- T6 (Issue6): RerankConfig(enabled/model=BAAI/bge-reranker-v2-m3/device)
- rag-layer-design §6.3 Rerank 精排:窗口=RRF top-10、候选≤top_k 跳过、
故障降级 RRF 原序;原 §6.3-6.6 顺延 6.4-6.7
- config-design §5 新增 rerank 段;design §5.5 检索策略加 rerank
- T11 (OV2): EmbeddingConfig.model 默认 bge-small-zh-v1.5 → BAAI/bge-m3(日文语料)
- rag-layer-design 选型表/依赖表/manifest/流程图同步 + 新增 §2.3 日文样本验证
- design.md / implementation-plan 4.3 / config-design embedding 同步
- 新增 test_rag_design_consistency.py 一致性门禁(6 用例防文档漂移)
- TDD: RED(默认模型仍旧 + rerank 字段不存在)→ GREEN → 全量 198 passed / 100.00%(996 stmts/252 br)
85 lines
2.6 KiB
Python
85 lines
2.6 KiB
Python
"""RAG 模型选型一致性门禁:T6/T11 整改(I6/OV2)的机器契约。
|
||
|
||
- T6(I6):v1 引入 rerank 精排 bge-reranker-v2-m3
|
||
- T11(OV2):embedding 默认切换多语言 bge-m3(实际语料为日文)
|
||
|
||
防文档漂移:代码默认值、fixture、当前设计文档四处必须一致;
|
||
当前设计文档不得残留 bge-small-zh-v1.5(历史评审记录/计划快照不改写,故不扫描)。
|
||
"""
|
||
|
||
from __future__ import annotations
|
||
|
||
from pathlib import Path
|
||
|
||
import yaml
|
||
|
||
from genesis.config import Settings
|
||
|
||
ROOT = Path(__file__).resolve().parents[1]
|
||
FIXTURES = Path(__file__).parent / "fixtures"
|
||
|
||
# 当前设计文档(排除历史评审记录 docs/design-review.md、docs/*-review*.md、
|
||
# docs/superpowers/plans/* 等历史快照)
|
||
_CURRENT_DOCS = [
|
||
"docs/rag-layer-design.md",
|
||
"docs/config-design.md",
|
||
"docs/design.md",
|
||
"docs/implementation-plan.md",
|
||
]
|
||
|
||
EMBEDDING_MODEL = "BAAI/bge-m3"
|
||
RERANK_MODEL = "BAAI/bge-reranker-v2-m3"
|
||
# 设计文档用短名(bge-m3);全名含短名子串,故 token 宽松匹配
|
||
_EMBEDDING_TOKEN = "bge-m3"
|
||
_RERANK_TOKEN = "bge-reranker-v2-m3"
|
||
|
||
|
||
def _doc_text(name: str) -> str:
|
||
return (ROOT / name).read_text(encoding="utf-8")
|
||
|
||
|
||
def _fixture_rag() -> dict:
|
||
with (FIXTURES / "rag.yaml").open("r", encoding="utf-8") as f:
|
||
return yaml.safe_load(f) or {}
|
||
|
||
|
||
# ---------- 代码默认值(T11) ----------
|
||
|
||
def test_code_default_embedding_is_bge_m3():
|
||
s = Settings()
|
||
assert s.rag.embedding.model == EMBEDDING_MODEL
|
||
|
||
|
||
# ---------- 代码默认值(T6) ----------
|
||
|
||
def test_code_rerank_config_defaults():
|
||
s = Settings()
|
||
assert s.rag.rerank.enabled is True
|
||
assert s.rag.rerank.model == RERANK_MODEL
|
||
|
||
|
||
# ---------- fixture 与代码一致 ----------
|
||
|
||
def test_fixture_synced_with_code_defaults():
|
||
rag = _fixture_rag()
|
||
assert rag["embedding"]["model"] == EMBEDDING_MODEL
|
||
assert rag["rerank"]["enabled"] is True
|
||
assert rag["rerank"]["model"] == RERANK_MODEL
|
||
|
||
|
||
# ---------- 文档与选型一致(T6/T11) ----------
|
||
|
||
def test_docs_use_bge_m3():
|
||
missing = [name for name in _CURRENT_DOCS if _EMBEDDING_TOKEN not in _doc_text(name)]
|
||
assert not missing, f"以下当前设计文档缺失 {_EMBEDDING_TOKEN}:{missing}"
|
||
|
||
|
||
def test_docs_use_rerank_model():
|
||
missing = [name for name in _CURRENT_DOCS if _RERANK_TOKEN not in _doc_text(name)]
|
||
assert not missing, f"以下当前设计文档缺失 {_RERANK_TOKEN}:{missing}"
|
||
|
||
|
||
def test_docs_no_legacy_bge_small_zh():
|
||
legacy = [name for name in _CURRENT_DOCS if "bge-small-zh-v1.5" in _doc_text(name)]
|
||
assert not legacy, f"当前设计文档残留旧模型 bge-small-zh-v1.5(请同步为 bge-m3):{legacy}"
|