Repository navigation
knowledge: distill Lily with portable AREX workflow - #188
Merged
Merged
Conversation
Distill exact-model layout, strict API/session state and native/benchmark qualification; retain missing compiler and Metal/model gaps. Agent-assisted-by: Codex / gpt-6.1-sol [effort=xhigh] (approved_arex_author; author and integrator) Agent-assisted-by: Codex / gpt-6.1-sol [effort=xhigh] (runtime_memory_author; author) Agent-assisted-by: Codex / gpt-6.1-sol [effort=xhigh] (serving_state_author; author) Agent-assisted-by: Codex / gpt-6.1-sol [effort=xhigh] (qualification_author; author) Signed-off-by: WeiQing Chen <530634352@qq.com>
Bind generated discovery abstracts to the committed exact-model source graph and registry. Agent-assisted-by: Codex / gpt-6.1-sol [effort=xhigh] (approved_arex_author; generated integration) Signed-off-by: WeiQing Chen <530634352@qq.com>
Skill evaluationPLANNED: 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. Evidence and logs · Retention: 30 days. Pipeline evidence only; offline fixed outputs do not establish skill efficacy. Plan bindings, budget and frozen examPlan SHA-256: Model/settings:
Uncovered / unsupported paths (134)
|
5 tasks
Dong1017
reviewed
Oct 27, 2026
Dong1017
left a comment
Dong1017
left a comment
Collaborator
There was a problem hiding this comment.
Scoped early draft review at ec0ed1c, base 4817f92, within #187's partial-delivery scope: no verified actionable blocker in the inspected helpers, graph/reference contracts and tests. Targeted pinned-source checks cover request/budget/state rules, prefix/key ranking and benchmark schema/FNV digest; the MLX version mismatch and supplied-record-only validation limits are accurately disclosed.
Ready CI 40492462061 now passes all four full-check/installed-package jobs on Python 3.10/3.11. Nonblocking cleanup: refresh the PR-body sentence saying exact-head CI/clean wheel are pending. Do not broaden that update into Rust/native qualification: compilation was not run and M5/macOS/Metal/model gaps remain.
The planning-only Skill Eval comment selects zero comparisons. Not a merge-ready LGTM: retain draft status, prescribed ordered main/catalog reconciliation and separate informed candidate admission. Targeted static early review, not exhaustive upstream/hash verification; no PR code, model or hardware execution.
david6666666
marked this pull request as ready for review
October 27, 2026 11:13
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. |
Preserve current main evidence and both admitted graphs; update only candidate discovery, counts and navigation. Original Lily runtime and backend limits unchanged. Signed-off-by: WeiQing Chen <530634352@qq.com> Agent-assisted-by: Codex / approved_arex_author (gpt-6.1-sol, xhigh)
Signed-off-by: WeiQing Chen <530634352@qq.com> Agent-assisted-by: Codex / approved_arex_author (gpt-6.1-sol, xhigh)
5 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Governing issue
Refs #187
User-approved native Rust/Metal adaptation and partial candidate drafting. Final informed limited-candidate admission, sequential merge and issue closure were explicitly authorized by the user on2026-10-27T12:08:19Z; candidate/backend-unverified and known failures remain. Strata explicitly skipped. Merge order DwarfStar → Splash → Lily → Husky; refresh this independent branch against the resulting main/catalog before integration.
Delivered
Pinned Lily subtree operating graph: model-runtime, serving-cache and qualification; exact bounded source hashes/readsets, LICENSE/NOTICE,3 offline helpers and10 assertion-backed cases. Registry/catalog, maintained matrices and exact packaging expectations update together. Actual compatible-agent AREX portable instructions were used, not an unexecuted DisCo CLI/model API run.
Actual checks and boundaries
-Ruff,3 root graph/helper tests and recursive license gate PASS (4 SKILL.md files).
-Author lanes exercised selected config fixtures4 positive/13 negative, request fixtures5 positive/16 negative and production/diagnostic report fixtures plus8 negative cases, with CLI/help/error checks. These test helpers, not Rust parity.
-Stable full scaffold suite: 1332 passed,8 skipped,135 subtests. Ready CI run40492462061 terminal SUCCESS (all4 jobs). Clean noneditable wheel outside checkout passes check_installed with109 records.
-Independent read-only runtime-only case, source40hash/LICENSE/NOTICE,20focused tests, helpers and exact archive wheel review found no outstanding blocker; report retained externally as lily-independent-review.md. This is not a GitHub APPROVE or native validation.
-Cargo/rustc are absent. Rust compilation/native tests were not run, not declared passing or silently skipped.
-BLOCKED_REQUIRED_BACKEND: compatible M5/macOS26/Metal and actual supported checkpoint/model/state execution. CPU-reference code is an oracle within selected GPU tests, not CPU-only GPU validation. Quality/performance/service/skill efficacy unverified.
-Exact Q4 with roleQ8, request lookahead+1, strict-token-prefix-before-key ranking and actually-fed state retention are preserved. Current MLX0.31.2 runner versus later0.32.2 report/harness mismatch and reported long-context prefill disadvantages remain explicit.
-NO_LICENSE is the actual unauthenticated gh fallback, not a legal absence claim. Subtree Apache-2.0 LICENSE and NOTICE are independently retained.
Artifacts:
tests/fixtures/repo_to_skill/lily/; source monorepo1ed972ed3f0bd5616c997c9507c25616c63394fc, subtreelily/; methodac3fe1afa80fb9a09775ecfb2b6cc3ba850a2db6. No toolchain install, weight download, GPU/paid provider invocation or upstream edit.Knowledge authorship
Candidate delivery is not native qualification, campaign PASS or efficacy. Issue closure remains separately assessable; Refs intentionally does not auto-close.
Sequential integration
DwarfStar then Splash merged. Current assessed main54a475d379b640e9df606e154251188f2cb26723 is preserved in full, including independently landed evidence/tooling; none of its frozen evidence or knowledge_operations bytes are altered. Lily runtime graph/fixtures unchanged from reviewed1639db2. Combined registry/navigation/counts63 roots/93 skill documents/111 records; real published source7766bcd57817aa80bfab3ffdb9fd1cd12e902c9c drives regenerated catalog.26 focused graph/index/library tests and ruff PASS. New exact-head Ready CI and independent delta review complete before merge; native blockers remain unchanged.