feat(eval): Phase 2A Diagnose Phase 3+4 — dispatcher + 3 측정 + decision (H3 bge-m3 유지)
phase-2a-embedding-diagnose.md v4 § 6 (dispatcher) + § 7 Phase 3 (51 case 측정) + § 7 Phase 4 (decision)
Round 2 review: round-2-review-mighty-starfish.md (R2-2 + R2-B1 페어 invariant + slug-based resolve)
코드 변경:
- app/services/search/retrieval_service.py:
- CANDIDATE_BACKEND_MAP allowlist (baseline / cand_me5_large_inst / cand_snowflake_l_v2)
- _resolve_backend(slug) → docs_table/chunks_table/embed_endpoint or None
- _embed_query_via_tei() — candidate TEI 엔드포인트 호출 (cache 미사용)
- _VALID_DOCS_TABLE + _VALID_CHUNKS_TABLE regex (R2-B1 2단계 gate)
- _search_vector_docs / _search_vector_chunks: docs_table/chunks_table + snapshot_*_id_max 파라미터
- search_vector + search_vector_multilingual: embedding_backend + snapshot_*_id_max 파라미터 + dispatch log
- app/services/search/search_pipeline.py: run_search() 시그니처 + 4 search_vector* 호출 threading
- app/api/search.py: 3 Query parameter + ValueError → HTTP 400 (allowed list 응답)
- tests/search_eval/run_eval.py: --embedding-backend + --snapshot-doc-id-max + --snapshot-chunk-id-max
+ call_search/call_search_full/evaluate threading + main 3 asyncio.run threading
측정 산출물 (51 case, scored=46, failure=5):
- reports/v0_2_phase2a_baseline_snapshot_2026-05-23.csv (snapshot filter 적용 production path)
- reports/v0_2_phase2a_me5_large_inst_2026-05-23.csv
- reports/v0_2_phase2a_snowflake_l_v2_2026-05-23.csv
- tests/search_eval/baselines/v0_2_phase2a_{baseline_snapshot,me5_large_inst,snowflake_l_v2}_2026-05-23.json (3개)
결과:
| Candidate | NDCG | Δ vs baseline | mixed | korean_only | p50 ms |
|------------------------------------|-----:|--------------:|------:|------------:|-------:|
| bge-m3 (baseline snapshot) | 0.659| — | 0.39 | 0.51 | 464 |
| cand_me5_large_inst | 0.477| -0.182 | 0.17 | 0.47 | 194 |
| cand_snowflake_l_v2 | 0.616| -0.043 | 0.35 | 0.52 | 254 |
Decision (H3): bge-m3 유지. 둘 다 net 회귀.
- mE5-large-instruct: 전 카테고리 회귀 (-0.182). prefix 미적용 변수 — 별 PR PR-2A-mE5-Prefix-Retry 후보.
- snowflake_l_v2: 가벼운 회귀 (-0.043). korean_only +0.01 미세 개선 신호.
- korean_only/mixed 약점 보완은 Phase 2B (Reranker) 또는 Phase 2Q (Query rewrite) 권고.
Decision report: reports/phase_2a_embedding_decision_2026-05-23.md (§ 1~8 포함, Closure gate 16 항목 모두 PASS).
후속 PR 백로그:
- PR-2A-mE5-Prefix-Retry (별 PR)
- PR-2A-Extended-Bge-Mgemma2 (별 PR, v3 결정)
- PR-2A-Cloud-Embedding-Scaffold-1 (Cohere/Voyage scaffold-only, 선택)
- PR-Search-Query-Rewrite-1 (Phase 2Q)
- PR-Search-Reranker-V2-Diagnose (Phase 2B)
- PR-2A-Chunks-Cand-Cleanup-1 (1주 후 cand 테이블 DROP)
production 영향:
- documents / document_chunks 컬럼/row 변경 0
- config.yaml 변경 0 (ollama bge-m3 unchanged)
- 추가된 endpoint = query parameter opt-in (미지정 시 production path 회귀 0)
- smoke 4건 PASS (baseline / baseline+snapshot / cand_me5 / cand_invalid → HTTP 400)
- dispatch log 박제 verify (snapshot_doc/chunk_id_max 박제)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
hyungi
+37
-9
@@ -156,17 +156,45 @@ async def search(
|
||||
description="QueryAnalyzer 활성화 (Phase 2.1, LLM 호출). Phase 2.1은 debug 노출만, 검색 경로 영향 X",
|
||||
),
|
||||
debug: bool = Query(False, description="단계별 candidates + timing 응답에 포함"),
|
||||
embedding_backend: str | None = Query(
|
||||
None,
|
||||
pattern=r"^(baseline|cand_[a-z0-9_]+)$",
|
||||
description="Phase 2A Diagnose dispatcher (R2-2 + R2-B1). slug 만 받음 (raw table name X). baseline|cand_<slug>. 미지정/baseline = production path.",
|
||||
),
|
||||
snapshot_doc_id_max: int | None = Query(
|
||||
None, ge=1,
|
||||
description="Phase 2A snapshot freeze (R2-D + R2-B2). documents.id <= 값 filter. baseline 측정 시에도 동일 filter 적용.",
|
||||
),
|
||||
snapshot_chunk_id_max: int | None = Query(
|
||||
None, ge=1,
|
||||
description="Phase 2A snapshot freeze (R2-D + R2-B2). document_chunks.id <= 값 filter. baseline 측정 시에도 동일 filter 적용.",
|
||||
),
|
||||
):
|
||||
"""문서 검색 — FTS + ILIKE + 벡터 결합 (Phase 3.1 이후 run_search wrapper)"""
|
||||
pr = await run_search(
|
||||
session,
|
||||
q,
|
||||
mode=mode, # type: ignore[arg-type]
|
||||
limit=limit,
|
||||
fusion=fusion,
|
||||
rerank=rerank,
|
||||
analyze=analyze,
|
||||
)
|
||||
try:
|
||||
pr = await run_search(
|
||||
session,
|
||||
q,
|
||||
mode=mode, # type: ignore[arg-type]
|
||||
limit=limit,
|
||||
fusion=fusion,
|
||||
rerank=rerank,
|
||||
analyze=analyze,
|
||||
embedding_backend=embedding_backend,
|
||||
snapshot_doc_id_max=snapshot_doc_id_max,
|
||||
snapshot_chunk_id_max=snapshot_chunk_id_max,
|
||||
)
|
||||
except ValueError as e:
|
||||
# _resolve_backend 가 unknown slug 시 ValueError → HTTP 400
|
||||
return JSONResponse(
|
||||
status_code=400,
|
||||
content={
|
||||
"error_reason": "unknown_embedding_backend",
|
||||
"backend_requested": embedding_backend,
|
||||
"allowed": ["baseline", "cand_me5_large_inst", "cand_snowflake_l_v2"],
|
||||
"detail": str(e),
|
||||
},
|
||||
)
|
||||
|
||||
# 사용자 feedback: 모든 단계 timing은 debug 응답과 별도로 항상 로그로 남긴다
|
||||
timing_str = " ".join(f"{k}={v:.0f}" for k, v in pr.timing_ms.items())
|
||||
|
||||
@@ -22,6 +22,7 @@ from __future__ import annotations
|
||||
|
||||
import asyncio
|
||||
import hashlib
|
||||
import re
|
||||
import time
|
||||
from typing import TYPE_CHECKING, Any
|
||||
|
||||
@@ -48,6 +49,61 @@ _QUERY_EMBED_CACHE: dict[str, dict[str, Any]] = {}
|
||||
QUERY_EMBED_TTL = 86400 # 24h
|
||||
QUERY_EMBED_MAXSIZE = 500
|
||||
|
||||
# ─── Phase 2A Diagnose dispatcher (R2-2 + R2-B1) ──────────────
|
||||
# server-side allowlist map. query parameter 가 raw table name 받지 않음.
|
||||
CANDIDATE_BACKEND_MAP: dict[str, dict[str, str] | None] = {
|
||||
"baseline": None,
|
||||
"cand_me5_large_inst": {
|
||||
"docs_table": "documents_cand_me5_large_inst",
|
||||
"chunks_table": "document_chunks_cand_me5_large_inst",
|
||||
"embed_endpoint": "http://embedding-cand-me5-inst:80/embed",
|
||||
},
|
||||
"cand_snowflake_l_v2": {
|
||||
"docs_table": "documents_cand_snowflake_l_v2",
|
||||
"chunks_table": "document_chunks_cand_snowflake_l_v2",
|
||||
"embed_endpoint": "http://embedding-cand-snowflake-l-v2:80/embed",
|
||||
},
|
||||
}
|
||||
|
||||
# 2단계 gate (R2-B1) — SQL string interpolation 직전 final allowlist.
|
||||
_VALID_DOCS_TABLE = re.compile(r"^(documents|documents_cand_[a-z0-9_]+)$")
|
||||
_VALID_CHUNKS_TABLE = re.compile(r"^(document_chunks|document_chunks_cand_[a-z0-9_]+)$")
|
||||
|
||||
|
||||
def _resolve_backend(slug: str | None) -> dict[str, str] | None:
|
||||
"""slug → (docs_table, chunks_table, embed_endpoint) | None (baseline).
|
||||
|
||||
Raises ValueError on unknown slug (caller 가 HTTP 400 으로 translate).
|
||||
"""
|
||||
if slug is None or slug == "baseline":
|
||||
return None
|
||||
if slug not in CANDIDATE_BACKEND_MAP:
|
||||
raise ValueError(f"unknown_embedding_backend: {slug!r}")
|
||||
cfg = CANDIDATE_BACKEND_MAP[slug]
|
||||
if cfg is None:
|
||||
return None
|
||||
if not all(k in cfg for k in ("docs_table", "chunks_table", "embed_endpoint")):
|
||||
raise RuntimeError(f"candidate_table_pair_misconfigured: {slug}")
|
||||
return cfg
|
||||
|
||||
|
||||
async def _embed_query_via_tei(endpoint: str, text_: str) -> list[float] | None:
|
||||
"""후보 TEI endpoint 호출 (cache 미사용 — slug 별 다른 모델 분포)."""
|
||||
if not text_:
|
||||
return None
|
||||
import httpx
|
||||
try:
|
||||
async with httpx.AsyncClient(timeout=30.0) as c:
|
||||
r = await c.post(endpoint, json={"inputs": [text_], "truncate": True})
|
||||
r.raise_for_status()
|
||||
data = r.json()
|
||||
if not isinstance(data, list) or not data or not isinstance(data[0], list):
|
||||
raise ValueError(f"unexpected TEI shape: {type(data).__name__}")
|
||||
return data[0]
|
||||
except Exception as exc:
|
||||
logger.warning("candidate TEI embed failed endpoint=%s err=%r", endpoint, exc)
|
||||
return None
|
||||
|
||||
|
||||
def _query_embed_key(text_: str) -> str:
|
||||
return hashlib.sha256(f"{text_}|bge-m3".encode("utf-8")).hexdigest()
|
||||
@@ -183,53 +239,78 @@ async def search_text(
|
||||
|
||||
|
||||
async def search_vector(
|
||||
session: AsyncSession, query: str, limit: int
|
||||
session: AsyncSession,
|
||||
query: str,
|
||||
limit: int,
|
||||
*,
|
||||
embedding_backend: str | None = None,
|
||||
snapshot_doc_id_max: int | None = None,
|
||||
snapshot_chunk_id_max: int | None = None,
|
||||
) -> list["SearchResult"]:
|
||||
"""Hybrid 벡터 검색 — doc + chunks 동시 retrieval (Phase 1.2-G).
|
||||
|
||||
Phase 1.2-C 진단:
|
||||
chunks-only는 segment 의미 손실로 자연어 query에서 catastrophic recall.
|
||||
doc embedding은 전체 본문 평균 → recall robust.
|
||||
→ 두 retrieval 동시 사용이 정석.
|
||||
Phase 2A v4 dispatcher (R2-2 + R2-B1):
|
||||
embedding_backend=None|"baseline" → production (documents + document_chunks).
|
||||
snapshot_*_id_max 지정 시 baseline 도 동일 filter (rebaseline measurement).
|
||||
embedding_backend=cand_<slug> → CANDIDATE_BACKEND_MAP 에서 페어 resolve.
|
||||
cand 테이블 자체가 snapshot 범위로 INSERT → snapshot filter 무시 (dispatch log 만 박제).
|
||||
|
||||
데이터 흐름:
|
||||
1. query embedding 1번 (bge-m3)
|
||||
2. asyncio.gather로 두 SQL 동시 호출:
|
||||
- _search_vector_docs: documents.embedding cosine top N
|
||||
- _search_vector_chunks: document_chunks.embedding window partition (doc당 top 2)
|
||||
3. _merge_doc_and_chunk_vectors로 가중치 + dedup:
|
||||
- chunk score * 1.2 (precision)
|
||||
- doc score * 1.0 (recall)
|
||||
- doc_id 기준 dedup, chunks 우선
|
||||
|
||||
Returns:
|
||||
list[SearchResult] — doc_id 중복 제거됨. compress_chunks_to_docs는 그대로 동작.
|
||||
chunks_by_doc은 search.py에서 group_by_doc으로 보존.
|
||||
1. query embedding 1번 (baseline=bge-m3 cache / cand=TEI endpoint no-cache)
|
||||
2. asyncio.gather 로 두 SQL 동시 호출:
|
||||
- _search_vector_docs(docs_table, snapshot_doc_id_max)
|
||||
- _search_vector_chunks(chunks_table, snapshot_chunk_id_max)
|
||||
3. _merge_doc_and_chunk_vectors 가중치 + dedup (chunk 1.2 / doc 1.0).
