Files
news/embedding/factory.py
T
simon 65ead54b4f docs: 文档清理与重构 — 统一为 3 个核心文档
- 删除 5 个过时/残留文档(project_plan/agent_prompt/optimization_plan/report_db_design/deploy/README)
- 新建 docs/architecture.md(项目架构:11 包职责+数据模型+配置+产物)
- 重写 docs/user-guide.md(CLI 全量+增量/断点续跑+MCP+FAQ)
- 重写 README.md(精简入口+文档索引)
- 更新 continuation.md(追加本次记录)
- 更新 .gitignore(排除 data/* 运行产物)
2026-08-22 17:10:39 +08:00

72 lines
2.3 KiB
Python

"""Embedding provider 工厂:根据配置构造合适后端。
配置优先级: 显式参数 > configs/llm_models.yaml scenes.embedding > .env > 默认。
"""
from __future__ import annotations
import os
from configs.loader import load_scene_config
from .base import AsyncEmbeddingProvider, EmbeddingProvider
from .models import EmbeddingError, EmbeddingProviderType
from .remote import (
DashScopeAsyncEmbeddingProvider,
DashScopeEmbeddingProvider,
)
def _read_env(key: str, default: str | None = None) -> str | None:
val = os.environ.get(key)
if val is None or val.strip() == "":
return default
return val.strip()
def resolve_provider_type(provider: str | None = None) -> EmbeddingProviderType:
"""根据 provider 参数 / YAML 场景 / env 解析出 EmbeddingProviderType。
映射:
dashscope / qwen / remote -> DASHSCOPE
local / local-bge / bge / bge-m3 -> LOCAL_BGE
默认 dashscope。
"""
scene_provider = load_scene_config("embedding").get("provider")
p = (
provider
or scene_provider
or _read_env("EMBEDDING_PROVIDER", "dashscope")
or "dashscope"
).lower()
if p in ("dashscope", "qwen", "remote"):
return EmbeddingProviderType.DASHSCOPE
if p in ("local", "local-bge", "bge", "bge-m3"):
return EmbeddingProviderType.LOCAL_BGE
raise EmbeddingError(f"未知 embedding provider: {provider!r}")
def make_sync_provider(
provider: str | None = None,
**kwargs: object,
) -> EmbeddingProvider:
"""构造同步 provider。"""
pt = resolve_provider_type(provider)
if pt == EmbeddingProviderType.DASHSCOPE:
return DashScopeEmbeddingProvider(**kwargs) # type: ignore[arg-type]
# 本地后端
from .local import LocalBGEEmbeddingProvider
return LocalBGEEmbeddingProvider(**kwargs) # type: ignore[arg-type]
def make_async_provider(
provider: str | None = None,
**kwargs: object,
) -> AsyncEmbeddingProvider:
"""构造异步 provider。"""
pt = resolve_provider_type(provider)
if pt == EmbeddingProviderType.DASHSCOPE:
return DashScopeAsyncEmbeddingProvider(**kwargs) # type: ignore[arg-type]
from .local import LocalBGEAsyncEmbeddingProvider
return LocalBGEAsyncEmbeddingProvider(**kwargs) # type: ignore[arg-type]