fix(embed-candle): address round-1 review
- commit track-spec + meta-spec/plan into branch (HIGH: dangling `amends:` ref) - inline parity evidence (cosine 1.0, max_abs_diff 2.01e-7) into HOTFIXES + release notes; drop refs to deleted IMPL_REPORT/SPIKE_REPORT (MEDIUM) - model guard: reject non-e5-large `model` before the 2GB download so model_id() can't mislabel vectors (MEDIUM) + unit test - parity test now covers BOTH query: and passage: prefixes (MEDIUM) - guard encodings.first() index; document zero-attention/pooling invariant; clarify embed_batch prefixing doc (LOW) Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
@@ -69,4 +69,6 @@ double-free 경로를 원천 차단한다. NUMA 노드 바인딩이 더 필요
|
||||
|
||||
듀얼소켓 NUMA 서버에서 `provider=candle` 로 5150-doc ingest 가 double-free
|
||||
없이 EXIT=0 완주하는지가 본 release 의 최종 인수 게이트다 (meta-spec §4.3).
|
||||
패리티 max abs diff 수치는 `IMPL_REPORT.md` 참조.
|
||||
패리티(candle vs onnxruntime): cosine_min = 1.000000, 차원별 max 절대오차 =
|
||||
2.01e-7 — 벡터가 사실상 동일하므로 `embedding_version` 유지(재색인 0). 재현은
|
||||
`crates/kebab-embed-candle/tests/parity.rs` (`--ignored`).
|
||||
|
||||
Reference in New Issue
Block a user