重构：添加模型服务模块，支持嵌入和重排服务的自动降级
新增功能： - 创建 app/model_services 模块，提供统一的模型服务获取接口 - 实现 BaseServiceProvider 基类和 FallbackServiceChain 降级链 - 实现 get_embedding_service()：优先本地 llama.cpp，降级到智谱 API - 实现 get_rerank_service()：优先本地 llama.cpp，降级到智谱 API - 支持单例管理，确保全局只有一个服务实例 修改内容： - 更新 app/config.py，添加智谱 API 相关配置 - 修改 rag_core/vector_store.py：支持接受外部传入的 embeddings - 修改 rag_core/retriever_factory.py：支持接受外部传入的 embeddings - 修改 app/agent/rag_initializer.py：使用 get_embedding_service() - 修改 app/rag/pipeline.py：使用 get_rerank_service() - 修改 app/memory/mem0_client.py：智能判断可用服务配置 mem0 - 修改 rag_indexer/index_builder.py：支持使用新服务，保持向后兼容 - 修改 rag_indexer/config.py：添加智谱配置 环境变量： - ZHIPUAI_API_KEY：智谱 API 密钥（必选） - ZHIPU_EMBEDDING_MODEL：可选，默认 embedding-3 - ZHIPU_RERANK_MODEL：可选，默认 rerank-2 - ZHIPU_API_BASE：可选，默认 https://open.bigmodel.cn/api/paas/v4
This commit is contained in:
31
backend/app/model_services/README.md
Normal file
31
backend/app/model_services/README.md
Normal file
@@ -0,0 +1,31 @@
|
||||
"""
|
||||
模型服务模块（model_services）
|
||||
|
||||
提供统一的嵌入和重排模型服务获取接口，支持自动降级：
|
||||
1. 优先使用本地 llama.cpp 服务
|
||||
2. 本地服务不可用时，自动降级到智谱 API 服务
|
||||
|
||||
使用方法：
|
||||
|
||||
from app.model_services import get_embedding_service, get_rerank_service, BaseReranker
|
||||
|
||||
# 获取嵌入服务（LangChain 兼容的 Embeddings）
|
||||
embeddings = get_embedding_service()
|
||||
|
||||
# 获取重排服务
|
||||
reranker = get_rerank_service()
|
||||
sorted_docs = reranker.compress_documents(documents, query, top_n=5)
|
||||
|
||||
环境变量配置：
|
||||
|
||||
# 智谱 API 配置
|
||||
ZHIPUAI_API_KEY=your_api_key
|
||||
ZHIPU_EMBEDDING_MODEL=embedding-3 # 可选：embedding-2、embedding-3
|
||||
ZHIPU_RERANK_MODEL=rerank-2 # 可选：rerank-1、rerank-2
|
||||
ZHIPU_API_BASE=https://open.bigmodel.cn/api/paas/v4
|
||||
|
||||
# 本地 llama.cpp 服务配置（原有配置保持不变）
|
||||
LLAMACPP_EMBEDDING_URL=http://localhost:port/v1
|
||||
LLAMACPP_RERANKER_URL=http://localhost:port/v1
|
||||
LLAMACPP_API_KEY=your_api_key
|
||||
"""
|
||||
14
backend/app/model_services/__init__.py
Normal file
14
backend/app/model_services/__init__.py
Normal file
@@ -0,0 +1,14 @@
|
||||
"""
|
||||
模型服务模块
|
||||
|
||||
提供统一的嵌入和重排模型服务获取接口，支持自动降级。
|
||||
"""
|
||||
|
||||
from .embedding_services import get_embedding_service
|
||||
from .rerank_services import get_rerank_service, BaseReranker
|
||||