|
||||
"""
|
||||
client = AIClient()
|
||||
try:
|
||||
query_embedding = await _get_query_embedding(client, query)
|
||||
finally:
|
||||
cfg = _resolve_backend(embedding_backend)
|
||||
|
||||
if cfg is None:
|
||||
docs_table = "documents"
|
||||
chunks_table = "document_chunks"
|
||||
client = AIClient()
|
||||
try:
|
||||
await client.close()
|
||||
except Exception:
|
||||
pass
|
||||
query_embedding = await _get_query_embedding(client, query)
|
||||
finally:
|
||||
try:
|
||||
await client.close()
|
||||
except Exception:
|
||||
pass
|
||||
else:
|
||||
docs_table = cfg["docs_table"]
|
||||
chunks_table = cfg["chunks_table"]
|
||||
query_embedding = await _embed_query_via_tei(cfg["embed_endpoint"], query)
|
||||
|
||||
logger.info(
|
||||
"[embedding-dispatch] backend=%s docs_table=%s chunks_table=%s snapshot_doc_id_max=%s snapshot_chunk_id_max=%s",
|
||||
embedding_backend or "baseline",
|
||||
docs_table,
|
||||
chunks_table,
|
||||
snapshot_doc_id_max,
|
||||
snapshot_chunk_id_max,
|
||||
)
|
||||
|
||||
if query_embedding is None:
|
||||
return []
|
||||
|
||||
embedding_str = str(query_embedding)
|
||||
|
||||
# 두 SQL 병렬 호출 — 각각 별도 session 사용 (asyncpg connection은 statement 단위 직렬)
|
||||
Session = async_sessionmaker(engine)
|
||||
|
||||
async def _docs_call() -> list["SearchResult"]:
|
||||
async with Session() as s:
|
||||
return await _search_vector_docs(s, embedding_str, limit * 4)
|
||||
return await _search_vector_docs(
|
||||
s, embedding_str, limit * 4,
|
||||
docs_table=docs_table,
|
||||
snapshot_doc_id_max=snapshot_doc_id_max,
|
||||
)
|
||||
|
||||
async def _chunks_call() -> list["SearchResult"]:
|
||||
async with Session() as s:
|
||||
return await _search_vector_chunks(s, embedding_str, limit * 4)
|
||||
return await _search_vector_chunks(
|
||||
s, embedding_str, limit * 4,
|
||||
chunks_table=chunks_table,
|
||||
snapshot_chunk_id_max=snapshot_chunk_id_max,
|
||||
)
|
||||
|
||||
doc_results, chunk_results = await asyncio.gather(_docs_call(), _chunks_call())
|
||||
|
||||
@@ -237,93 +318,116 @@ async def search_vector(
|
||||
|
||||
|
||||
async def _search_vector_docs(
|
||||
session: AsyncSession, embedding_str: str, limit: int
|
||||
session: AsyncSession,
|
||||
embedding_str: str,
|
||||
limit: int,
|
||||
*,
|
||||
docs_table: str = "documents",
|
||||
snapshot_doc_id_max: int | None = None,
|
||||
) -> list["SearchResult"]:
|
||||
"""documents.embedding 직접 검색 — recall robust (자연어 매칭).
|
||||
"""documents (또는 documents_cand_<slug>).embedding 직접 검색.
|
||||
|
||||
chunks가 없는 doc도 매칭 가능. score는 cosine similarity (1 - distance).
|
||||
chunk_id/chunk_index/section_title은 None.
|
||||
docs_table = "documents": production path. snapshot_doc_id_max 지정 시 id <= max filter.
|
||||
docs_table = "documents_cand_<slug>": 후보 path. cand 테이블이 이미 snapshot 범위로 INSERT됨 →
|
||||
snapshot_doc_id_max 무시. metadata 는 production documents 와 JOIN.
|
||||
|
||||
R2-B1 final gate: docs_table 은 _VALID_DOCS_TABLE allowlist 통과 후 SQL interpolation.
|
||||
"""
|
||||
from api.search import SearchResult # 순환 import 회피
|
||||
|
||||
result = await session.execute(
|
||||
text("""
|
||||
SELECT
|
||||
id,
|
||||
title,
|
||||
ai_domain,
|
||||
ai_summary,
|
||||
file_format,
|
||||
(1 - (embedding <=> cast(:embedding AS vector))) AS score,
|
||||
left(extracted_text, 1200) AS snippet,
|
||||
'vector_doc' AS match_reason,
|
||||
NULL::bigint AS chunk_id,
|
||||
NULL::integer AS chunk_index,
|
||||
NULL::text AS section_title
|
||||
if not _VALID_DOCS_TABLE.match(docs_table):
|
||||
raise RuntimeError(f"invalid_docs_table: {docs_table!r}")
|
||||
|
||||
params: dict[str, Any] = {"embedding": embedding_str, "limit": limit}
|
||||
|
||||
if docs_table == "documents":
|
||||
snapshot_clause = ""
|
||||
if snapshot_doc_id_max is not None:
|
||||
snapshot_clause = " AND id <= :snapshot_doc_id_max"
|
||||
params["snapshot_doc_id_max"] = snapshot_doc_id_max
|
||||
sql = f"""
|
||||
SELECT id, title, ai_domain, ai_summary, file_format,
|
||||
(1 - (embedding <=> cast(:embedding AS vector))) AS score,
|
||||
left(extracted_text, 1200) AS snippet,
|
||||
'vector_doc' AS match_reason,
|
||||
NULL::bigint AS chunk_id, NULL::integer AS chunk_index, NULL::text AS section_title
|
||||
FROM documents
|
||||
WHERE embedding IS NOT NULL AND deleted_at IS NULL
|
||||
WHERE embedding IS NOT NULL AND deleted_at IS NULL{snapshot_clause}
|
||||
ORDER BY embedding <=> cast(:embedding AS vector)
|
||||
LIMIT :limit
|
||||
"""),
|
||||
{"embedding": embedding_str, "limit": limit},
|
||||
)
|
||||
"""
|
||||
else:
|
||||
# candidate: docs_table 은 (doc_id, embed_input, embed_input_hash, embedding) 만 보유 → JOIN documents
|
||||
sql = f"""
|
||||
SELECT d.id, d.title, d.ai_domain, d.ai_summary, d.file_format,
|
||||
(1 - (c.embedding <=> cast(:embedding AS vector))) AS score,
|
||||
left(d.extracted_text, 1200) AS snippet,
|
||||
'vector_doc' AS match_reason,
|
||||
NULL::bigint AS chunk_id, NULL::integer AS chunk_index, NULL::text AS section_title
|
||||
FROM {docs_table} c
|
||||
JOIN documents d ON d.id = c.doc_id
|
||||
WHERE d.deleted_at IS NULL
|
||||
ORDER BY c.embedding <=> cast(:embedding AS vector)
|
||||
LIMIT :limit
|
||||
"""
|
||||
result = await session.execute(text(sql), params)
|
||||
return [SearchResult(**row._mapping) for row in result]
|
||||
|
||||
|
||||
async def _search_vector_chunks(
|
||||
session: AsyncSession, embedding_str: str, limit: int
|
||||
session: AsyncSession,
|
||||
embedding_str: str,
|
||||
limit: int,
|
||||
*,
|
||||
chunks_table: str = "document_chunks",
|
||||
snapshot_chunk_id_max: int | None = None,
|
||||
) -> list["SearchResult"]:
|
||||
"""document_chunks.embedding 검색 + window partition (doc당 top 2 chunks).
|
||||
"""document_chunks (또는 document_chunks_cand_<slug>).embedding window partition.
|
||||
|
||||
SQL 흐름:
|
||||
1. inner CTE topk: ivfflat 인덱스로 top-K chunks 추출
|
||||
2. ranked CTE: doc_id PARTITION + ROW_NUMBER (score 내림차순)
|
||||
3. outer: rn <= 2 (doc당 max 2 chunks) + JOIN documents
|
||||
chunks_table = "document_chunks": production path. snapshot_chunk_id_max 지정 시 c.id <= max filter.
|
||||
chunks_table = "document_chunks_cand_<slug>": cand 테이블 (이미 snapshot 범위로 INSERT) → filter 무시.
|
||||
|
||||
R2-B1 final gate: chunks_table 은 _VALID_CHUNKS_TABLE allowlist 통과 후 SQL interpolation.
|
||||
"""
|
||||
from api.search import SearchResult # 순환 import 회피
|
||||
|
||||
if not _VALID_CHUNKS_TABLE.match(chunks_table):
|
||||
raise RuntimeError(f"invalid_chunks_table: {chunks_table!r}")
|
||||
|
||||
inner_k = max(limit * 5, 500)
|
||||
result = await session.execute(
|
||||
text("""
|
||||
WITH topk AS (
|
||||
SELECT
|
||||
c.id AS chunk_id,
|
||||
c.doc_id,
|
||||
c.chunk_index,
|
||||
c.section_title,
|
||||
c.text,
|
||||
c.embedding <=> cast(:embedding AS vector) AS dist
|
||||
FROM document_chunks c
|
||||
WHERE c.embedding IS NOT NULL
|
||||
ORDER BY c.embedding <=> cast(:embedding AS vector)
|
||||
LIMIT :inner_k
|
||||
),
|
||||
ranked AS (
|
||||
SELECT
|
||||
chunk_id, doc_id, chunk_index, section_title, text, dist,
|
||||
ROW_NUMBER() OVER (PARTITION BY doc_id ORDER BY dist ASC) AS rn
|
||||
FROM topk
|
||||
)
|
||||
SELECT
|
||||
d.id AS id,
|
||||
d.title AS title,
|
||||
d.ai_domain AS ai_domain,
|
||||
d.ai_summary AS ai_summary,
|
||||
d.file_format AS file_format,
|
||||
(1 - r.dist) AS score,
|
||||
left(r.text, 1200) AS snippet,
|
||||
'vector_chunk' AS match_reason,
|
||||
r.chunk_id AS chunk_id,
|
||||
r.chunk_index AS chunk_index,
|
||||
r.section_title AS section_title
|
||||
FROM ranked r
|
||||
JOIN documents d ON d.id = r.doc_id
|
||||
WHERE r.rn <= 2 AND d.deleted_at IS NULL
|
||||
ORDER BY r.dist
|
||||
LIMIT :limit
|
||||
"""),
|
||||
{"embedding": embedding_str, "inner_k": inner_k, "limit": limit},
|
||||
)
|
||||
params: dict[str, Any] = {"embedding": embedding_str, "inner_k": inner_k, "limit": limit}
|
||||
|
||||
snapshot_clause = ""
|
||||
if chunks_table == "document_chunks" and snapshot_chunk_id_max is not None:
|
||||
snapshot_clause = " AND c.id <= :snapshot_chunk_id_max"
|
||||
params["snapshot_chunk_id_max"] = snapshot_chunk_id_max
|
||||
|
||||
sql = f"""
|
||||
WITH topk AS (
|
||||
SELECT c.id AS chunk_id, c.doc_id, c.chunk_index, c.section_title, c.text,
|
||||
c.embedding <=> cast(:embedding AS vector) AS dist
|
||||
FROM {chunks_table} c
|
||||
WHERE c.embedding IS NOT NULL{snapshot_clause}
|
||||
ORDER BY c.embedding <=> cast(:embedding AS vector)
|
||||
LIMIT :inner_k
|
||||
),
|
||||
ranked AS (
|
||||
SELECT chunk_id, doc_id, chunk_index, section_title, text, dist,
|
||||
ROW_NUMBER() OVER (PARTITION BY doc_id ORDER BY dist ASC) AS rn
|
||||
FROM topk
|
||||
)
|
||||
SELECT d.id AS id, d.title AS title, d.ai_domain AS ai_domain,
|
||||
d.ai_summary AS ai_summary, d.file_format AS file_format,
|
||||
(1 - r.dist) AS score, left(r.text, 1200) AS snippet,
|
||||
'vector_chunk' AS match_reason,
|
||||
r.chunk_id AS chunk_id, r.chunk_index AS chunk_index, r.section_title AS section_title
|
||||
FROM ranked r
|
||||
JOIN documents d ON d.id = r.doc_id
|
||||
WHERE r.rn <= 2 AND d.deleted_at IS NULL
|
||||
ORDER BY r.dist
|
||||
LIMIT :limit
|
||||
"""
|
||||
result = await session.execute(text(sql), params)
|
||||
return [SearchResult(**row._mapping) for row in result]
|
||||
|
||||
|
||||
@@ -369,6 +473,10 @@ async def search_vector_multilingual(
|
||||
session: AsyncSession,
|
||||
normalized_queries: list[dict],
|
||||
limit: int,
|
||||
*,
|
||||
embedding_backend: str | None = None,
|
||||
snapshot_doc_id_max: int | None = None,
|
||||
snapshot_chunk_id_max: int | None = None,
|
||||
) -> list["SearchResult"]:
|
||||
"""Phase 2.2 — 다국어 normalized_queries 배열로 vector retrieval.
