配置说明
项目采用 config.toml + src/core/config.py 双层配置。
配置文件位置
| 文件 | 职责 |
|---|---|
config.toml | 参数数据源（TOML 格式，人工编辑） |
src/core/config.py | Python 读取接口与路径常量，供各模块统一调用 |
主要配置段
[paths] — 目录路径
toml
[paths]
raw_dir = "data/raw"
processed_dir = "data/processed"
evaluation_dir = "data/evaluation"
log_base_dir = "outputs/log" # 实验跑次目录
reports_base_dir = "outputs/reports" # 定稿发布目录
figures_dir = "outputs/figures"[ocr] — OCR 配置
toml
[ocr]
device = "cuda" # 推理设备："cuda" / "cpu"
mfr_batch_size = 8 # 公式识别批量大小
render_dpi = 200 # PDF 渲染 DPI[terms_gen] — 术语结构化生成
toml
[terms_gen]
engine = "api" # "api" / "local"
api_base = "https://api.deepseek.com/v1"
model = "deepseek-chat"
api_key_env = "API-KEY-TERMS"
local_model_dir = ""
temperature = 0.3[rag_gen] — RAG 回答生成
toml
[rag_gen]
engine = "api" # "api" / "local"
api_base = "https://api.deepseek.com/v1"
api_model = "deepseek-chat"
api_key_env = "API-KEY-RAG"
local_model_dir = "../Qwen3.5-4B"
temperature = 0.7[retrieval] — 检索参数
toml
[retrieval]
recall_factor = 10
rrf_k = 60
bm25_default_weight = 0.7
vector_default_weight = 0.3
default_vector_model = "BAAI/bge-base-zh-v1.5"[reports_generation] — 报告生成
toml
[reports_generation]
report_footer_note = "..."
defense_output_subdir = "defense"Python 配置接口
python
from core import config
# 路径常量
config.RAW_DIR
config.PROCESSED_DIR
config.LOG_BASE_DIR
config.REPORTS_BASE_DIR
# 配置读取函数
config.getPathsConfig()
config.get_ocr_config()
config.getRetrievalConfig()
config.getGenerationConfig()
config.getOutputController()详细 API 说明见 config.py 文档。