|
||||
__all__ = [
|
||||
"get_embedding_service",
|
||||
"get_rerank_service",
|
||||
"BaseReranker"
|
||||
]
|
||||
139
backend/app/model_services/base.py
Normal file
139
backend/app/model_services/base.py
Normal file
@@ -0,0 +1,139 @@
|
||||
"""
|
||||
模型服务获取器基类和自动降级机制模块
|
||||
|
||||
本模块提供：
|
||||
1. 统一的服务获取器基类，支持服务可用性检查和自动降级
|
||||
2. 单例模式的服务管理器，确保全局只有一个服务实例
|
||||
3. 支持链式降级策略，主服务失败时自动尝试备用服务
|
||||
|
||||
主要功能：
|
||||
- BaseServiceProvider：所有服务获取器的基类
|
||||
- FallbackServiceChain：链式降级处理器
|
||||
- SingletonServiceManager：单例服务管理器
|
||||
"""
|
||||
|
||||
import abc
|
||||
from typing import Generic, TypeVar, List, Optional, Any, Callable
|
||||
from functools import wraps
|
||||
import logging
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
T = TypeVar('T')
|
||||
|
||||
|
||||
class BaseServiceProvider(abc.ABC, Generic[T]):
|
||||
"""
|
||||
服务获取器基类，所有具体服务获取器都需要继承此类
|
||||
"""
|
||||
|
||||
def __init__(self, name: str):
|
||||
self._name = name
|
||||
self._service_instance: Optional[T] = None
|
||||
|
||||
@abc.abstractmethod
|
||||
def is_available(self) -> bool:
|
||||
"""
|
||||
检查服务是否可用
|
||||
|
||||
Returns:
|
||||
bool: 服务是否可用
|
||||
"""
|
||||
pass
|
||||
|
||||
@abc.abstractmethod
|
||||
def get_service(self) -> T:
|
||||
"""
|
||||
获取服务实例
|
||||
|
||||
Returns:
|
||||
T: 服务实例
|
||||
"""
|
||||
pass
|
||||
|
||||
@property
|
||||
def name(self) -> str:
|
||||
"""获取服务名称"""
|
||||
return self._name
|
||||
|
||||
|
||||
class FallbackServiceChain(Generic[T]):
|
||||
"""
|
||||
链式降级处理器，支持多级备用服务
|
||||
"""
|
||||
|
||||
def __init__(self, primary: BaseServiceProvider[T], fallbacks: List[BaseServiceProvider[T]]):
|
||||
self._primary = primary
|
||||
self._fallbacks = fallbacks
|
||||
self._providers = [primary] + fallbacks
|
||||
|
||||
def get_available_service(self) -> T:
|
||||
"""
|
||||
获取第一个可用的服务
|
||||
|
||||
Returns:
|
||||
T: 可用的服务实例
|
||||
|
||||
Raises:
|
||||
RuntimeError: 如果没有可用的服务
|
||||
"""
|
||||
for provider in self._providers:
|
||||
try:
|
||||
if provider.is_available():
|
||||
logger.info(f"使用服务: {provider.name}")
|
||||
return provider.get_service()
|
||||
else:
|
||||
logger.warning(f"服务不可用: {provider.name}，尝试下一个...")
|
||||
except Exception as e:
|
||||
logger.warning(f"服务 {provider.name} 检查失败: {e}，尝试下一个...")