|
||||
|
||||
@@ -393,18 +501,24 @@ async def search_vector_multilingual(
|
||||
if not normalized_queries:
|
||||
return []
|
||||
|
||||
# 1. 각 lang별 embedding 병렬 (cache hit 활용)
|
||||
client = AIClient()
|
||||
try:
|
||||
embed_tasks = [
|
||||
_get_query_embedding(client, q["text"]) for q in normalized_queries
|
||||
]
|
||||
embeddings = await asyncio.gather(*embed_tasks)
|
||||
finally:
|
||||
# 1. 각 lang별 embedding 병렬 (baseline=AIClient.embed cache / cand=TEI endpoint no-cache)
|
||||
_cfg_for_embed = _resolve_backend(embedding_backend)
|
||||
if _cfg_for_embed is None:
|
||||
client = AIClient()
|
||||
try:
|
||||
await client.close()
|
||||
except Exception:
|
||||
pass
|
||||
embed_tasks = [
|
||||
_get_query_embedding(client, q["text"]) for q in normalized_queries
|
||||
]
|
||||
embeddings = await asyncio.gather(*embed_tasks)
|
||||
finally:
|
||||
try:
|
||||
await client.close()
|
||||
except Exception:
|
||||
pass
|
||||
else:
|
||||
ep = _cfg_for_embed["embed_endpoint"]
|
||||
embed_tasks = [_embed_query_via_tei(ep, q["text"]) for q in normalized_queries]
|
||||
embeddings = await asyncio.gather(*embed_tasks)
|
||||
|
||||
# embedding 실패한 query는 skip (weight 재정규화 없이 조용히 drop)
|
||||
per_query_plan: list[tuple[dict, str]] = []
|
||||
@@ -417,17 +531,38 @@ async def search_vector_multilingual(
|
||||
if not per_query_plan:
|
||||
return []
|
||||
|
||||
# 2. 각 embedding에 대해 doc + chunks 병렬 retrieval
|
||||
# 2. multilingual dispatcher resolve (모든 lang query 가 동일 backend 사용)
|
||||
cfg = _resolve_backend(embedding_backend)
|
||||
docs_table = cfg["docs_table"] if cfg else "documents"
|
||||
chunks_table = cfg["chunks_table"] if cfg else "document_chunks"
|
||||
logger.info(
|
||||
"[embedding-dispatch] backend=%s docs_table=%s chunks_table=%s snapshot_doc_id_max=%s snapshot_chunk_id_max=%s multilingual=true",
|
||||
embedding_backend or "baseline",
|
||||
docs_table,
|
||||
chunks_table,
|
||||
snapshot_doc_id_max,
|
||||
snapshot_chunk_id_max,
|
||||
)
|
||||
|
||||
# 3. 각 embedding에 대해 doc + chunks 병렬 retrieval
|
||||
Session = async_sessionmaker(engine)
|
||||
|
||||
async def _one_query(q_meta: dict, embedding_str: str) -> list["SearchResult"]:
|
||||
async def _docs() -> list["SearchResult"]:
|
||||
async with Session() as s:
|
||||
return await _search_vector_docs(s, embedding_str, limit * 4)
|
||||
return await _search_vector_docs(
|
||||
s, embedding_str, limit * 4,
|
||||
docs_table=docs_table,
|
||||
snapshot_doc_id_max=snapshot_doc_id_max,
|
||||
)
|
||||
|
||||
async def _chunks() -> list["SearchResult"]:
|
||||
async with Session() as s:
|
||||
return await _search_vector_chunks(s, embedding_str, limit * 4)
|
||||
return await _search_vector_chunks(
|
||||
s, embedding_str, limit * 4,
|
||||
chunks_table=chunks_table,
|
||||
snapshot_chunk_id_max=snapshot_chunk_id_max,
|
||||
)
|
||||
|
||||
doc_r, chunk_r = await asyncio.gather(_docs(), _chunks())
|
||||
return _merge_doc_and_chunk_vectors(doc_r, chunk_r)
|
||||
|
||||
@@ -121,6 +121,9 @@ async def run_search(
|
||||
fusion: str = DEFAULT_FUSION,
|
||||
rerank: bool = True,
|
||||
analyze: bool = False,
|
||||
embedding_backend: str | None = None,
|
||||
snapshot_doc_id_max: int | None = None,
|
||||
snapshot_chunk_id_max: int | None = None,
|
||||
) -> PipelineResult:
|
||||
"""검색 파이프라인 실행.
|
||||
|
||||
@@ -214,9 +217,19 @@ async def run_search(
|
||||
if mode == "vector":
|
||||
t0 = time.perf_counter()
|
||||
if use_multilingual:
|
||||
raw_chunks = await search_vector_multilingual(session, normalized_queries, limit)
|
||||
raw_chunks = await search_vector_multilingual(
|
||||
session, normalized_queries, limit,
|
||||
embedding_backend=embedding_backend,
|
||||
snapshot_doc_id_max=snapshot_doc_id_max,
|
||||
snapshot_chunk_id_max=snapshot_chunk_id_max,
|
||||
)
|
||||
else:
|
||||
raw_chunks = await search_vector(session, q, limit)
|
||||
raw_chunks = await search_vector(
|
||||
session, q, limit,
|
||||
embedding_backend=embedding_backend,
|
||||
snapshot_doc_id_max=snapshot_doc_id_max,
|
||||
snapshot_chunk_id_max=snapshot_chunk_id_max,
|
||||
)
|
||||
timing["vector_ms"] = (time.perf_counter() - t0) * 1000
|
||||
if not raw_chunks:
|
||||
notes.append("vector_search_returned_empty (AI client error or no embeddings)")
|
||||
@@ -231,9 +244,19 @@ async def run_search(
|
||||
if mode == "hybrid":
|
||||
t1 = time.perf_counter()
|
||||
if use_multilingual:
|
||||
raw_chunks = await search_vector_multilingual(session, normalized_queries, limit)
|
||||
raw_chunks = await search_vector_multilingual(
|
||||
session, normalized_queries, limit,
|
||||
embedding_backend=embedding_backend,
|
||||
snapshot_doc_id_max=snapshot_doc_id_max,
|
||||
snapshot_chunk_id_max=snapshot_chunk_id_max,
|
||||
)
|
||||
else:
|
||||
raw_chunks = await search_vector(session, q, limit)
|
||||
raw_chunks = await search_vector(
|
||||
session, q, limit,
|
||||
embedding_backend=embedding_backend,
|
||||
snapshot_doc_id_max=snapshot_doc_id_max,
|
||||
snapshot_chunk_id_max=snapshot_chunk_id_max,
|
||||
)
|
||||
timing["vector_ms"] = (time.perf_counter() - t1) * 1000
|
||||
|
||||
# chunk-level → doc-level 압축 (raw chunks는 chunks_by_doc에 보존)
|
||||
|
||||
@@ -0,0 +1,97 @@
|
||||
# Phase 2A Embedding Decision Report (2026-05-23)
|
||||
|
||||
> Parent: `phase-2a-embedding-diagnose.md` v4
|
||||
>
|
||||
> Round 2 review: `round-2-review-mighty-starfish.md` 채택
|
||||
>
|
||||
> 본 보고서 = Phase 4 산출물. Decision Tree H1~H4 중 권고 1개 + 후속 PR 후보.
|
||||
|
||||
## 1. Summary
|
||||
|
||||
| | Value |
|
||||
|---|---|
|
||||
| baseline (bge-m3, snapshot 범위) | NDCG@10 (graded) **0.659** / mixed 0.39 / korean_only 0.51 / failure 0/5 / p50 464ms / p95 1582ms |
|
||||
| baseline rebaseline (snapshot filter 적용) | 위와 동일 (snapshot 범위 = corpus 전부와 거의 동일, 측정 가능 확인) |
|
||||
| 후보 2종 측정 완료 | me5_large_inst (mE5-instruct), snowflake_l_v2 (Snowflake Arctic L v2.0) |
|
||||
| 이관 (별 PR) | bge_mgemma2 (9B FP16 → 16GB GPU OOM risk → PR-2A-Extended-Bge-Mgemma2) |
|
||||
| 폐기 | ko_me5 (HF 401 Unauthorized) |
|
||||
|
||||
## 2. 후보별 Δ NDCG (vs baseline rebaseline)
|
||||
|
||||
| Candidate | overall NDCG | Δ overall | mixed | Δ mixed | korean_only | Δ korean | standards | english_only | exam | failure | p50 ms | p95 ms |
|
||||
|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|
|
||||
| **bge-m3 snapshot rebaseline** | **0.659** | — | **0.39** | — | **0.51** | — | 0.87 | 0.78 | 0.74 | 0/5 | 464 | 1582 |
|
||||
| mE5-large-instruct | 0.477 | **-0.182** | 0.17 | **-0.22** | 0.47 | -0.04 | 0.54 | 0.63 | 0.62 | 0/5 | 194 | 1348 |
|
||||
| snowflake-arctic-embed-l-v2.0 | 0.616 | -0.043 | 0.35 | -0.04 | 0.52 | +0.01 | 0.87 | 0.74 | 0.56 | 0/5 | 254 | 1412 |
|
||||
|
||||
**관찰**:
|
||||
- **mE5-large-instruct**: 전 카테고리 큰 회귀. Δ -0.182 overall. mixed 절반 회귀 (0.39 → 0.17). standards 도 큰 회귀 (0.87 → 0.54). 단 latency p50 270ms 단축 (mE5 의 512 context = 적은 compute).
|
||||
- **snowflake_l_v2**: 가벼운 회귀 (Δ -0.043). standards / korean_only 거의 동일. mixed 약간 회귀. exam 명확 회귀 (-0.18). latency p50 210ms 단축.
|
||||
|
||||
**ambiguous note** (LLM 단독 결정, [[feedback_user_block_minimize]]):
|
||||
- mE5-instruct 는 query input 에 `Instruct: <task>\nQuery: <query>` prefix 권장 (intfloat 모델 카드). 본 PR 측정은 plain query → prefix 효과 미반영. prefix 적용 시 +0.05~0.15 회복 가능성 있으나 측정 외 — 별 PR 후보 `PR-2A-mE5-Prefix-Retry`.
|
||||
- snowflake_l_v2 의 한국어 specific 벤치마크 공개 부재. 본 측정 = 사실상 한국어 specific 첫 audit. korean_only +0.01 미세 개선 신호 있으나 통계적 의미 없음 (n=9, 0.51 vs 0.52).
|
||||
|
||||
## 3. Latency 영향
|
||||
|
||||
- mE5 (512 ctx): p50 464 → 194 (**−270ms**), p95 1582 → 1348 (−234ms). 빠름.
|
||||
- snowflake (8192 ctx): p50 464 → 254 (**−210ms**), p95 1582 → 1412 (−170ms). 빠름.
|
||||
- 둘 다 baseline 보다 빠르지만 quality 회귀. **trade-off favor quality** ([[feedback_quant_expectation_not_hard_gate]] 룰, 정량 hard gate 없으나 +0 NDCG 회복 시 latency 30% 단축 가치는 별 평가).
|
||||
|
||||
## 4. Decision (H3 — bge-m3 유지)
|
||||
|
||||
| | H1 swap 권고 | H2 query rewrite 보완 | **H3 bge-m3 유지 (✅ 선택)** | H4 latency 회귀 |
|
||||
|---|---|---|---|---|
|
||||
| 조건 | mixed + korean_only 둘 다 명확 개선 | korean_only 만 개선 / mixed 미개선 | 모든 후보 bge-m3 대비 개선 없음 | latency p95 ≥ 3000ms |
|
||||
| 결과 | ❌ 둘 다 회귀 | ❌ korean_only 미세 개선 (+0.01) 만, mixed 회귀 | ✅ **확정** | ❌ 둘 다 baseline 보다 빠름 |
|
||||
|
||||
**최종 권고**: **bge-m3 유지** (Apply PR 진입 안 함).
|
||||
|
||||
근거:
|
||||
- mE5 -0.182 / snowflake -0.043 — 둘 다 net 회귀.
|
||||
- korean_only 약점 보완 도구로 embedding swap 보다 query rewrite (Phase 2Q) 또는 reranker 튜닝 (Phase 2B) 가 더 유망.
|
||||
- mE5 prefix retry 는 별 PR 로 분리 — diagnose 본 PR scope 외.
|
||||
|
||||
## 5. Apply / 보완 / 보류 권고
|
||||
|
||||
- **Apply** (production embedding swap): **하지 않음**.
|
||||
- **보완** (다른 트랙): **Phase 2B (Reranker)** 또는 **Phase 2Q (Query rewrite)** 우선 — korean_only / mixed 약점 다른 layer 에서 공략.
|
||||
- **보류** (Phase 2A-Extended): bge_mgemma2 (별 PR), mE5 prefix retry (별 PR), Cloud embedding (Cohere/Voyage) scaffold-only (별 PR).
|
||||
|
||||
## 6. 후보 cleanup 일정
|
||||
|
||||
- 미선택 후보 4 테이블 (`documents_cand_me5_large_inst` / `document_chunks_cand_me5_large_inst` / `documents_cand_snowflake_l_v2` / `document_chunks_cand_snowflake_l_v2`) = **1주 dormant 유지** (mE5 prefix retry / Phase 2Q 비교 baseline 사용 가능성).
|
||||
- 1주 후 별 chore `PR-2A-Chunks-Cand-Cleanup-1` 에서 DROP + 컨테이너 docker-compose.override 제거.
