maya, Metal-engine, Flare and FastH3 material, ownership, registry, licenses, maturity/evidence fields, fixtures, runtime, policy, tools and catalog-cap fix. This PR adds no skill or profile: the source inventory remains 73 registered roots and 117 SKILL documents.Final head
c9af5f81f7b905251c30d3d61923a1f83e2c46b4(treefaf3a56573ea61997328f5701a6d72e8e82213f2) is the generated-only sole child of that source, changing exactly 75 paths: 74 discovery-index files anddocs/skill-coverage.md. The unchanged native index render returned and its same-invocation output/readback assertion passed. Recovered verification binds all 488 indexed source hashes to the committed source. Of 73 records, 71 change provenance only; the two focused records also refresh their actual root-document hashes. Metadata, scope, attribution, license and candidate/not_run states remain unchanged. Coverage text and rows remain identical to main except for input digestsha256:15f6aa9237afc144617167a4bdd44696f9ed141d88e2a63b93a8636d1f8f069d, independently recomputed with the unchanged native digest function. Source immutability, the exact output/path set and the contribution whitespace check passed.The native coverage generation/check logs report successful output and matching sources, but their return codes were not durably recorded before a 300-second Windows post-generation timeout. Full catalog, package-resource and clean installed-package checks were not completed locally. Fresh independent exact-head review and current-head Linux Ready full-render/full-check/installed-package jobs remain required before merge; older reviews and CI remain historical. No local pytest or upstream-helper, model, device or paid-trial execution was added. Candidate and empirical-validation limits remain: this integration establishes no backend/model conformance, quality, performance or agent-efficacy result.