|
||||
|
||||
raise RuntimeError(f"没有可用的服务，尝试了: {[p.name for p in self._providers]}")
|
||||
|
||||
def get_all_providers(self) -> List[BaseServiceProvider[T]]:
|
||||
"""
|
||||
获取所有服务提供者（主服务 + 备用服务）
|
||||
|
||||
Returns:
|
||||
List[BaseServiceProvider[T]]: 服务提供者列表
|
||||
"""
|
||||
return self._providers.copy()
|
||||
|
||||
|
||||
class SingletonServiceManager:
|
||||
"""
|
||||
单例服务管理器，确保全局只有一个服务实例
|
||||
"""
|
||||
|
||||
_instances: dict = {}
|
||||
|
||||
@classmethod
|
||||
def get_or_create(cls, key: str, creator: Callable[[], Any]) -> Any:
|
||||
"""
|
||||
获取或创建单例实例
|
||||
|
||||
Args:
|
||||
key: 单例键
|
||||
creator: 创建函数
|
||||
|
||||
Returns:
|
||||
Any: 单例实例
|
||||
"""
|
||||
if key not in cls._instances:
|
||||
cls._instances[key] = creator()
|
||||
logger.debug(f"创建单例实例: {key}")
|
||||
return cls._instances[key]
|
||||
|
||||
@classmethod
|
||||
def clear(cls, key: Optional[str] = None):
|
||||
"""
|
||||
清除单例实例
|
||||
|
||||
Args:
|
||||
key: 单例键，如果为 None 则清除所有
|
||||
"""
|
||||
if key is None:
|
||||
cls._instances.clear()
|
||||
logger.debug("清除所有单例实例")
|
||||
elif key in cls._instances:
|
||||
del cls._instances[key]
|
||||
logger.debug(f"清除单例实例: {key}")
|
||||
213
backend/app/model_services/embedding_services.py
Normal file
213
backend/app/model_services/embedding_services.py
Normal file
@@ -0,0 +1,213 @@
|
||||
"""
|
||||
嵌入模型服务模块
|
||||
|
||||
本模块提供统一的嵌入模型服务获取接口，支持自动降级：
|
||||
1. 优先使用本地 llama.cpp 嵌入服务
|
||||
2. 本地服务不可用时，自动降级到智谱 API 嵌入服务
|
||||
|
||||
主要功能：
|
||||
- LocalLlamaCppEmbeddingProvider：本地 llama.cpp 嵌入服务提供者
|
||||
- ZhipuEmbeddingProvider：智谱 API 嵌入服务提供者
|
||||
- get_embedding_service()：获取嵌入服务的统一接口
|
||||
"""
|
||||
|
||||
import logging
|
||||
from typing import List
|
||||
import httpx
|
||||
from langchain_core.embeddings import Embeddings
|
||||
|
||||
from .base import (
|
||||
BaseServiceProvider,
|
||||
FallbackServiceChain,
|
||||
SingletonServiceManager
|
||||
)
|
||||
from ..config import (
|
||||
LLAMACPP_EMBEDDING_URL,
|
||||
LLAMACPP_API_KEY,
|
||||
ZHIPUAI_API_KEY,
|
||||
ZHIPU_EMBEDDING_MODEL,
|
||||
ZHIPU_API_BASE
|
||||
)
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
|
||||
class LocalLlamaCppEmbeddingProvider(BaseServiceProvider[Embeddings]):
|
||||
"""
|
||||
本地 llama.cpp 嵌入服务提供者
|
||||
"""
|
||||
|
||||
def __init__(self, model: str = "Qwen3-Embedding-0.6B-Q8_0"):
|
||||
super().__init__("local_llamacpp_embedding")
|
||||
self._model = model
|
||||
|
||||
def is_available(self) -> bool:
|
||||
"""
|
||||
检查本地 llama.cpp 嵌入服务是否可用
|
||||
|
||||
Returns:
|
||||
bool: 服务是否可用
|
||||
"""
|
||||
if not LLAMACPP_EMBEDDING_URL:
|
||||
logger.warning("LLAMACPP_EMBEDDING_URL 未配置")
|
||||
return False
|
||||
|
||||
try:
|
||||
# 尝试嵌入一个测试字符串
|
||||
embedder = LocalLlamaCppEmbedder(model=self._model)
|
||||
test_embedding = embedder.embed_query("test")
|
||||