|
||||
|
||||
## 7. 후속 PR 후보 (백로그)
|
||||
|
||||
| PR 가칭 | trigger | scope |
|
||||
|---|---|---|
|
||||
| `PR-2A-mE5-Prefix-Retry` | 본 PR 결과 + ambiguous note | mE5-instruct query prefix 적용 후 재측정. 페어 reindex 재실행 + 51 case 재측정. 본 PR 의 dispatcher 재사용 (`CANDIDATE_BACKEND_MAP` 에 신규 slug 추가). |
|
||||
| `PR-2A-Extended-Bge-Mgemma2` | v3 short-list swap 결정 | 9B FP16 OOM 회피 (quantization int8 또는 sentence-transformers). 별 컨테이너 + reindex + 측정. |
|
||||
| `PR-2A-Cloud-Embedding-Scaffold-1` | (선택) self-hosted 무개선 확정 | Cohere / Voyage scaffold-only (`[[feedback_scaffold_first_for_external_cost_pr]]`). 실비 0. |
|
||||
| `PR-Search-Query-Rewrite-1` (Phase 2Q) | korean_only / mixed 약점 보완 | 자연어 query → SQL/keyword 강화. |
|
||||
| `PR-Search-Reranker-V2-Diagnose` (Phase 2B) | korean_only / mixed 약점 보완 | bge-reranker-v2-m3 swap 후보 측정. |
|
||||
| `PR-2A-Chunks-Cand-Cleanup-1` | 본 PR closure 후 1주 | 4 cand 테이블 DROP + 컨테이너 정리. |
|
||||
|
||||
## 8. Closure gate verify (§ 8 본 plan)
|
||||
|
||||
- [x] G0-1 + G0-2 fixture 박제 (Phase 1 closure 시 commit `943ac5f`)
|
||||
- [x] snapshot json 박제 (`v0_2_phase2a_snapshot_2026-05-23.json`, commit `a67df0a`)
|
||||
- [x] 2 후보 (me5_large_inst + snowflake_l_v2) 51 case 측정 완료 (`overall.n = 46`, 5 failure 제외)
|
||||
- [x] baseline rebaseline 51 case 측정 완료 (snapshot filter 적용)
|
||||
- [x] 후보별 baseline json 2개 + baseline_snapshot json 1개 박제
|
||||
- [x] documents_cand_<slug> row count = 21365 verify (2 후보 동일)
|
||||
- [x] document_chunks_cand_<slug> row count = 30605 verify (2 후보 동일)
|
||||
- [x] baseline rebaseline 측정도 동일 snapshot_doc/chunk_id_max filter 통과 verify (dispatch log)
|
||||
- [x] dispatcher 호출 시 unknown slug → HTTP 400 verify (smoke test `cand_invalid` 통과)
|
||||
- [x] decision md 박제 (본 파일)
|
||||
- [x] Apply 권고 1줄 작성 (H3)
|
||||
- [x] production embedding (bge-m3 ollama) 변경 0 verify (`docker compose ps`, `ollama list`, `config.yaml` diff 0)
|
||||
- [x] production `documents` row count + embedding 변경 0 verify
|
||||
- [x] production `document_chunks` row count + content 변경 0 verify
|
||||
- [x] 후보 cleanup 일정 명시 (1주 dormant → `PR-2A-Chunks-Cand-Cleanup-1`)
|
||||
- [x] dispatch log audit (silent fallback 0, `embedding_backend_unavailable` 0, snapshot id 박제 verify)
|
||||
- [x] DOCSRV_TOKEN 만료 사고 0 (3 측정 모두 15분 이내 완주)
|
||||
|
||||
**Phase 2A Diagnose PR closure: PASS**.
|
||||
@@ -0,0 +1,52 @@
|
||||
label,id,category,legacy_category,intent,domain_hint,language,ocr_derived,failure_expected,query,relevant_ids,graded_relevance,returned_ids_top10,latency_ms,recall_at_10,mrr_at_10,ndcg_at_10,top3_hit,graded_ndcg_at_10,graded_recall_at_10_t2,graded_recall_at_10_t3,error
|
||||
single,kw_001,standards,exact_keyword,fact_lookup,document,ko,0,0,산업안전보건법 제6장,3856;3868;3879,3856:3;3868:2;3879:2,3868;3879;3856;3851;4041;3890;3917;3863;3908;3855,418.5,1.000,1.000,1.000,1,0.808,1.000,1.000,
|
||||
single,kw_002,standards,exact_keyword,fact_lookup,document,ko,0,0,중대재해 처벌 등에 관한 법률 제2장 중대산업재해,3917;3921,3917:3;3921:2,3921;3917;3919;10573;10571;3916;3874;3918;3854;3922,464.3,1.000,1.000,1.000,1,0.834,1.000,1.000,
|
||||
single,kw_003,standards,exact_keyword,fact_lookup,document,ko,0,0,화학물질관리법 유해화학물질 영업자,3981,3981:3,3981;3985;3984;3993;3857;3978;3983;3957;3980;3903,291.5,1.000,1.000,1.000,1,1.000,1.000,1.000,
|
||||
single,kw_004,standards,exact_keyword,fact_lookup,document,ko,0,0,근로기준법 안전과 보건,4041,4041:3,4041;3852;3851;3877;3905;3878;3858;3903;3781;3881,478.4,1.000,1.000,1.000,1,1.000,1.000,1.000,
|
||||
single,kw_005,standards,exact_keyword,fact_lookup,document,ko,0,0,산업안전보건기준에 관한 규칙 보호구,3888,3888:3,10570;3888;3912;3913;3911;3905;3909;3906;3910;3893,489.4,1.000,0.500,0.631,1,0.631,1.000,1.000,
|
||||
single,nl_001,korean_only,natural_language_ko,semantic_search,document,ko,0,0,기계로 인한 산업재해 관련 법령,3856;3868;3879;3854,3854:1;3856:3;3868:2;3879:2,5244;5249;3897;3863;5253;3856;3895;3867;3879;3851,505.3,0.500,0.167,0.257,0,0.314,0.667,1.000,
|
||||
single,nl_002,korean_only,natural_language_ko,semantic_search,document,ko,0,0,사업주가 도급을 줄 때 산업재해를 예방하기 위해 해야 할 일,3855;3867;3878,3855:3;3867:2;3878:2,3855;5227;3854;5244;3851;3867;3878;3863;3908;10573,460.3,1.000,1.000,0.793,1,0.873,1.000,1.000,
|
||||
single,nl_003,korean_only,natural_language_ko,semantic_search,document,ko,0,0,유해화학물질을 다루는 회사가 지켜야 할 안전 의무,3980;3981;3982,3980:2;3981:2;3982:2,3980;3904;3903;3909;3905;3981;3760;5253;3985;3896,400.1,0.667,1.000,0.636,1,0.636,0.667,0.000,
|
||||
single,nl_004,korean_only,natural_language_ko,semantic_search,document,ko,0,0,중대재해가 발생했을 때 경영책임자가 처벌받는 기준,3916;3917;3920;3921,3916:2;3917:3;3920:2;3921:2,10572;3917;3916;3918;5227;3854;3877;3922;5240;5226,363.9,0.500,0.500,0.441,1,0.506,0.500,1.000,
|
||||
single,nl_005,korean_only,natural_language_ko,semantic_search,document,ko,0,0,안전보건교육은 누가 받아야 하고 어떤 내용을 다루는가,3853;3865,3853:3;3865:2,3853;3876;5249;5234;4025;6675;11677;10573;3757;3811,593.8,0.500,1.000,0.613,1,0.787,0.500,1.000,
|
||||
single,news_001,korean_only,news_ko,semantic_search,news,ko,0,0,이란과 미국의 군사 충돌,4303;4304;4307;4316;4322;4323;4327;4335,4303:2;4304:2;4307:2;4316:2;4322:2;4323:2;4327:2;4335:2,16081;18077;22048;12213;23984;15793;4321;21273;21276;4307,465.2,0.125,0.100,0.073,1,0.073,0.125,0.000,
|
||||
single,news_002,korean_only,news_ko,semantic_search,news,ko,0,0,호르무즈 해협 봉쇄,4316;4320;4322;4327,4316:3;4320:2;4322:2;4327:2,16532;21706;15922;17123;21890;22049;4346;9022;4767;6067,284.7,0.000,0.000,0.000,0,0.000,0.000,0.000,
|
||||
single,misc_001,korean_only,other_domain,fact_lookup,document,ko,0,0,강체의 평면 운동학,4063;4065,4063:3;4065:2,4063;4064;4065;4066;4071;4068;4069;5063;5105;4067,568.8,1.000,1.000,0.920,1,0.956,1.000,1.000,
|
||||
single,misc_002,korean_only,other_domain,semantic_search,document,ko,0,0,질점의 운동역학,4060;4061;4062,4060:2;4061:2;4062:2,4070;4062;4059;4058;4060;4063;4066;4071;4064;5095,515.5,0.667,0.500,0.478,1,0.478,0.667,0.000,
|
||||
single,news_003,english_only,news_en,semantic_search,news,en,0,0,Trump Iran ultimatum,4258;4260;4262,4258:2;4260:2;4262:2,4775;23446;4776;4202;4679;24382;21155;4668;4199;21855,269.1,0.000,0.000,0.000,1,0.000,0.000,0.000,
|
||||
single,cl_001,mixed,crosslingual_ko_en,semantic_search,document,mixed,0,0,기계 안전 가드 설계 원리,3770;3856,3770:3;3856:2,5239;3770;3817;4540;5244;3762;3789;5249;3791;3793,545.1,0.500,0.500,0.387,1,0.497,0.500,1.000,
|
||||
single,cl_002,mixed,crosslingual_ko_en,semantic_search,document,mixed,0,0,산업 안전 입문서,3755;3775;3776;3777,3755:2;3775:2;3776:2;3777:2,5244;5236;5249;5229;3755;3774;3761;5230;10573;3787,470.0,0.250,0.200,0.151,1,0.151,0.250,0.000,
|
||||
single,cl_003,mixed,crosslingual_ko_en,semantic_search,document,mixed,0,0,전기 안전 위험,3772;3790,3772:2;3790:2,3790;3772;5260;3897;5248;3771;3769;11671;13936;3755,739.8,1.000,1.000,1.000,1,1.000,1.000,0.000,
|
||||
single,news_004,mixed,news_fr,semantic_search,news,mixed,0,0,guerre en Iran,4199;4202;4210;4361;4363;4507;4519;4521,4199:2;4202:2;4210:2;4361:2;4363:2;4507:2;4519:2;4521:2,22342;19576;17069;15924;16935;23149;16019;16462;16010;4776,324.6,0.000,0.000,0.000,1,0.000,0.000,0.000,
|
||||
single,news_005,mixed,news_crosslingual,semantic_search,news,mixed,0,0,이란 미국 전쟁 글로벌 반응,4202;4258;4262;4536;4303;4304;4316,4202:2;4258:2;4262:2;4303:2;4304:2;4316:2;4536:2,16761;21275;16927;20893;16771;17242;4329;20886;4457;4307,513.2,0.000,0.000,0.000,1,0.000,0.000,0.000,
|
||||
single,fail_001,failure_expected,failure_expected,semantic_search,document,mixed,0,1,Rust async runtime tokio scheduler 내부 구조,,,5161;5262;23732;24155;4546;20758;5145;4547;3774;5180,394.6,0.000,0.000,0.000,1,0.000,0.000,0.000,
|
||||
single,fail_002,failure_expected,failure_expected,semantic_search,document,ko,0,1,양자컴퓨터 큐비트 디코히어런스,,,16289;5089;5092;5250;22202;20507;5070;5118;5173;23605,301.6,0.000,0.000,0.000,1,0.000,0.000,0.000,
|
||||
single,fail_003,failure_expected,failure_expected,semantic_search,news,ko,0,1,재즈 보컬리스트 빌리 홀리데이,,,20022;20470;4634;15361;16059;9102;23336;18286;16218;5738,264.4,0.000,0.000,0.000,1,0.000,0.000,0.000,
|
||||
single,kw_006,standards,standards,fact_lookup,document,ko,0,0,산업안전보건기준에 관한 규칙 작업장 통로,3886;3887,3886:3;3887:2,3886;3902;3887;3895;3898;3885;3905;3908;3911;3915,338.8,1.000,1.000,0.920,1,0.956,1.000,1.000,
|
||||
single,kw_007,standards,standards,fact_lookup,document,ko,0,0,산업안전보건기준 폭발 화재 위험물 누출 방지,3896;3766,3766:1;3896:3,3896;13930;3895;3911;13929;3866;3903;3890;3910;3909,295.6,0.500,1.000,0.613,1,0.917,1.000,1.000,
|
||||
single,kw_008,standards,standards,fact_lookup,document,ko,0,0,고압가스 안전관리법 전문,4025;4026,4025:3;4026:2,11644;11579;4025;4026;11645;13750;11676;13299;13749;13766,444.4,1.000,0.333,0.571,1,0.539,1.000,1.000,
|
||||
single,kw_009,standards,standards,fact_lookup,document,ko,0,0,KGS FP111 가스설비 배관설비 기준,13305,13305:3,13305;13311;13306;13312;13302;13304;13309;13299;13313;13918,420.0,1.000,1.000,1.000,1,1.000,1.000,1.000,
|
||||
single,kw_010,standards,standards,fact_lookup,document,ko,0,0,KGS FU551 가스설비 압력조정기 가스계량기,13652,13652:3,13652;11689;13657;13655;13656;13649;13651;13752;13659;13650,333.7,1.000,1.000,1.000,1,1.000,1.000,1.000,
|
||||
single,kw_011,standards,standards,fact_lookup,document,ko,0,0,산업안전보건기준 전기로 인한 위험 방지,3897;3772,3772:1;3897:3,3897;3895;3902;3896;3887;13935;13938;3877;3900;3899,450.7,0.500,1.000,0.613,1,0.917,1.000,1.000,
|
||||
single,en_001,english_only,english_only,semantic_search,document,en,0,0,pressure vessel flange design,5144;5136,5136:2;5144:3,5144;5136;5180;5193;5140;5137;5149;5178;5207;5148,1618.0,1.000,1.000,1.000,1,1.000,1.000,1.000,
|
||||
single,en_002,english_only,english_only,semantic_search,document,en,0,0,ASME Section VIII introduction,5204;5206,5204:3;5206:2,5204;5180;5208;5210;5143;5206;5137;5207;5182;5140,1458.1,1.000,1.000,0.832,1,0.907,1.000,1.000,
|
||||
single,en_003,english_only,english_only,semantic_search,document,en,0,0,impact test requirements ASME,5205;5148,5148:1;5205:3,5205;5204;5178;5214;5224;5210;5148;5145;5186;5190,1600.2,1.000,1.000,0.818,1,0.961,1.000,1.000,
|
||||
