altair823-org/kebab
0
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases 40 Wiki Activity

feat(embed): arctic-embed-l-v2.0 임베더(candle+ollama) #203

Merged
altair823 merged 5 commits from feat/arctic-embedder into main 2026-06-03 06:27:58 +00:00
Conversation 2 Commits 5 Files Changed 16 +1018 -130

5 Commits

Author SHA1 Message Date
altair823
095c9f37a2 docs(smoke): embedding config 블록 v0.26.0 동기화 
SMOKE.md 의 [models.embedding] 예시 주석이 stale: provider 목록에 ollama 누락 +
"candle 은 e5-large 만 지원"(arctic 추가로 더 이상 사실 아님) + endpoint/arctic
미기재. CLAUDE.md §"README Configuration + SMOKE config 블록 동시 갱신" 규칙대로
보완 — provider 4종, arctic 모델(candle/ollama 태그), endpoint(ollama 전용, llm
endpoint fallback), e5↔arctic cascade 주석 추가.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
 2026-06-03 05:11:03 +00:00
altair823
16ddb1dfc3 docs: arctic 임베더 문서 동기화 (README/ARCHITECTURE/HANDOFF/HOTFIXES) 
README Configuration: provider candle/ollama + arctic 모델(candle CLS / ollama 태그)
+ endpoint + e5→arctic cascade 경고. ARCHITECTURE: 백엔드 그래프 노드(embedollama)
+ 임베딩 백엔드 결정표(채택 근거 측정 recall@10 130) + 디렉토리 트리. HANDOFF 1줄.
HOTFIXES 2026-06-03 arctic dated entry(레지스트리/pooling/prefix/cascade + 수동
cosine 0.999984 실측 결과).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
 2026-06-03 04:59:23 +00:00
altair823
72c99c452c feat(config,app): embedding provider=ollama 배선 + endpoint, version 0.26.0 
kebab-config: EmbeddingModelCfg.endpoint: Option<String>(serde default, ollama용,
None→models.llm.endpoint 폴백) + provider 문서에 ollama + env
KEBAB_MODELS_EMBEDDING_ENDPOINT. kebab-app embedder(): provider match 에 ollama
분기(facade 경유). workspace member += kebab-embed-ollama, app dep 추가.
version 0.25.0 → 0.26.0(minor, +Cargo.lock) — 신규 임베딩 백엔드/모델은 CLAUDE.md
§Release 의 surface 변경 트리거.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
 2026-06-03 04:59:23 +00:00
altair823
cbcae69abf feat(embed): candle 모델 레지스트리 + arctic-embed-l-v2.0 (CLS pooling) 
e5 하드코딩(HF_MODEL/SUPPORTED_MODEL/mean/query:+passage:) → 모델 레지스트리
EmbedModelSpec{name,hf_repo,pooling,query_prefix,doc_prefix,dim,version_tag}.
e5(mean, query:/passage:) + arctic(CLS, query:/무접두어). pooling 모델별 분기
(mean=attention-mask-weighted / CLS=hidden[:,0,:]), tokenize/forward/L2 공유.
arctic pooling=CLS 는 HF 1_Pooling/config.json(pooling_mode_cls_token:true) 확인.
model_version 은 arctic 일 때 +arctic-cls 태그(embedding_version cascade 트리거);
e5 는 fastembed-e5 호환(NUMA 드롭인) 위해 plain config.version 유지.

correctness 게이트: tests/arctic_ollama_parity.rs (#[ignore], live Ollama) —
candle arctic vs Ollama snowflake-arctic-embed2 per-sentence 코사인>0.99.
수동 실측 cosine_min=0.999984 (recall@10 130 재현 보장).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
 2026-06-03 04:59:11 +00:00
altair823
7505645008 feat(embed): kebab-embed-ollama 신규 크레이트 — Ollama /api/embed Embedder 
arctic-embed-l-v2.0 의 폴백 백엔드(측정에 쓴 경로 그대로). reqwest::blocking
POST {endpoint}/api/embed {model,input:[...]} → embeddings. batch 48 +
fail-soft 재시도 3, 결과 L2 정규화(Ollama raw 반환 → 일관성), dim 검증.
query/doc prefix 는 모델 태그로 추론(arctic-embed→query:/무접두어, e5→query:/passage:).
model_version=ollama:{model}. endpoint=models.embedding.endpoint ?? models.llm.endpoint.
wiremock 테스트 3종(L2 정규화/dim mismatch/empty no-op) + 단위 5.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
 2026-06-03 04:59:11 +00:00