logger.info(f"本地 llama.cpp 嵌入服务可用，维度: {len(test_embedding)}")
|
||||
return True
|
||||
except Exception as e:
|
||||
logger.warning(f"本地 llama.cpp 嵌入服务不可用: {e}")
|
||||
return False
|
||||
|
||||
def get_service(self) -> Embeddings:
|
||||
"""
|
||||
获取本地 llama.cpp 嵌入服务
|
||||
|
||||
Returns:
|
||||
Embeddings: LangChain 兼容的嵌入实例
|
||||
"""
|
||||
if self._service_instance is None:
|
||||
embedder = LocalLlamaCppEmbedder(model=self._model)
|
||||
self._service_instance = embedder.as_langchain_embeddings()
|
||||
return self._service_instance
|
||||
|
||||
|
||||
class ZhipuEmbeddingProvider(BaseServiceProvider[Embeddings]):
|
||||
"""
|
||||
智谱 API 嵌入服务提供者
|
||||
"""
|
||||
|
||||
def __init__(self, model: str | None = None):
|
||||
super().__init__("zhipu_embedding")
|
||||
self._model = model or ZHIPU_EMBEDDING_MODEL
|
||||
|
||||
def is_available(self) -> bool:
|
||||
"""
|
||||
检查智谱 API 嵌入服务是否可用
|
||||
|
||||
Returns:
|
||||
bool: 服务是否可用
|
||||
"""
|
||||
if not ZHIPUAI_API_KEY:
|
||||
logger.warning("ZHIPUAI_API_KEY 未配置")
|
||||
return False
|
||||
|
||||
try:
|
||||
# 测试智谱 API 是否可用
|
||||
from zhipuai import ZhipuAI
|
||||
client = ZhipuAI(api_key=ZHIPUAI_API_KEY)
|
||||
response = client.embeddings.create(
|
||||
model=self._model,
|
||||
input=["test"]
|
||||
)
|
||||
logger.info(f"智谱嵌入服务可用，维度: {len(response.data[0].embedding)}")
|
||||
return True
|
||||
except ImportError:
|
||||
logger.warning("zhipuai 库未安装")
|
||||
return False
|
||||
except Exception as e:
|
||||
logger.warning(f"智谱嵌入服务不可用: {e}")
|
||||
return False
|
||||
|
||||
def get_service(self) -> Embeddings:
|
||||
"""
|
||||
获取智谱 API 嵌入服务
|
||||
|
||||
Returns:
|
||||
Embeddings: LangChain 兼容的嵌入实例
|
||||
"""
|
||||
if self._service_instance is None:
|
||||
from langchain_zhipu import ZhipuAIEmbeddings
|
||||
self._service_instance = ZhipuAIEmbeddings(
|
||||
model=self._model,
|
||||
api_key=ZHIPUAI_API_KEY
|
||||
)
|
||||
return self._service_instance
|
||||
|
||||
|
||||
class LocalLlamaCppEmbedder:
|
||||
"""
|
||||
通过 OpenAI 兼容 API 封装 llama.cpp 嵌入服务
|
||||
"""
|
||||
|
||||
def __init__(self, model: str = "Qwen3-Embedding-0.6B-Q8_0"):
|
||||
self.base_url = LLAMACPP_EMBEDDING_URL
|
||||
self.api_key = LLAMACPP_API_KEY
|
||||
self.model = model
|
||||
|
||||
def as_langchain_embeddings(self) -> Embeddings:
|
||||
"""创建 LangChain 兼容的嵌入实例"""
|
||||
return _LlamaCppLangchainAdapter(self)
|
||||
|
||||
def embed_documents(self, texts: List[str]) -> List[List[float]]:
|
||||
"""嵌入一批文档"""
|
||||
return self._call_embedding_api(texts)
|
||||
|
||||
def embed_query(self, text: str) -> List[float]:
|
||||
"""嵌入单个查询"""
|
||||
return self._call_embedding_api([text])[0]
|
||||
|
||||
def _call_embedding_api(self, texts: List[str]) -> List[List[float]]:
|
||||
"""直接调用 llama.cpp 嵌入 API"""
|
||||
headers = {"Content-Type": "application/json"}
|
||||
if self.api_key:
