2026-04-10 16:52:07 +08:00
|
|
|
|
# Text2SQL Agent 环境配置示例
|
|
|
|
|
|
# 复制为 .env 并填入实际值
|
|
|
|
|
|
|
|
|
|
|
|
# ========== DeepSeek API 配置 ==========
|
|
|
|
|
|
DEEPSEEK_API_KEY=sk-2e3cc9f79a2c40e9a6b7c960d8c090d9
|
|
|
|
|
|
DEEPSEEK_BASE_URL=https://api.deepseek.com
|
|
|
|
|
|
|
|
|
|
|
|
# 模型配置
|
|
|
|
|
|
# 主生成模型
|
2026-04-14 10:28:22 +08:00
|
|
|
|
MODEL_PRIMARY=deepseek-chat
|
2026-04-10 16:52:07 +08:00
|
|
|
|
TEMPERATURE=0 # 确定性模式：每次生成相同结果
|
|
|
|
|
|
MAX_TOKENS=4096
|
|
|
|
|
|
# 生成与校验默认 SQL Server（T-SQL）；可改为 mysql / postgresql 等
|
|
|
|
|
|
TEXT2SQL_DIALECT=sqlserver
|
|
|
|
|
|
|
|
|
|
|
|
# ========== Embedding 配置 ==========
|
2026-04-14 18:02:12 +08:00
|
|
|
|
# 默认 false：使用远程 /v1/embeddings（须配置下方 OPENAI_* 或 MODELSCOPE_* 等）
|
|
|
|
|
|
# 仅当 true 时需设置 EMBEDDING_MODEL_PATH 指向本地 HuggingFace 模型目录
|
2026-04-10 16:52:07 +08:00
|
|
|
|
USE_LOCAL_EMBEDDING=false
|
|
|
|
|
|
# ModelScope 推理 API：https://www.modelscope.cn/docs/inference-api
|
|
|
|
|
|
# MODELSCOPE_API_KEY=ms-b95f3233-4797-4efd-90e6-b3728743571d
|
|
|
|
|
|
# MODELSCOPE_BASE_URL=https://api-inference.modelscope.cn/v1
|
|
|
|
|
|
# MODELSCOPE_EMBEDDING_MODEL=Qwen/Qwen3-Embedding-8B
|
|
|
|
|
|
# # 单次请求最大条数，按服务端限制调整
|
|
|
|
|
|
# MODELSCOPE_EMBEDDING_MAX_BATCH=32
|
|
|
|
|
|
|
|
|
|
|
|
# 兼容 OpenAI 的网关：地址含 /v1（若网关根路径即 OpenAI API 则按实际填写）
|
|
|
|
|
|
OPENAI_BASE_URL=http://113.192.49.54:9080/v1
|
|
|
|
|
|
OPENAI_API_KEY=sk-proj-FEb2ChHZK5Llm0tBkmNIT3BlbkFJbu0b0zQpTW8762yD6HDv
|
|
|
|
|
|
OPENAI_EMBEDDING_MODEL=text-embedding-ada-002
|
|
|
|
|
|
|
2026-04-14 18:02:12 +08:00
|
|
|
|
# ========== 数据文件与向量库路径（均相对仓库根目录）==========
|
|
|
|
|
|
# G3SB 表结构；api_server 用 SCHEMA_PATH / SCHEMA_META_PATH（勿用 SCHEMA_FILE，代码不读）
|
|
|
|
|
|
SCHEMA_PATH=./data/schemas/G3SB_MCDataDictionary_table_structure.json
|
|
|
|
|
|
# G3SB 表注释，与 structure 同名配对；粗筛 embedding 文本会含表注释
|
|
|
|
|
|
SCHEMA_META_PATH=./data/schemas/G3SB_MCDataDictionary_table_meta.json
|
|
|
|
|
|
# Schema 表级向量持久化目录（Chroma 集合名固定为 schema_tables）；改 structure/meta 后需强制重建索引
|
2026-04-10 16:52:07 +08:00
|
|
|
|
VECTOR_DB_PATH=./data/embeddings/chroma
|
2026-04-14 18:02:12 +08:00
|
|
|
|
# Few-shot 源 JSONL（build_fewshot_chroma_index.py 默认读此变量）
|
|
|
|
|
|
FEWSHOT_DATA_PATH=./data/experiences/all_samples.jsonl
|
|
|
|
|
|
# Few-shot 向量持久化目录（Chroma 集合名默认 fewshot_samples；勿与 VECTOR_DB_PATH 混用同一目录）
|
|
|
|
|
|
FEWSHOT_CHROMA_PATH=./data/embeddings/chroma_fewshot
|
2026-04-10 16:52:07 +08:00
|
|
|
|
|
|
|
|
|
|
# ========== 验证配置 ==========
|
|
|
|
|
|
ENABLE_SQL_VALIDATION=true
|
|
|
|
|
|
BLOCK_DANGEROUS_SQL=true
|
|
|
|
|
|
|
|
|
|
|
|
# ========== 重试配置 ==========
|
2026-04-14 10:28:22 +08:00
|
|
|
|
MAX_RETRY=2 # 确定性模式：只尝试一次，不重试
|
2026-04-10 16:52:07 +08:00
|
|
|
|
RETRY_DELAY=1.0
|
|
|
|
|
|
|
|
|
|
|
|
# ========== Few-shot 配置 ==========
|
|
|
|
|
|
# 是否启用经验数据集few-shot增强
|
|
|
|
|
|
FEWSHOT_ENABLED=true
|
|
|
|
|
|
# 每次使用的示例数量
|
|
|
|
|
|
FEWSHOT_TOP_K=3
|
|
|
|
|
|
# 示例最低评分（1-10）
|
|
|
|
|
|
FEWSHOT_MIN_RATING=7
|
2026-04-14 18:02:12 +08:00
|
|
|
|
# 优先使用 Chroma 向量库（运行时检索不依赖 JSONL；灌库: python scripts/build_fewshot_chroma_index.py --force）
|
|
|
|
|
|
FEWSHOT_USE_CHROMA=true
|
|
|
|
|
|
# 留空则用代码默认集合名 fewshot_samples
|
|
|
|
|
|
FEWSHOT_CHROMA_COLLECTION=
|
2026-04-10 16:52:07 +08:00
|
|
|
|
|
|
|
|
|
|
# ========== 日志配置 ==========
|
|
|
|
|
|
LOG_LEVEL=INFO
|
2026-04-14 10:28:22 +08:00
|
|
|
|
LOG_FILE=./logs/text2sql.log
|
|
|
|
|
|
|
2026-04-14 18:02:12 +08:00
|
|
|
|
# 业务库 SQLAlchemy URL（试执行探针 / dbhub）；也可用 DATABASE_URL
|
|
|
|
|
|
database_url=mssql+pymssql://sa:123456@192.168.3.201:1433/G3SB_PROD_HW
|