single,en_004,english_only,english_only,semantic_search,document,en,0,0,design of vessel supports,5149,5149:3,5149;5141;5137;5139;5136;5140;5186;5178;5145;5143,1564.2,1.000,1.000,1.000,1,1.000,1.000,1.000,
|
||||
single,en_005,english_only,english_only,semantic_search,document,en,0,0,hydrogen piping ASME code,5178,5178:3,5178;5210;5180;5182;5143;5204;5211;5207;5185;5186,1311.5,1.000,1.000,1.000,1,1.000,1.000,1.000,
|
||||
single,en_006,english_only,english_only,semantic_search,document,en,0,0,ASME welding qualification requirements,5209;3771,3771:1;5209:3,5204;5224;5208;5209;5205;5178;5180;5225;5187;5186,1417.4,0.500,0.250,0.264,0,0.395,1.000,1.000,
|
||||
single,en_007,english_only,english_only,semantic_search,document,en,0,0,pressure vessel fabrication and inspection,5208;5145,5145:2;5208:3,5208;5189;5192;5180;5187;5186;5212;5188;5182;5137,1664.3,0.500,1.000,0.613,1,0.787,0.500,1.000,
|
||||
single,en_008,english_only,english_only,semantic_search,document,en,0,0,Industrial Safety and Health Management ergonomics,3763;3755,3755:1;3763:3,3763;3759;3774;3755;3818;3812;3778;3756;3761;3771,1076.8,1.000,1.000,0.877,1,0.974,1.000,1.000,
|
||||
single,cl_004,mixed,mixed,semantic_search,document,mixed,0,0,ASME 압력용기 설계 실무,5207;5210;5139;5135,5135:2;5139:2;5207:3;5210:3,5139;5207;5204;5225;5206;5208;5210;5137;5182;5145,749.0,0.750,1.000,0.767,1,0.686,0.750,1.000,
|
||||
single,cl_005,mixed,mixed,semantic_search,document,mixed,0,0,ASME 용접 코드 해설,5224;5209,5209:2;5224:3,5224;5222;5225;5209;5180;5204;5210;5205;5178;5143,709.8,1.000,1.000,0.877,1,0.932,1.000,1.000,
|
||||
single,cl_006,mixed,mixed,semantic_search,document,mixed,0,0,pressure vessel Korean industrial safety regulation,4025;3856;5136,3856:2;4025:2;5136:1,5210;5186;13913;5143;13760;13749;5145;5180;5240;5137,741.3,0.000,0.000,0.000,1,0.000,0.000,0.000,
|
||||
single,cl_007,mixed,mixed,semantic_search,document,mixed,0,0,KGS 코드 LPG safety standard,11647;11689;11645;4025,4025:1;11645:2;11647:3;11689:2,11647;13760;13674;13669;13774;13773;13675;13755;13924;13772,371.6,0.250,1.000,0.390,1,0.647,0.333,1.000,
|
||||
single,cl_008,mixed,mixed,semantic_search,document,mixed,0,0,수소 가스 안전 기준 hydrogen safety,5178;5169,5169:2;5178:3,10575;11671;11649;11648;13915;5241;11563;5173;5177;11653,636.2,0.000,0.000,0.000,0,0.000,0.000,0.000,
|
||||
single,exam_001,exam,exam,fact_lookup,document,ko,0,0,레이놀즈수 정의 공식,11504;11505,11504:3;11505:2,11504;11533;5081;11509;11476;11486;5064;3788;5134;5075,528.8,0.500,1.000,0.613,1,0.787,0.500,1.000,
|
||||
single,exam_002,exam,exam,fact_lookup,document,ko,0,0,탱크 바닥 구멍 유체 유속 토리첼리,11500;11495;11496,11495:2;11496:2;11500:3,11500;11495;11501;5139;5090;5178;11515;5210;11493;11719,329.0,0.667,1.000,0.765,1,0.856,0.667,1.000,
|
||||
single,exam_003,exam,exam,fact_lookup,document,ko,0,0,이상기체 음속 마하수,11514;11515,11514:3;11515:2,11514;11475;5090;5084;11531;11476;11473;5093;11479;5124,585.5,0.500,1.000,0.613,1,0.787,0.500,1.000,
|
||||
single,exam_004,exam,exam,fact_lookup,document,ko,0,0,고압가스 용기 내압시험 영구증가량,11591;11644;11691,11591:3;11644:2;11691:2,11591;11664;13948;13660;5177;13652;11665;13917;11660;13752,351.0,0.333,1.000,0.469,1,0.674,0.333,1.000,
|
||||
single,exam_005,exam,exam,fact_lookup,document,ko,0,0,도시가스 배관 매설 이격거리,11627;11625;11646,11625:2;11627:3;11646:1,11627;11658;11600;11625;11692;13918;13751;5177;13653;13753,359.0,0.667,1.000,0.671,1,0.883,1.000,1.000,
|
||||
single,exam_006,exam,exam,fact_lookup,document,ko,0,0,LPG 저장탱크 안전거리 분말소화기,11617;11669;11620,11617:3;11620:1;11669:2,11595;11616;13669;11617;11649;11655;11690;11658;11653;11689,299.7,0.333,0.250,0.202,0,0.321,0.500,1.000,
|
||||
single,exam_007,exam,exam,fact_lookup,document,ko,0,0,오리피스 차압식 유량계,11712;11711;11503,11503:2;11711:2;11712:3,11711;11712;11503;11500;11713;11714;13930;11717;11701;11502,357.6,1.000,1.000,1.000,1,0.858,1.000,1.000,
|
||||
single,fail_004,failure_expected,failure_expected,fact_lookup,document,ko,0,1,KGS AC999 임의 가스 코드,,,11691;11693;11692;13665;13661;13664;13666;13670;13773;13934,360.5,0.000,0.000,0.000,1,0.000,0.000,0.000,
|
||||
single,fail_005,failure_expected,failure_expected,fact_lookup,document,ko,0,1,초전도 안전 관리법 시행규칙,,,4026;5236;3977;3971;3966;4018;3972;3973;3974;3895,420.6,0.000,0.000,0.000,1,0.000,0.000,0.000,
|
||||
|
@@ -0,0 +1,52 @@
|
||||
label,id,category,legacy_category,intent,domain_hint,language,ocr_derived,failure_expected,query,relevant_ids,graded_relevance,returned_ids_top10,latency_ms,recall_at_10,mrr_at_10,ndcg_at_10,top3_hit,graded_ndcg_at_10,graded_recall_at_10_t2,graded_recall_at_10_t3,error
|
||||
single,kw_001,standards,exact_keyword,fact_lookup,document,ko,0,0,산업안전보건법 제6장,3856;3868;3879,3856:3;3868:2;3879:2,10573;3868;3854;3879;3890;3856;3971;3867;3910;3876,129.9,1.000,0.500,0.665,0,0.546,1.000,1.000,
|
||||
single,kw_002,standards,exact_keyword,fact_lookup,document,ko,0,0,중대재해 처벌 등에 관한 법률 제2장 중대산업재해,3917;3921,3917:3;3921:2,3921;3917;10573;3919;3923;3916;3874;3854;3918;3922,243.4,1.000,1.000,1.000,1,0.834,1.000,1.000,
|
||||
single,kw_003,standards,exact_keyword,fact_lookup,document,ko,0,0,화학물질관리법 유해화학물질 영업자,3981,3981:3,3985;3981;3978;3984;3983;3980;3904;3869;3979;3988,115.7,1.000,0.500,0.631,1,0.631,1.000,1.000,
|
||||
single,kw_004,standards,exact_keyword,fact_lookup,document,ko,0,0,근로기준법 안전과 보건,4041,4041:3,4041;3851;3858;5227;3881;4036;4040;4045;10573;3853,273.0,1.000,1.000,1.000,1,1.000,1.000,1.000,
|
||||
single,kw_005,standards,exact_keyword,fact_lookup,document,ko,0,0,산업안전보건기준에 관한 규칙 보호구,3888,3888:3,3895;3890;3901;3899;3910;3905;3915;3911;3894;3913,476.0,0.000,0.000,0.000,0,0.000,0.000,0.000,
|
||||
single,nl_001,korean_only,natural_language_ko,semantic_search,document,ko,0,0,기계로 인한 산업재해 관련 법령,3856;3868;3879;3854,3854:1;3856:3;3868:2;3879:2,5244;5253;3895;3868;3879;3856;3921;3854;3923;3915,289.2,1.000,0.250,0.581,0,0.486,1.000,1.000,
|
||||
single,nl_002,korean_only,natural_language_ko,semantic_search,document,ko,0,0,사업주가 도급을 줄 때 산업재해를 예방하기 위해 해야 할 일,3855;3867;3878,3855:3;3867:2;3878:2,3867;3855;10571;10573;3917;5231;3878;3918;3851;3854,182.6,1.000,1.000,0.922,1,0.810,1.000,1.000,
|
||||
single,nl_003,korean_only,natural_language_ko,semantic_search,document,ko,0,0,유해화학물질을 다루는 회사가 지켜야 할 안전 의무,3980;3981;3982,3980:2;3981:2;3982:2,11686;3980;3903;3869;3918;3981;3985;3854;3896;3955,127.6,0.667,0.500,0.463,1,0.463,0.667,0.000,
|
||||
single,nl_004,korean_only,natural_language_ko,semantic_search,document,ko,0,0,중대재해가 발생했을 때 경영책임자가 처벌받는 기준,3916;3917;3920;3921,3916:2;3917:3;3920:2;3921:2,10571;10572;10573;3918;3917;3921;3877;3923;3854;11677,99.8,0.500,0.200,0.290,0,0.323,0.500,1.000,
|
||||
single,nl_005,korean_only,natural_language_ko,semantic_search,document,ko,0,0,안전보건교육은 누가 받아야 하고 어떤 내용을 다루는가,3853;3865,3853:3;3865:2,11677;10573;5234;3876;13926;5249;13935;11676;3853;3921,274.7,0.500,0.111,0.185,0,0.237,0.500,1.000,
|
||||
single,news_001,korean_only,news_ko,semantic_search,news,ko,0,0,이란과 미국의 군사 충돌,4303;4304;4307;4316;4322;4323;4327;4335,4303:2;4304:2;4307:2;4316:2;4322:2;4323:2;4327:2;4335:2,22907;21273;23757;21276;23571;18077;16526;15922;15911;15919,194.4,0.000,0.000,0.000,1,0.000,0.000,0.000,
|
||||
single,news_002,korean_only,news_ko,semantic_search,news,ko,0,0,호르무즈 해협 봉쇄,4316;4320;4322;4327,4316:3;4320:2;4322:2;4327:2,20240;23572;22067;20898;16532;18087;17123;15922;15918;4327,100.8,0.250,0.100,0.113,0,0.074,0.250,0.000,
|
||||
single,misc_001,korean_only,other_domain,fact_lookup,document,ko,0,0,강체의 평면 운동학,4063;4065,4063:3;4065:2,4063;4071;4066;4064;4065;5105;4067;5086;5064;4068,198.8,1.000,1.000,0.850,1,0.918,1.000,1.000,
|
||||
single,misc_002,korean_only,other_domain,semantic_search,document,ko,0,0,질점의 운동역학,4060;4061;4062,4060:2;4061:2;4062:2,4062;4060;4064;4059;4070;4058;4068;4061;4066;5086,171.1,1.000,1.000,0.913,1,0.913,1.000,0.000,
|
||||
single,news_003,english_only,news_en,semantic_search,news,en,0,0,Trump Iran ultimatum,4258;4260;4262,4258:2;4260:2;4262:2,21441;22075;20274;23242;21897;20440;18428;16404;17008;16823,70.8,0.000,0.000,0.000,1,0.000,0.000,0.000,
|
||||
single,cl_001,mixed,crosslingual_ko_en,semantic_search,document,mixed,0,0,기계 안전 가드 설계 원리,3770;3856,3770:3;3856:2,5239;5244;11736;11638;11675;11634;11656;11737;11648;5236,214.6,0.000,0.000,0.000,0,0.000,0.000,0.000,
|
||||
single,cl_002,mixed,crosslingual_ko_en,semantic_search,document,mixed,0,0,산업 안전 입문서,3755;3775;3776;3777,3755:2;3775:2;3776:2;3777:2,5249;11637;6674;5230;11737;11638;11676;3876;3867;3859,153.1,0.000,0.000,0.000,1,0.000,0.000,0.000,
|
||||
single,cl_003,mixed,crosslingual_ko_en,semantic_search,document,mixed,0,0,전기 안전 위험,3772;3790,3772:2;3790:2,13938;11565;13937;11572;11737;13769;13943;3897;5260;4020,516.6,0.000,0.000,0.000,1,0.000,0.000,0.000,
|
||||
single,news_004,mixed,news_fr,semantic_search,news,mixed,0,0,guerre en Iran,4199;4202;4210;4361;4363;4507;4519;4521,4199:2;4202:2;4210:2;4361:2;4363:2;4507:2;4519:2;4521:2,23149;25056;22342;16351;16842;17069;16457;4688;4670;4507,119.4,0.125,0.100,0.073,1,0.073,0.125,0.000,
|
||||
single,news_005,mixed,news_crosslingual,semantic_search,news,mixed,0,0,이란 미국 전쟁 글로벌 반응,4202;4258;4262;4536;4303;4304;4316,4202:2;4258:2;4262:2;4303:2;4304:2;4316:2;4536:2,22907;17128;17242;19111;16526;16761;4761;4307;4457;4452,313.2,0.000,0.000,0.000,1,0.000,0.000,0.000,
|
||||
single,fail_001,failure_expected,failure_expected,semantic_search,document,mixed,0,1,Rust async runtime tokio scheduler 내부 구조,,,11733;11698;11735;11613;11711;11736;24508;24268;5215;20238,90.2,0.000,0.000,0.000,1,0.000,0.000,0.000,
|
||||
single,fail_002,failure_expected,failure_expected,semantic_search,document,ko,0,1,양자컴퓨터 큐비트 디코히어런스,,,11513;11510;11711;11739;11736;11738;11508;11735;11523;11509,44.4,0.000,0.000,0.000,1,0.000,0.000,0.000,
|
||||
single,fail_003,failure_expected,failure_expected,semantic_search,news,ko,0,1,재즈 보컬리스트 빌리 홀리데이,,,11510;11735;23082;23336;11711;11513;11507;11712;11698;11508,43.0,0.000,0.000,0.000,1,0.000,0.000,0.000,
|
||||
single,kw_006,standards,standards,fact_lookup,document,ko,0,0,산업안전보건기준에 관한 규칙 작업장 통로,3886;3887,3886:3;3887:2,3895;3913;3901;3899;3910;3905;3890;3915;3908;3911,160.4,0.000,0.000,0.000,0,0.000,0.000,0.000,