|
||||
headers["Authorization"] = f"Bearer {self.api_key}"
|
||||
|
||||
base = self.base_url.rstrip("/")
|
||||
if not base.endswith("/v1"):
|
||||
base = base + "/v1"
|
||||
|
||||
payload = {
|
||||
"input": texts,
|
||||
"model": self.model,
|
||||
}
|
||||
|
||||
with httpx.Client(timeout=120) as client:
|
||||
response = client.post(
|
||||
f"{base}/embeddings",
|
||||
headers=headers,
|
||||
json=payload,
|
||||
)
|
||||
response.raise_for_status()
|
||||
data = response.json()
|
||||
|
||||
if isinstance(data, list):
|
||||
return [item["embedding"] for item in data]
|
||||
elif isinstance(data, dict) and "data" in data:
|
||||
return [item["embedding"] for item in sorted(data["data"], key=lambda x: x["index"])]
|
||||
else:
|
||||
raise ValueError(f"未知的嵌入 API 响应格式: {data}")
|
||||
|
||||
|
||||
class _LlamaCppLangchainAdapter(Embeddings):
|
||||
"""
|
||||
将 LlamaCppEmbedder 适配为 LangChain Embeddings 接口
|
||||
"""
|
||||
|
||||
def __init__(self, embedder: "LocalLlamaCppEmbedder"):
|
||||
self._embedder = embedder
|
||||
|
||||
def embed_documents(self, texts: List[str]) -> List[List[float]]:
|
||||
return self._embedder.embed_documents(texts)
|
||||
|
||||
def embed_query(self, text: str) -> List[float]:
|
||||
return self._embedder.embed_query(text)
|
||||
|
||||
|
||||
def get_embedding_service() -> Embeddings:
|
||||
"""
|
||||
获取嵌入服务（带自动降级）
|
||||
|
||||
Returns:
|
||||
Embeddings: LangChain 兼容的嵌入实例
|
||||
"""
|
||||
def _create_chain():
|
||||
primary = LocalLlamaCppEmbeddingProvider()
|
||||
fallback = ZhipuEmbeddingProvider()
|
||||
return FallbackServiceChain(primary, [fallback])
|
||||
|
||||
chain = SingletonServiceManager.get_or_create("embedding_service_chain", _create_chain)
|
||||
return chain.get_available_service()
|
||||
233
backend/app/model_services/rerank_services.py
Normal file
233
backend/app/model_services/rerank_services.py
Normal file
@@ -0,0 +1,233 @@
|
||||
"""
|
||||
重排模型服务模块
|
||||
|
||||
本模块提供统一的重排模型服务获取接口，支持自动降级：
|
||||
1. 优先使用本地 llama.cpp 重排服务
|
||||
2. 本地服务不可用时，自动降级到智谱 API 重排服务
|
||||
|
||||
主要功能：
|
||||
- LocalLlamaCppRerankProvider：本地 llama.cpp 重排服务提供者
|
||||
- ZhipuRerankProvider：智谱 API 重排服务提供者
|
||||
- get_rerank_service()：获取重排服务的统一接口
|
||||
"""
|
||||
|
||||
import logging
|
||||
from typing import List
|
||||
import requests
|
||||
from langchain_core.documents import Document
|
||||
|
||||
from .base import (
|
||||
BaseServiceProvider,
|
||||
FallbackServiceChain,
|
||||
SingletonServiceManager
|
||||
)
|
||||
from ..config import (
|
||||
LLAMACPP_RERANKER_URL,
|
||||
LLAMACPP_API_KEY,
|
||||
ZHIPUAI_API_KEY,
|
||||
ZHIPU_RERANK_MODEL,
|
||||
ZHIPU_API_BASE
|
||||
)
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
|
||||
class BaseReranker:
|
||||
"""
|
||||
重排器基类，定义统一的接口
|
||||
"""
|
||||
|
||||
def compress_documents(self, documents: List[Document], query: str, top_n: int = 5) -> List[Document]:
