Repository navigation
docs: add cache-aware ASR stream-boundary reference - #301
Conversation
Extend the existing benchmark-design method with pinned NeMo source guidance for startup geometry, decoded progress, slot reuse, reset and EOU/EOF. Keep native test definitions and author reports separate from execution. Agent-assisted-by: Codex / gpt-6.1-sol [effort=max] (author) Signed-off-by: Dong1017 <55723162+Dong1017@users.noreply.github.com>
Refresh the standard knowledge index and coverage view from committed canonical sources, preserving all registered IDs and evidence states. Agent-assisted-by: Codex / gpt-6.1-sol [effort=max] (author) Signed-off-by: Dong1017 <55723162+Dong1017@users.noreply.github.com>
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. |
Skill evaluationPLANNED: not executed; there are no task outcomes yet.
Author verification steps (nonpaid; not yet run)
python tools/build_skill_index.py --source-revision 1e57c434b98137d3e517e15a65163b0e217d2116 --checkThe declared source pin matches this head's canonical library bytes. The command has not run: retain its exit code and full drift/success output. It checks the generated projection, not claim truth or skill efficacy.
Evidence and logs · Retention: 30 days. Plan bindings, budget, frozen exam and advisory detailsAdvisory: obtain missing evidenceRetained evidence availability: planned / unexecuted. No approval or merge authorization is issued.
Model/settings:
Selection/task metadata (not a relevance verdict): {"adapter":"audit","basis":"trusted_path_mapping","matched_paths":["src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md","src/inferrsi/data/skill_library/methods/inference-benchmark-design/metadata.json","src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/cache-aware-asr-stream-boundaries.md"],"tasks":[{"evidence_scope":"http_fixed_workload","id":"audit-threshold","request":"Audit the fixed input report using its contract. Compute the median of per-trial request means, including every request in each trial, and the percentage reduction relative to the baseline aggregate.","split":"holdout"}]}Automatic execution / cumulative PR budget: {"reason":"Trusted configuration or pricing missing, invalid or expired","remaining_usd":"unknown","reserved_or_unresolved_usd":"unknown","spent_or_legacy_usd":"unknown","state":"blocked"}Projection check selection (not an executed result): {"declared_in":"src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md","source_revision":"1e57c434b98137d3e517e15a65163b0e217d2116","status":"ready"}Provider/settings: Uncovered / unsupported paths (84)
|
Dong1017
added
the
area:knowledge
ChatGPT Codex Connector
Dong1017
left a comment
Dong1017
left a comment
There was a problem hiding this comment.
Source-review LGTM for the ASR knowledge slice at 31cddd29 (COMMENTED; no formal GitHub approval).
- The pinned upstream code supports the documented first/steady geometry, decoded clock, slot/reset lifecycle and EOU/EOF distinction. The reference also correctly separates shape-only success from optional native-model coverage and preserves the benchmark's precision, feature-extraction and repetition limits. No blocking source finding was identified.
- The three changed canonical resources match their packaged hashes and material pin. The other 76 catalog cards change only their source URL/commit; candidate and
not_runboundaries remain. Current full-check, clean-package, qualification and authorship/policy checks pass. - SkillEval disposition: its snapshot remains unsupported and execution was skipped. The mapped HTTP fixed-workload audit is not validation of the new ASR state/geometry claims. This source review covers the documentation addition; ASR model/device/service behavior and task efficacy remain unverified. Unsupported snapshot safety must stay intact; no paid run is requested.
- This source-review conclusion applies to the standalone head against current main. If #302 lands first, preserve both canonical additions and regenerate the combined projection before re-review; do not resolve their shared owner files by discarding one source slice.
Evidence and verification limits
Reviewed head: 31cddd29a602c47ac67cc0ecd8d750a2827dd588; API base / observed main: eabad4ad57c0af9407d33da83bb117201d5e5cce. NeMo-Speech.cpp source: 20999cc921d2b14aa58708f61756a8b2e787ba20; material pin: 1e57c434b98137d3e517e15a65163b0e217d2116.
Inspected all changed-file patches, the new reference, metadata/routing/matrices and cited upstream implementation/test/benchmark/attribution regions. Head and tested merge 4b5d4527ad7addcbef349cfa1a4731aaec46dcbf both resolve to tree 379189a235b9a031b4443998e5c1e22534295a37. Ready checks, policy, and planned/unexecuted SkillEval were checked separately.
Independent SHA-256 checks of the reference / SKILL / metadata match the packaged bindings: 4e0ea9c1bf6b096a49a91ad5266e90b8a456f4f5aca15eba8ac270a53c2309b5, 1071b359685b4423fb968ec163912eee4daa24897d0d6567ade7587245a3f2ff, 7e33307f47bdb0afb1eb2e96c7230361c91c5d407b0f44e1011f51915e16534d. Their bytes match the material pin, whose subsequent 79-path delta contains only generated outputs.
Other integration input: #302 head 19ab07ebb7c5fedcde81741a3b6df4e4be56a738. This source-review conclusion does not cover that PR, a future combined/rebased head, upstream decoder/kernel internals outside the cited readset, Issue-level acceptance or performance/efficacy. The GitHub connection uses the same account as the PR author, so GitHub rejected an APPROVE and no formal approval was created. This COMMENTED record is a separate maintenance source review; formal approval requires an eligible reviewer account. This maintenance reviewer did not implement or edit this PR and ran no repository tests, upstream tests, models or devices; software validation is retained hosted evidence.
Refs #91
Add one cache-aware ASR reference to the existing
inference-benchmark-designmethod, grounded in NVIDIA/NeMo-Speech.cpp at20999cc921d2b14aa58708f61756a8b2e787ba20. It separates startup/steady mel geometry and decoded progress, stream-slot reset/reuse, EOU, EOF and full stream reset, and explains the upstream optional-model test boundary.The parent skill, source metadata and two matrices route to that reference. Existing source records, matrix rows, license limits and candidate/not_run states remain. Native catalog and coverage outputs are generated from source commit
1e57c434b98137d3e517e15a65163b0e217d2116; all 77 registered IDs remain, and the other 76 catalog cards change only their source commit and URL.Validation
4b5d4527ad7addcbef349cfa1a4731aaec46dcbfhas the same complete tree as this head. Knowledge-model-policy passed.The upstream tests were read, not executed. Model/device, ASR quality, long-audio/load and service evidence remain not_run. Benchmark values remain author-reported, with Q8_0/FP32, feature-extraction and repetition differences stated. This is one source slice; broader queue completion and candidate promotion are not claimed.
Knowledge authorship declaration