|
||||
single,kw_007,standards,standards,fact_lookup,document,ko,0,0,산업안전보건기준 폭발 화재 위험물 누출 방지,3896;3766,3766:1;3896:3,11565;11637;11572;11636;11568;11673;11678;11634;3896;3894,110.8,0.500,0.111,0.185,0,0.276,1.000,1.000,
|
||||
single,kw_008,standards,standards,fact_lookup,document,ko,0,0,고압가스 안전관리법 전문,4025;4026,4025:3;4026:2,11644;11676;11693;13941;13299;13749;13766;13306;4026;13302,139.7,0.500,0.111,0.185,0,0.102,0.500,0.000,
|
||||
single,kw_009,standards,standards,fact_lookup,document,ko,0,0,KGS FP111 가스설비 배관설비 기준,13305,13305:3,13305;13941;13311;13913;13653;13307;13306;13317;13310;13313,111.7,1.000,1.000,1.000,1,1.000,1.000,1.000,
|
||||
single,kw_010,standards,standards,fact_lookup,document,ko,0,0,KGS FU551 가스설비 압력조정기 가스계량기,13652,13652:3,13652;13941;11689;13752;13655;13319;13653;11690;11612;11693,73.1,1.000,1.000,1.000,1,1.000,1.000,1.000,
|
||||
single,kw_011,standards,standards,fact_lookup,document,ko,0,0,산업안전보건기준 전기로 인한 위험 방지,3897;3772,3772:1;3897:3,13938;3897;13937;3854;3895;3901;3915;3890;3899;3867,194.5,0.500,0.500,0.387,1,0.579,1.000,1.000,
|
||||
single,en_001,english_only,english_only,semantic_search,document,en,0,0,pressure vessel flange design,5144;5136,5136:2;5144:3,5144;5136;5147;5137;5210;5180;5140;5149;5133;5145,1273.4,1.000,1.000,1.000,1,1.000,1.000,1.000,
|
||||
single,en_002,english_only,english_only,semantic_search,document,en,0,0,ASME Section VIII introduction,5204;5206,5204:3;5206:2,5210;5137;5212;5178;5144;5180;5145;5147;11634;5141,1116.0,0.000,0.000,0.000,0,0.000,0.000,0.000,
|
||||
single,en_003,english_only,english_only,semantic_search,document,en,0,0,impact test requirements ASME,5205;5148,5148:1;5205:3,5205;5178;5214;5210;5148;5145;5190;5186;5192;5209,1449.5,1.000,1.000,0.850,1,0.968,1.000,1.000,
|
||||
single,en_004,english_only,english_only,semantic_search,document,en,0,0,design of vessel supports,5149,5149:3,5149;5141;5136;5178;5186;5145;5207;5140;5143;5204,1383.7,1.000,1.000,1.000,1,1.000,1.000,1.000,
|
||||
single,en_005,english_only,english_only,semantic_search,document,en,0,0,hydrogen piping ASME code,5178,5178:3,5178;5143;5180;5148;5207;5210;5179;5182;5133;5208,1312.7,1.000,1.000,1.000,1,1.000,1.000,1.000,
|
||||
single,en_006,english_only,english_only,semantic_search,document,en,0,0,ASME welding qualification requirements,5209;3771,3771:1;5209:3,5178;5180;5205;5209;5212;5145;5186;4835;4826;5182,1252.0,0.500,0.250,0.264,0,0.395,1.000,1.000,
|
||||
single,en_007,english_only,english_only,semantic_search,document,en,0,0,pressure vessel fabrication and inspection,5208;5145,5145:2;5208:3,5208;5187;5191;5186;5188;5190;5148;5182;5143;5210,1480.7,0.500,1.000,0.613,1,0.787,0.500,1.000,
|
||||
single,en_008,english_only,english_only,semantic_search,document,en,0,0,Industrial Safety and Health Management ergonomics,3763;3755,3755:1;3763:3,3774;3755;3763;3812;3815;3756;3758;3757;3773;3770,744.7,1.000,0.500,0.693,1,0.541,1.000,1.000,
|
||||
single,cl_004,mixed,mixed,semantic_search,document,mixed,0,0,ASME 압력용기 설계 실무,5207;5210;5139;5135,5135:2;5139:2;5207:3;5210:3,5139;5207;5135;5204;5225;5133;5195;5224;5180;5209,529.6,0.750,1.000,0.832,1,0.628,0.750,0.500,
|
||||
single,cl_005,mixed,mixed,semantic_search,document,mixed,0,0,ASME 용접 코드 해설,5224;5209,5209:2;5224:3,5224;5222;5225;5209;5135;5133;5208;5205;11601;5206,540.8,1.000,1.000,0.877,1,0.932,1.000,1.000,
|
||||
single,cl_006,mixed,mixed,semantic_search,document,mixed,0,0,pressure vessel Korean industrial safety regulation,4025;3856;5136,3856:2;4025:2;5136:1,11585;11633;11634;11737;11590;13942;13919;13917;11693;3895,560.0,0.000,0.000,0.000,1,0.000,0.000,0.000,
|
||||
single,cl_007,mixed,mixed,semantic_search,document,mixed,0,0,KGS 코드 LPG safety standard,11647;11689;11645;4025,4025:1;11645:2;11647:3;11689:2,13764;11652;11690;11693;13941;11689;11650;11669;11651;11653,152.6,0.250,0.167,0.139,0,0.099,0.333,0.000,
|
||||
single,cl_008,mixed,mixed,semantic_search,document,mixed,0,0,수소 가스 안전 기준 hydrogen safety,5178;5169,5169:2;5178:3,5173;5177;11671;11559;11651;11672;11588;11477;11652;5179,403.3,0.000,0.000,0.000,0,0.000,0.000,0.000,
|
||||
single,exam_001,exam,exam,fact_lookup,document,ko,0,0,레이놀즈수 정의 공식,11504;11505,11504:3;11505:2,11533;11504;5090;11482;11509;11505;11513;11510;11514;11534,236.5,1.000,0.500,0.605,1,0.617,1.000,1.000,
|
||||
single,exam_002,exam,exam,fact_lookup,document,ko,0,0,탱크 바닥 구멍 유체 유속 토리첼리,11500;11495;11496,11495:2;11496:2;11500:3,11500;11501;11503;11719;11517;11713;11715;11712;11594;11514,105.6,0.333,1.000,0.469,1,0.674,0.333,1.000,
|
||||
single,exam_003,exam,exam,fact_lookup,document,ko,0,0,이상기체 음속 마하수,11514;11515,11514:3;11515:2,11514;11479;5090;11475;11518;11515;11516;11517;11694;11478,339.3,1.000,1.000,0.832,1,0.907,1.000,1.000,
|
||||
single,exam_004,exam,exam,fact_lookup,document,ko,0,0,고압가스 용기 내압시험 영구증가량,11591;11644;11691,11591:3;11644:2;11691:2,13660;11691;11591;13943;13942;13917;13653;13752;11579;13753,132.6,0.667,0.500,0.531,1,0.519,0.667,1.000,
|
||||
single,exam_005,exam,exam,fact_lookup,document,ko,0,0,도시가스 배관 매설 이격거리,11627;11625;11646,11625:2;11627:3;11646:1,11658;11692;11627;11655;13753;11651;11670;11646;11690;11617,151.5,0.667,0.333,0.383,1,0.406,0.500,1.000,
|
||||
single,exam_006,exam,exam,fact_lookup,document,ko,0,0,LPG 저장탱크 안전거리 분말소화기,11617;11669;11620,11617:3;11620:1;11669:2,11658;11595;11690;11669;11581;11639;11655;11650;11649;11617,113.8,0.667,0.250,0.338,0,0.353,1.000,1.000,
|
||||
single,exam_007,exam,exam,fact_lookup,document,ko,0,0,오리피스 차압식 유량계,11712;11711;11503,11503:2;11711:2;11712:3,11711;11712;11503;11713;11719;11715;11717;11716;11613;11502,149.2,1.000,1.000,1.000,1,0.858,1.000,1.000,
|
||||
single,fail_004,failure_expected,failure_expected,fact_lookup,document,ko,0,1,KGS AC999 임의 가스 코드,,,11691;11647;11668;5177;11693;11692;13665;13661;13666;13663,160.4,0.000,0.000,0.000,1,0.000,0.000,0.000,
|
||||
single,fail_005,failure_expected,failure_expected,fact_lookup,document,ko,0,1,초전도 안전 관리법 시행규칙,,,5260;4026;3977;3971;3966;3972;3973;3974;3895;4019,218.3,0.000,0.000,0.000,1,0.000,0.000,0.000,
|
||||
|
@@ -0,0 +1,52 @@
|
||||
label,id,category,legacy_category,intent,domain_hint,language,ocr_derived,failure_expected,query,relevant_ids,graded_relevance,returned_ids_top10,latency_ms,recall_at_10,mrr_at_10,ndcg_at_10,top3_hit,graded_ndcg_at_10,graded_recall_at_10_t2,graded_recall_at_10_t3,error
|
||||
single,kw_001,standards,exact_keyword,fact_lookup,document,ko,0,0,산업안전보건법 제6장,3856;3868;3879,3856:3;3868:2;3879:2,3879;3868;3890;3863;3856;3908;3851;4041;3862;3873,236.8,1.000,1.000,0.947,0,0.731,1.000,1.000,
|
||||
single,kw_002,standards,exact_keyword,fact_lookup,document,ko,0,0,중대재해 처벌 등에 관한 법률 제2장 중대산업재해,3917;3921,3917:3;3921:2,3921;3917;3919;3923;10573;10571;3916;3874;3918;3854,243.0,1.000,1.000,1.000,1,0.834,1.000,1.000,
|
||||
single,kw_003,standards,exact_keyword,fact_lookup,document,ko,0,0,화학물질관리법 유해화학물질 영업자,3981,3981:3,3985;3981;3984;3978;3983;3986;3957;3980;3992;3869,118.4,1.000,0.500,0.631,1,0.631,1.000,1.000,
|
||||
single,kw_004,standards,exact_keyword,fact_lookup,document,ko,0,0,근로기준법 안전과 보건,4041,4041:3,4041;3852;3851;3858;3881;4036;4040;4045;3913;3912,281.7,1.000,1.000,1.000,1,1.000,1.000,1.000,
|
||||
single,kw_005,standards,exact_keyword,fact_lookup,document,ko,0,0,산업안전보건기준에 관한 규칙 보호구,3888,3888:3,3888;3893;3887;3897;3892;3890;3896;3895;3902;3889,287.1,1.000,1.000,1.000,1,1.000,1.000,1.000,
|
||||
single,nl_001,korean_only,natural_language_ko,semantic_search,document,ko,0,0,기계로 인한 산업재해 관련 법령,3856;3868;3879;3854,3854:1;3856:3;3868:2;3879:2,3878;5249;3863;3868;3856;3879;3867;3921;3851;3923,288.6,0.750,0.250,0.458,0,0.468,1.000,1.000,
|
||||
single,nl_002,korean_only,natural_language_ko,semantic_search,document,ko,0,0,사업주가 도급을 줄 때 산업재해를 예방하기 위해 해야 할 일,3855;3867;3878,3855:3;3867:2;3878:2,3917;5246;3854;5227;3851;3867;3855;3878;3863;10573,233.7,1.000,0.167,0.472,0,0.418,1.000,1.000,
|
||||
single,nl_003,korean_only,natural_language_ko,semantic_search,document,ko,0,0,유해화학물질을 다루는 회사가 지켜야 할 안전 의무,3980;3981;3982,3980:2;3981:2;3982:2,3980;3904;3903;3760;3985;3916;3851;3978;3905;3981,175.1,0.667,1.000,0.605,1,0.605,0.667,0.000,
|
||||
single,nl_004,korean_only,natural_language_ko,semantic_search,document,ko,0,0,중대재해가 발생했을 때 경영책임자가 처벌받는 기준,3916;3917;3920;3921,3916:2;3917:3;3920:2;3921:2,10572;10573;3917;3916;3918;10571;5244;3919;5227;3854,163.1,0.500,0.333,0.363,1,0.410,0.500,1.000,
|
||||
single,nl_005,korean_only,natural_language_ko,semantic_search,document,ko,0,0,안전보건교육은 누가 받아야 하고 어떤 내용을 다루는가,3853;3865,3853:3;3865:2,3853;5249;4025;5240;10573;11677;3876;3757;3811;3921,358.4,0.500,1.000,0.613,1,0.787,0.500,1.000,
|
||||
single,news_001,korean_only,news_ko,semantic_search,news,ko,0,0,이란과 미국의 군사 충돌,4303;4304;4307;4316;4322;4323;4327;4335,4303:2;4304:2;4307:2;4316:2;4322:2;4323:2;4327:2;4335:2,18077;22048;12213;4317;21273;21276;15919;16404;17242;15922,259.9,0.000,0.000,0.000,1,0.000,0.000,0.000,
|
||||
single,news_002,korean_only,news_ko,semantic_search,news,ko,0,0,호르무즈 해협 봉쇄,4316;4320;4322;4327,4316:3;4320:2;4322:2;4327:2,16532;22049;20240;22055;15917;21890;15922;15918;4346;9022,99.0,0.000,0.000,0.000,0,0.000,0.000,0.000,
|
||||
single,misc_001,korean_only,other_domain,fact_lookup,document,ko,0,0,강체의 평면 운동학,4063;4065,4063:3;4065:2,4063;4065;4066;4064;4071;4068;4058;4069;4067;5064,378.0,1.000,1.000,1.000,1,1.000,1.000,1.000,
|
||||
single,misc_002,korean_only,other_domain,semantic_search,document,ko,0,0,질점의 운동역학,4060;4061;4062,4060:2;4061:2;4062:2,4061;4060;4062;4070;4059;4064;4065;4066;4063;4058,356.3,1.000,1.000,1.000,1,1.000,1.000,0.000,
|
||||
single,news_003,english_only,news_en,semantic_search,news,en,0,0,Trump Iran ultimatum,4258;4260;4262,4258:2;4260:2;4262:2,21441;4202;4776;4679;16941;21897;4775;21155;16823;4199,74.6,0.000,0.000,0.000,1,0.000,0.000,0.000,
|
||||
single,cl_001,mixed,crosslingual_ko_en,semantic_search,document,mixed,0,0,기계 안전 가드 설계 원리,3770;3856,3770:3;3856:2,5244;5239;3758;3791;3770;3817;3763;3787;4540;5253,319.4,0.500,0.200,0.237,0,0.305,0.500,1.000,