|
||||
"""
|
||||
对文档进行重排序
|
||||
|
||||
Args:
|
||||
documents: 待排序的文档列表
|
||||
query: 查询字符串
|
||||
top_n: 返回前 N 个结果
|
||||
|
||||
Returns:
|
||||
排序后的文档列表
|
||||
"""
|
||||
raise NotImplementedError
|
||||
|
||||
|
||||
class LocalLlamaCppReranker(BaseReranker):
|
||||
"""
|
||||
使用远程 llama.cpp 服务对检索结果重排序
|
||||
"""
|
||||
|
||||
def __init__(self, base_url: str, api_key: str, model: str = "bge-reranker-v2-m3", timeout: int = 60):
|
||||
self.base_url = base_url
|
||||
self.api_key = api_key
|
||||
self.model = model
|
||||
self.timeout = timeout
|
||||
self.endpoint = f"{self.base_url}/rerank"
|
||||
|
||||
def compress_documents(self, documents: List[Document], query: str, top_n: int = 5) -> List[Document]:
|
||||
"""
|
||||
对文档进行重排序
|
||||
"""
|
||||
if not documents:
|
||||
return []
|
||||
|
||||
# 准备请求体
|
||||
payload = {
|
||||
"model": self.model,
|
||||
"query": query,
|
||||
"documents": [doc.page_content for doc in documents],
|
||||
"top_n": top_n
|
||||
}
|
||||
headers = {
|
||||
"Content-Type": "application/json",
|
||||
"Authorization": f"Bearer {self.api_key}"
|
||||
}
|
||||
|
||||
try:
|
||||
response = requests.post(self.endpoint, json=payload, headers=headers, timeout=self.timeout)
|
||||
response.raise_for_status()
|
||||
results = response.json()
|
||||
|
||||
# 解析返回结果
|
||||
sorted_indices = [item["index"] for item in results["results"]]
|
||||
sorted_docs = [documents[idx] for idx in sorted_indices]
|
||||
return sorted_docs
|
||||
except Exception as e:
|
||||
logger.warning(f"远程重排序过程出错，返回原始前 {top_n} 个结果: {e}")
|
||||
return documents[:top_n]
|
||||
|
||||
|
||||
class ZhipuReranker(BaseReranker):
|
||||
"""
|
||||
使用智谱 API 对检索结果重排序
|
||||
"""
|
||||
|
||||
def __init__(self, model: str | None = None):
|
||||
self.model = model or ZHIPU_RERANK_MODEL
|
||||
self.api_key = ZHIPUAI_API_KEY
|
||||
|
||||
def compress_documents(self, documents: List[Document], query: str, top_n: int = 5) -> List[Document]:
|
||||
"""
|
||||
对文档进行重排序
|
||||
"""
|
||||
if not documents:
|
||||
return []
|
||||
|
||||
try:
|
||||
from zhipuai import ZhipuAI
|
||||
client = ZhipuAI(api_key=self.api_key)
|
||||
|
||||
response = client.rerank.create(
|
||||
model=self.model,
|
||||
query=query,
|
||||
documents=[doc.page_content for doc in documents],
|
||||
top_n=top_n
|
||||
)
|
||||
|
||||
sorted_indices = [item.index for item in response.results]
|
||||
sorted_docs = [documents[idx] for idx in sorted_indices]
|
||||
return sorted_docs
|
||||
except Exception as e:
|
||||
logger.warning(f"智谱重排序过程出错，返回原始前 {top_n} 个结果: {e}")
|
||||
return documents[:top_n]
|
||||
|
||||
|
||||
class LocalLlamaCppRerankProvider(BaseServiceProvider[BaseReranker]):
|
||||
"""
|
||||
本地 llama.cpp 重排服务提供者
|
||||
"""
|
||||
|
||||
def __init__(self, model: str = "bge-reranker-v2-m3"):
|