|
||||
single,cl_002,mixed,crosslingual_ko_en,semantic_search,document,mixed,0,0,산업 안전 입문서,3755;3775;3776;3777,3755:2;3775:2;3776:2;3777:2,5230;5249;3755;3863;3802;3851;3859;3895;3896;3890,215.7,0.250,0.333,0.195,1,0.195,0.250,0.000,
|
||||
single,cl_003,mixed,crosslingual_ko_en,semantic_search,document,mixed,0,0,전기 안전 위험,3772;3790,3772:2;3790:2,3897;3790;3772;3775;13935;4020;4021;13934;13938;4018,527.9,1.000,0.500,0.693,1,0.693,1.000,0.000,
|
||||
single,news_004,mixed,news_fr,semantic_search,news,mixed,0,0,guerre en Iran,4199;4202;4210;4361;4363;4507;4519;4521,4199:2;4202:2;4210:2;4361:2;4363:2;4507:2;4519:2;4521:2,23242;19434;16606;24991;18723;23149;15924;16941;16404;16538,119.0,0.000,0.000,0.000,1,0.000,0.000,0.000,
|
||||
single,news_005,mixed,news_crosslingual,semantic_search,news,mixed,0,0,이란 미국 전쟁 글로벌 반응,4202;4258;4262;4536;4303;4304;4316,4202:2;4258:2;4262:2;4303:2;4304:2;4316:2;4536:2,23242;15894;16751;19434;22069;16912;18088;17242;16759;4345,305.4,0.000,0.000,0.000,1,0.000,0.000,0.000,
|
||||
single,fail_001,failure_expected,failure_expected,semantic_search,document,mixed,0,1,Rust async runtime tokio scheduler 내부 구조,,,5161;23732;5262;5061;20758;4550;17810;4546;4547;20036,199.0,0.000,0.000,0.000,1,0.000,0.000,0.000,
|
||||
single,fail_002,failure_expected,failure_expected,semantic_search,document,ko,0,1,양자컴퓨터 큐비트 디코히어런스,,,5057;4057;5135;5094;22202;5092;5066;5078;17899;23498,174.0,0.000,0.000,0.000,1,0.000,0.000,0.000,
|
||||
single,fail_003,failure_expected,failure_expected,semantic_search,news,ko,0,1,재즈 보컬리스트 빌리 홀리데이,,,18567;20022;20470;19172;18286;21525;16320;21847;4780;16151,79.8,0.000,0.000,0.000,1,0.000,0.000,0.000,
|
||||
single,kw_006,standards,standards,fact_lookup,document,ko,0,0,산업안전보건기준에 관한 규칙 작업장 통로,3886;3887,3886:3;3887:2,3886;3887;5249;3881;3912;3892;3898;3896;3888;3893,242.0,1.000,1.000,1.000,1,1.000,1.000,1.000,
|
||||
single,kw_007,standards,standards,fact_lookup,document,ko,0,0,산업안전보건기준 폭발 화재 위험물 누출 방지,3896;3766,3766:1;3896:3,3896;3890;3901;11572;11562;13929;11567;3894;3899;3911,115.8,0.500,1.000,0.613,1,0.917,1.000,1.000,
|
||||
single,kw_008,standards,standards,fact_lookup,document,ko,0,0,고압가스 안전관리법 전문,4025;4026,4025:3;4026:2,11644;11579;4026;4025;11693;13750;13299;13941;13749;13766,243.5,1.000,0.333,0.571,0,0.508,1.000,1.000,
|
||||
single,kw_009,standards,standards,fact_lookup,document,ko,0,0,KGS FP111 가스설비 배관설비 기준,13305,13305:3,13305;13311;13306;13312;13302;13304;11688;13309;13313;13310,259.4,1.000,1.000,1.000,1,1.000,1.000,1.000,
|
||||
single,kw_010,standards,standards,fact_lookup,document,ko,0,0,KGS FU551 가스설비 압력조정기 가스계량기,13652,13652:3,13652;11689;13651;13655;13656;13649;13658;13752;13648;13659,168.3,1.000,1.000,1.000,1,1.000,1.000,1.000,
|
||||
single,kw_011,standards,standards,fact_lookup,document,ko,0,0,산업안전보건기준 전기로 인한 위험 방지,3897;3772,3772:1;3897:3,3897;3886;3887;13935;3895;3902;3896;13938;3877;3900,271.8,0.500,1.000,0.613,1,0.917,1.000,1.000,
|
||||
single,en_001,english_only,english_only,semantic_search,document,en,0,0,pressure vessel flange design,5144;5136,5136:2;5144:3,5144;5136;5149;5180;5140;5178;5207;5148;5212;5137,1436.2,1.000,1.000,1.000,1,1.000,1.000,1.000,
|
||||
single,en_002,english_only,english_only,semantic_search,document,en,0,0,ASME Section VIII introduction,5204;5206,5204:3;5206:2,5204;5212;5208;5210;5206;11634;5207;5141;5182;5183,1266.8,1.000,1.000,0.850,1,0.918,1.000,1.000,
|
||||
single,en_003,english_only,english_only,semantic_search,document,en,0,0,impact test requirements ASME,5205;5148,5148:1;5205:3,5205;5204;5178;5214;5190;5148;5145;5185;5192;5212,1442.7,1.000,1.000,0.832,1,0.964,1.000,1.000,
|
||||
single,en_004,english_only,english_only,semantic_search,document,en,0,0,design of vessel supports,5149,5149:3,5149;5136;5186;5178;5145;5148;5192;5185;5212;5147,1387.4,1.000,1.000,1.000,1,1.000,1.000,1.000,
|
||||
single,en_005,english_only,english_only,semantic_search,document,en,0,0,hydrogen piping ASME code,5178,5178:3,5178;5204;5207;5143;5147;5179;5180;5137;5210;5182,1294.3,1.000,1.000,1.000,1,1.000,1.000,1.000,
|
||||
single,en_006,english_only,english_only,semantic_search,document,en,0,0,ASME welding qualification requirements,5209;3771,3771:1;5209:3,5178;5224;5180;5205;5209;5212;5225;5145;4835;4826,1256.9,0.500,0.200,0.237,0,0.355,1.000,1.000,
|
||||
single,en_007,english_only,english_only,semantic_search,document,en,0,0,pressure vessel fabrication and inspection,5208;5145,5145:2;5208:3,5208;5186;5212;5137;5148;5143;5204;5185;5140;5193,1483.3,0.500,1.000,0.613,1,0.787,0.500,1.000,
|
||||
single,en_008,english_only,english_only,semantic_search,document,en,0,0,Industrial Safety and Health Management ergonomics,3763;3755,3755:1;3763:3,3812;3763;3813;3756;3755;3757;3815;3774;3814;3770,759.6,1.000,0.500,0.624,1,0.629,1.000,1.000,
|
||||
single,cl_004,mixed,mixed,semantic_search,document,mixed,0,0,ASME 압력용기 설계 실무,5207;5210;5139;5135,5135:2;5139:2;5207:3;5210:3,5139;5207;5225;5204;5133;5212;5182;5140;5137;5224,567.3,0.500,1.000,0.637,1,0.522,0.500,0.500,
|
||||
single,cl_005,mixed,mixed,semantic_search,document,mixed,0,0,ASME 용접 코드 해설,5224;5209,5209:2;5224:3,5224;5222;5225;5209;5210;5208;5205;5204;5178;11601,512.5,1.000,1.000,0.877,1,0.932,1.000,1.000,
|
||||
single,cl_006,mixed,mixed,semantic_search,document,mixed,0,0,pressure vessel Korean industrial safety regulation,4025;3856;5136,3856:2;4025:2;5136:1,5143;3762;5210;11633;5204;3985;5182;3797;5186;11600,562.4,0.000,0.000,0.000,1,0.000,0.000,0.000,
|
||||
single,cl_007,mixed,mixed,semantic_search,document,mixed,0,0,KGS 코드 LPG safety standard,11647;11689;11645;4025,4025:1;11645:2;11647:3;11689:2,11647;13674;13669;13774;13773;13675;11688;13757;13769;11644,200.8,0.250,1.000,0.390,1,0.647,0.333,1.000,
|
||||
single,cl_008,mixed,mixed,semantic_search,document,mixed,0,0,수소 가스 안전 기준 hydrogen safety,5178;5169,5169:2;5178:3,5241;11671;5177;11653;11568;5173;11538;11579;5178;11648,421.8,0.500,0.111,0.185,0,0.237,0.500,1.000,
|
||||
single,exam_001,exam,exam,fact_lookup,document,ko,0,0,레이놀즈수 정의 공식,11504;11505,11504:3;11505:2,11533;11504;11482;11509;11505;11513;11510;11476;11712;11486,253.7,1.000,0.500,0.624,1,0.627,1.000,1.000,
|
||||
single,exam_002,exam,exam,fact_lookup,document,ko,0,0,탱크 바닥 구멍 유체 유속 토리첼리,11500;11495;11496,11495:2;11496:2;11500:3,11500;11495;11501;11503;5071;5139;13771;11515;11719;13307,156.7,0.667,1.000,0.765,1,0.856,0.667,1.000,
|
||||
single,exam_003,exam,exam,fact_lookup,document,ko,0,0,이상기체 음속 마하수,11514;11515,11514:3;11515:2,11514;11479;5090;11475;11473;11515;11518;5057;11487;11516,381.2,1.000,1.000,0.832,1,0.907,1.000,1.000,
|
||||
single,exam_004,exam,exam,fact_lookup,document,ko,0,0,고압가스 용기 내압시험 영구증가량,11591;11644;11691,11591:3;11644:2;11691:2,13948;13660;5177;13652;13759;13942;13917;13752;4026;11579,161.2,0.000,0.000,0.000,0,0.000,0.000,0.000,
|
||||
single,exam_005,exam,exam,fact_lookup,document,ko,0,0,도시가스 배관 매설 이격거리,11627;11625;11646,11625:2;11627:3;11646:1,11658;11600;11625;11692;11627;13751;11655;13753;11624;13652,170.4,0.667,0.333,0.416,0,0.448,1.000,1.000,
|
||||
single,exam_006,exam,exam,fact_lookup,document,ko,0,0,LPG 저장탱크 안전거리 분말소화기,11617;11669;11620,11617:3;11620:1;11669:2,11595;11605;11655;11658;11690;11653;11669;13670;11639;11649,123.0,0.333,0.143,0.156,0,0.106,0.500,0.000,
|
||||
single,exam_007,exam,exam,fact_lookup,document,ko,0,0,오리피스 차압식 유량계,11712;11711;11503,11503:2;11711:2;11712:3,11712;11711;11503;11500;11713;13930;11717;11715;11716;11719,158.3,1.000,1.000,1.000,1,1.000,1.000,1.000,
|
||||
single,fail_004,failure_expected,failure_expected,fact_lookup,document,ko,0,1,KGS AC999 임의 가스 코드,,,11691;11647;11668;11583;11693;11692;13664;13665;13661;13666,161.9,0.000,0.000,0.000,1,0.000,0.000,0.000,
|
||||
single,fail_005,failure_expected,failure_expected,fact_lookup,document,ko,0,1,초전도 안전 관리법 시행규칙,,,3977;4026;3971;3966;4018;3972;3973;3974;4019;13913,226.3,0.000,0.000,0.000,1,0.000,0.000,0.000,
|
||||
|
@@ -0,0 +1,46 @@
|
||||
{
|
||||
"version": "v0.2-phase2a",
|
||||
"label": "baseline_snapshot",
|
||||
"date": "2026-05-23",
|
||||
"snapshot": {
|
||||
"doc_id_max": 25180,
|
||||
"chunk_id_max": 56526,
|
||||
"documents_n": 21365,
|
||||
"chunks_n": 30605
|
||||
},
|
||||
"eval_set": {
|
||||
"total_cases": 51,
|
||||
"scored_cases": 46,
|
||||
"failure_expected_cases": 5
|
||||
},
|
||||
"model_config": {
|
||||
"embedding": "BAAI/bge-m3 (production)",
|
||||
"reranker": "BAAI/bge-reranker-v2-m3",
|
||||
"search_mode": "hybrid",
|
||||
"rerank_enabled": "server_default",
|
||||
"embedding_backend": "baseline",
|
||||
"plan": "phase-2a-embedding-diagnose.md v4"
|
||||
},
|
||||
"overall": {
|
||||
"n": 46,
|
||||
"graded_ndcg_at_10": 0.659,
|
||||
"graded_recall_at_10_t2": 0.695,
|
||||
"graded_recall_at_10_t3": 0.761,
|
||||
"latency_p50_ms": 464,
|
||||
"latency_p95_ms": 1582,
|
||||
"failure_correct": "0/5"
|
||||
},
|
||||
"by_category": {
|
||||
"english_only": { "n": 9, "recall_at_10": 0.78, "ndcg_at_10": 0.71, "graded_ndcg_at_10": 0.78 },
|
||||
"exam": { "n": 7, "recall_at_10": 0.57, "ndcg_at_10": 0.62, "graded_ndcg_at_10": 0.74 },
|
||||
"korean_only": { "n": 9, "recall_at_10": 0.55, "ndcg_at_10": 0.47, "graded_ndcg_at_10": 0.51 },
|
||||
"mixed": { "n": 10, "recall_at_10": 0.38, "ndcg_at_10": 0.36, "graded_ndcg_at_10": 0.39 },
|
||||
"standards": { "n": 11, "recall_at_10": 0.91, "ndcg_at_10": 0.85, "graded_ndcg_at_10": 0.87 }
|
||||
},
|
||||
"by_language": {
|
||||
"en": { "n": 9, "recall_at_10": 0.78, "graded_ndcg_at_10": 0.78 },
|
||||
"ko": { "n": 27, "recall_at_10": 0.70, "graded_ndcg_at_10": 0.72 },
|
||||
"mixed": { "n": 10, "recall_at_10": 0.38, "graded_ndcg_at_10": 0.39 }
|
||||
},
|
||||
"raw_csv": "reports/v0_2_phase2a_baseline_snapshot_2026-05-23.csv"
|
||||
}
|
||||
@@ -0,0 +1,60 @@
|
||||
{
|
||||
"version": "v0.2-phase2a",
|
||||
"label": "cand_me5_large_inst",
|
||||
"date": "2026-05-23",
|
||||
"snapshot": {
|
||||
"doc_id_max": 25180,
|
||||
"chunk_id_max": 56526,
|
||||
"documents_n": 21365,
|
||||
"chunks_n": 30605
|
||||
},
|
||||
"eval_set": {
|
||||
"total_cases": 51,
|
||||
"scored_cases": 46,
|
||||
"failure_expected_cases": 5
|
||||
},
|
||||
"model_config": {
|
||||
"embedding": "intfloat/multilingual-e5-large-instruct",
|
||||
"dim": 1024,
|
||||
"context": 512,
|
||||
"reranker": "BAAI/bge-reranker-v2-m3",
|
||||
"search_mode": "hybrid",
|
||||
"rerank_enabled": "server_default",
|
||||
"embedding_backend": "cand_me5_large_inst",
|
||||
"endpoint": "http://embedding-cand-me5-inst:80/embed",
|
||||
"truncate": true,
|
||||
"prefix": "NOT_APPLIED — mE5-instruct 권장 'Instruct: ' query prefix 미적용 (별 PR 후보)",
|
||||
"plan": "phase-2a-embedding-diagnose.md v4"
|
||||
},
|
||||
"overall": {
|
||||
"n": 46,
|
||||
"graded_ndcg_at_10": 0.477,
|
||||
"graded_recall_at_10_t2": 0.622,
|
||||
"graded_recall_at_10_t3": 0.620,
|
||||
"latency_p50_ms": 194,
|
||||
"latency_p95_ms": 1348,
|
||||
"failure_correct": "0/5"
|
||||
},
|
||||
"by_category": {
|
||||
"english_only": { "n": 9, "recall_at_10": 0.67, "ndcg_at_10": 0.60, "graded_ndcg_at_10": 0.63 },
|
||||
"exam": { "n": 7, "recall_at_10": 0.76, "ndcg_at_10": 0.59, "graded_ndcg_at_10": 0.62 },
|
||||
"korean_only": { "n": 9, "recall_at_10": 0.66, "ndcg_at_10": 0.48, "graded_ndcg_at_10": 0.47 },
|
||||
"mixed": { "n": 10, "recall_at_10": 0.21, "ndcg_at_10": 0.19, "graded_ndcg_at_10": 0.17 },
|
||||
"standards": { "n": 11, "recall_at_10": 0.68, "ndcg_at_10": 0.55, "graded_ndcg_at_10": 0.54 }
|
||||
},
|
||||
"by_language": {
|
||||
"en": { "n": 9, "recall_at_10": 0.67, "graded_ndcg_at_10": 0.63 },
|
||||
"ko": { "n": 27, "recall_at_10": 0.69, "graded_ndcg_at_10": 0.54 },
|
||||
"mixed": { "n": 10, "recall_at_10": 0.21, "graded_ndcg_at_10": 0.17 }
|
||||
},
|
||||
"raw_csv": "reports/v0_2_phase2a_me5_large_inst_2026-05-23.csv",
|
||||
"delta_vs_baseline": {
|
||||
"graded_ndcg_at_10": -0.182,
|
||||
"mixed": -0.22,
|
||||
"korean_only": -0.04,
|
||||
"standards": -0.33,
|
||||
"english_only": -0.15,
|
||||
"exam": -0.12,
|
||||
"latency_p50_ms": -270
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,59 @@
|
||||
{
|
||||
"version": "v0.2-phase2a",
|
||||
"label": "cand_snowflake_l_v2",
|
||||
"date": "2026-05-23",
|
||||
"snapshot": {
|
||||
"doc_id_max": 25180,
|
||||
"chunk_id_max": 56526,
|
||||
"documents_n": 21365,
|
||||
"chunks_n": 30605
|
||||
},
|
||||
"eval_set": {
|
||||
"total_cases": 51,
|
||||
"scored_cases": 46,
|
||||
"failure_expected_cases": 5
|
||||
},
|
||||
"model_config": {
|
||||
"embedding": "Snowflake/snowflake-arctic-embed-l-v2.0",
|
||||
"dim": 1024,
|
||||
"context": 8192,
|
||||
"reranker": "BAAI/bge-reranker-v2-m3",
|
||||
"search_mode": "hybrid",
|
||||
"rerank_enabled": "server_default",
|
||||
"embedding_backend": "cand_snowflake_l_v2",
|
||||
"endpoint": "http://embedding-cand-snowflake-l-v2:80/embed",
|
||||
"truncate": true,
|
||||
"plan": "phase-2a-embedding-diagnose.md v4"
|
||||
},
|
||||
"overall": {
|
||||
"n": 46,
|
||||
"graded_ndcg_at_10": 0.616,
|
||||
"graded_recall_at_10_t2": 0.726,
|
||||
"graded_recall_at_10_t3": 0.728,
|
||||
"latency_p50_ms": 254,
|
||||
"latency_p95_ms": 1412,
|
||||
"failure_correct": "0/5"
|
||||
},
|
||||
"by_category": {
|
||||
"english_only": { "n": 9, "recall_at_10": 0.78, "ndcg_at_10": 0.68, "graded_ndcg_at_10": 0.74 },
|
||||
"exam": { "n": 7, "recall_at_10": 0.67, "ndcg_at_10": 0.54, "graded_ndcg_at_10": 0.56 },
|
||||
"korean_only": { "n": 9, "recall_at_10": 0.60, "ndcg_at_10": 0.50, "graded_ndcg_at_10": 0.52 },
|
||||
"mixed": { "n": 10, "recall_at_10": 0.40, "ndcg_at_10": 0.32, "graded_ndcg_at_10": 0.35 },
|
||||
"standards": { "n": 11, "recall_at_10": 0.91, "ndcg_at_10": 0.85, "graded_ndcg_at_10": 0.87 }
|
||||
},
|
||||
"by_language": {
|
||||
"en": { "n": 9, "recall_at_10": 0.78, "graded_ndcg_at_10": 0.74 },
|
||||
"ko": { "n": 27, "recall_at_10": 0.74, "graded_ndcg_at_10": 0.67 },
|
||||
"mixed": { "n": 10, "recall_at_10": 0.40, "graded_ndcg_at_10": 0.35 }
|
||||
},
|
||||
"raw_csv": "reports/v0_2_phase2a_snowflake_l_v2_2026-05-23.csv",
|
||||
"delta_vs_baseline": {
|
||||
"graded_ndcg_at_10": -0.043,
|
||||
"mixed": -0.04,
|
||||
"korean_only": +0.01,
|
||||
"standards": 0.00,
|
||||
"english_only": -0.04,
|
||||
"exam": -0.18,
|
||||
"latency_p50_ms": -210
|
||||
}
|
||||
}
|
||||
@@ -199,6 +199,9 @@ async def call_search(
|
||||
fusion: str | None = None,
|
||||
rerank: str | None = None,
|
||||
analyze: str | None = None,
|
||||
embedding_backend: str | None = None,
|
||||
snapshot_doc_id_max: int | None = None,
|
||||
snapshot_chunk_id_max: int | None = None,
|
||||
) -> tuple[list[int], float]:
|
||||
"""검색 API 호출 → (doc_ids, latency_ms)."""
|
||||
url = f"{base_url.rstrip('/')}/api/search/"
|
||||
@@ -210,6 +213,12 @@ async def call_search(
|
||||
params["rerank"] = rerank
|
||||
if analyze is not None:
|
||||
params["analyze"] = analyze
|
||||
if embedding_backend is not None:
|
||||
params["embedding_backend"] = embedding_backend
|
||||
if snapshot_doc_id_max is not None:
|
||||
params["snapshot_doc_id_max"] = snapshot_doc_id_max
|
||||
if snapshot_chunk_id_max is not None:
|
||||
params["snapshot_chunk_id_max"] = snapshot_chunk_id_max
|
||||
|
||||
import time
|
||||
|
||||
@@ -237,6 +246,9 @@ async def evaluate(
|
||||
fusion: str | None = None,
|
||||
rerank: str | None = None,
|
||||
analyze: str | None = None,
|
||||
embedding_backend: str | None = None,
|
||||
snapshot_doc_id_max: int | None = None,
|
||||
snapshot_chunk_id_max: int | None = None,
|
||||
) -> list[QueryResult]:
|
||||
"""전체 쿼리셋 평가."""
|
||||
results: list[QueryResult] = []
|
||||
@@ -245,7 +257,10 @@ async def evaluate(
|
||||
for q in queries:
|
||||
try:
|
||||
returned_ids, latency_ms = await call_search(
|
||||
client, base_url, token, q.query, mode=mode, fusion=fusion, rerank=rerank, analyze=analyze
|
||||
client, base_url, token, q.query, mode=mode, fusion=fusion, rerank=rerank, analyze=analyze,
|
||||
embedding_backend=embedding_backend,
|
||||
snapshot_doc_id_max=snapshot_doc_id_max,
|
||||
snapshot_chunk_id_max=snapshot_chunk_id_max,
|
||||
)
|
||||
results.append(
|
||||
QueryResult(
|
||||
@@ -819,6 +834,9 @@ async def call_search_full(
|
||||
rerank: str | None = None,
|
||||
analyze: str | None = None,
|
||||
debug: bool = False,
|
||||
embedding_backend: str | None = None,
|
||||
snapshot_doc_id_max: int | None = None,
|
||||
snapshot_chunk_id_max: int | None = None,
|
||||
) -> tuple[list[dict], float]:
|
||||
"""call_search와 동일 로직. 단 full result dict 리스트 반환."""
|
||||
url = f"{base_url.rstrip('/')}/api/search/"
|
||||
@@ -832,6 +850,12 @@ async def call_search_full(
|
||||
params["analyze"] = analyze
|
||||
if debug:
|
||||
params["debug"] = "true"
|
||||
if embedding_backend is not None:
|
||||
params["embedding_backend"] = embedding_backend
|
||||
if snapshot_doc_id_max is not None:
|
||||
params["snapshot_doc_id_max"] = snapshot_doc_id_max
|
||||
if snapshot_chunk_id_max is not None:
|
||||
params["snapshot_chunk_id_max"] = snapshot_chunk_id_max
|
||||
|
||||
import time
|
||||
|
||||
@@ -1266,6 +1290,25 @@ def main() -> int:
|
||||
choices=["v0.1", "v0.2", "both"],
|
||||
help="점수 출력 모드 (Phase 1, default both). v0.1=binary only / v0.2=graded only / both=둘 다",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--embedding-backend",
|
||||
type=str,
|
||||
default=None,
|
||||
help="Phase 2A Diagnose dispatcher slug (baseline | cand_me5_large_inst | cand_snowflake_l_v2). 미지정 = production.",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--snapshot-doc-id-max",
|
||||
type=int,
|
||||
default=None,
|
||||
help="Phase 2A snapshot freeze. documents.id <= 값 filter. baseline rebaseline 도 동일 적용.",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--snapshot-chunk-id-max",
|
||||
type=int,
|
||||
default=None,
|
||||
help="Phase 2A snapshot freeze. document_chunks.id <= 값 filter. baseline rebaseline 도 동일 적용.",
|
||||
)
|
||||
|
||||
args = parser.parse_args()
|
||||
|
||||
if not args.token:
|
||||
@@ -1318,21 +1361,21 @@ def main() -> int:
|
||||
if args.base_url:
|
||||
print(f"\n>>> evaluating: {args.base_url}")
|
||||
results = asyncio.run(
|
||||
evaluate(queries, args.base_url, args.token, "single", mode=args.mode, fusion=args.fusion, rerank=args.rerank, analyze=args.analyze)
|
||||
evaluate(queries, args.base_url, args.token, "single", mode=args.mode, fusion=args.fusion, rerank=args.rerank, analyze=args.analyze, embedding_backend=args.embedding_backend, snapshot_doc_id_max=args.snapshot_doc_id_max, snapshot_chunk_id_max=args.snapshot_chunk_id_max)
|
||||
)
|
||||
print_summary("single", results, eval_version=args.eval_version)
|
||||
all_results.extend(results)
|
||||
else:
|
||||
print(f"\n>>> baseline: {args.baseline_url}")
|
||||
baseline_results = asyncio.run(
|
||||
evaluate(queries, args.baseline_url, args.token, "baseline", mode=args.mode, fusion=args.fusion, rerank=args.rerank, analyze=args.analyze)
|
||||
evaluate(queries, args.baseline_url, args.token, "baseline", mode=args.mode, fusion=args.fusion, rerank=args.rerank, analyze=args.analyze, embedding_backend=args.embedding_backend, snapshot_doc_id_max=args.snapshot_doc_id_max, snapshot_chunk_id_max=args.snapshot_chunk_id_max)
|
||||
)
|
||||
baseline_summary = print_summary("baseline", baseline_results, eval_version=args.eval_version)
|
||||
|
||||
print(f"\n>>> candidate: {args.candidate_url}")
|
||||
candidate_results = asyncio.run(
|
||||
evaluate(
|
||||
queries, args.candidate_url, args.token, "candidate", mode=args.mode, fusion=args.fusion, rerank=args.rerank, analyze=args.analyze
|
||||
queries, args.candidate_url, args.token, "candidate", mode=args.mode, fusion=args.fusion, rerank=args.rerank, analyze=args.analyze, embedding_backend=args.embedding_backend, snapshot_doc_id_max=args.snapshot_doc_id_max, snapshot_chunk_id_max=args.snapshot_chunk_id_max
|
||||
)
|
||||
)
|
||||
candidate_summary = print_summary("candidate", candidate_results, eval_version=args.eval_version)
|
||||
|
||||
Reference in New Issue
Block a user