||||
super().__init__("local_llamacpp_rerank")
|
||||
self._model = model
|
||||
|
||||
def is_available(self) -> bool:
|
||||
"""
|
||||
检查本地 llama.cpp 重排服务是否可用
|
||||
"""
|
||||
if not LLAMACPP_RERANKER_URL:
|
||||
logger.warning("LLAMACPP_RERANKER_URL 未配置")
|
||||
return False
|
||||
|
||||
try:
|
||||
# 测试重排服务
|
||||
test_docs = [Document(page_content="test document 1"), Document(page_content="test document 2")]
|
||||
reranker = LocalLlamaCppReranker(
|
||||
base_url=LLAMACPP_RERANKER_URL,
|
||||
api_key=LLAMACPP_API_KEY,
|
||||
model=self._model
|
||||
)
|
||||
result = reranker.compress_documents(test_docs, "test query", top_n=1)
|
||||
logger.info(f"本地 llama.cpp 重排服务可用")
|
||||
return True
|
||||
except Exception as e:
|
||||
logger.warning(f"本地 llama.cpp 重排服务不可用: {e}")
|
||||
return False
|
||||
|
||||
def get_service(self) -> BaseReranker:
|
||||
"""
|
||||
获取本地 llama.cpp 重排服务
|
||||
"""
|
||||
if self._service_instance is None:
|
||||
self._service_instance = LocalLlamaCppReranker(
|
||||
base_url=LLAMACPP_RERANKER_URL,
|
||||
api_key=LLAMACPP_API_KEY,
|
||||
model=self._model
|
||||
)
|
||||
return self._service_instance
|
||||
|
||||
|
||||
class ZhipuRerankProvider(BaseServiceProvider[BaseReranker]):
|
||||
"""
|
||||
智谱 API 重排服务提供者
|
||||
"""
|
||||
|
||||
def __init__(self, model: str | None = None):
|
||||
super().__init__("zhipu_rerank")
|
||||
self._model = model or ZHIPU_RERANK_MODEL
|
||||
|
||||
def is_available(self) -> bool:
|
||||
"""
|
||||
检查智谱 API 重排服务是否可用
|
||||
"""
|
||||
if not ZHIPUAI_API_KEY:
|
||||
logger.warning("ZHIPUAI_API_KEY 未配置")
|
||||
return False
|
||||
|
||||
try:
|
||||
# 测试重排服务
|
||||
test_docs = [Document(page_content="test document 1"), Document(page_content="test document 2")]
|
||||
reranker = ZhipuReranker(model=self._model)
|
||||
result = reranker.compress_documents(test_docs, "test query", top_n=1)
|
||||
logger.info(f"智谱重排服务可用")
|
||||
return True
|
||||
except ImportError:
|
||||
logger.warning("zhipuai 库未安装")
|
||||
return False
|
||||
except Exception as e:
|
||||
logger.warning(f"智谱重排服务不可用: {e}")
|
||||
return False
|
||||
|
||||
def get_service(self) -> BaseReranker:
|
||||
"""
|
||||
获取智谱 API 重排服务
|
||||
"""
|
||||
if self._service_instance is None:
|
||||
self._service_instance = ZhipuReranker(model=self._model)
|
||||
return self._service_instance
|
||||
|
||||
|
||||
def get_rerank_service() -> BaseReranker:
|
||||
"""
|
||||
获取重排服务（带自动降级）
|
||||
|
||||
Returns:
|
||||
BaseReranker: 重排服务实例
|
||||
"""
|
||||
def _create_chain():
|
||||
primary = LocalLlamaCppRerankProvider()
|
||||
fallback = ZhipuRerankProvider()
|
||||
return FallbackServiceChain(primary, [fallback])
|
||||
|
||||
chain = SingletonServiceManager.get_or_create("rerank_service_chain", _create_chain)
|
||||
return chain.get_available_service()
|
||||
Reference in New Issue
Block a user