199 lines
9.2 KiB
Python
199 lines
9.2 KiB
Python
|
from loguru import logger
|
|||
|
|
from typing import List, Dict, Optional, Union, AsyncGenerator, Generator, Any
|
|||
|
|
|
|||
|
|
# 核心：导入 LangChain 的基础语言模型抽象类
|
|||
|
|
from langchain_core.language_models import BaseLanguageModel
|
|||
|
|
from langchain_core.language_models.chat_models import BaseChatModel
|
|||
|
|
from langchain_core.messages import BaseMessage
|
|||
|
|
from langchain_core.outputs import ChatResult
|
|||
|
|
from langchain_core.callbacks import CallbackManagerForLLMRun
|
|||
|
|
from dataclasses import dataclass, field
|
|||
|
|
from typing import Optional, Dict, Any, List
|
|||
|
|
from datetime import datetime
|
|||
|
|
|
|||
|
|
@dataclass
|
|||
|
|
class LLMConfig_DataClass:
|
|||
|
|
"""
|
|||
|
|
统一的LLM配置基类，覆盖在线/本地/嵌入式模型所有配置，映射数据库完整字段
|
|||
|
|
通过 provider + is_embedding 区分模型类型：
|
|||
|
|
- 在线模型：provider in ['openai', 'zhipu', 'baidu'] + is_embedding=False
|
|||
|
|
- 本地模型：provider in ['llama', 'qwen', 'yi'] + is_embedding=False
|
|||
|
|
- 嵌入式模型：provider in ['bge', 'text2vec'] + is_embedding=True
|
|||
|
|
"""
|
|||
|
|
# ====================== 数据库核心公共字段（必选/可选） ======================
|
|||
|
|
# 基础标识字段
|
|||
|
|
name: str # 模型自定义名称（如 "gpt-5"）
|
|||
|
|
model_name: str # 模型官方标识名（如 "gpt-5"、"BAAI/bge-small-zh-v1.5"）
|
|||
|
|
provider: str # 提供商（openai/llama/bge/zhipu 等）
|
|||
|
|
id: Optional[int] = None # 数据库主键ID
|
|||
|
|
description: Optional[str] = None # 模型描述
|
|||
|
|
is_active: bool = True # 是否启用
|
|||
|
|
is_default: bool = False # 是否默认模型
|
|||
|
|
is_embedding: bool = False # 是否为嵌入式模型（核心区分标识）
|
|||
|
|
|
|||
|
|
# ====================== 通用生成参数（所有推理模型共用） ======================
|
|||
|
|
temperature: float = 0.7 # 生成温度（默认值对齐数据库示例）
|
|||
|
|
max_tokens: int = 3000 # 最大生成长度（默认值对齐数据库示例）
|
|||
|
|
top_p: float = 0.6 # 采样Top-P
|
|||
|
|
frequency_penalty: float = 0.0 # 频率惩罚
|
|||
|
|
presence_penalty: float = 0.0 # 存在惩罚
|
|||
|
|
|
|||
|
|
# ====================== 在线模型专属参数（非必填，仅在线模型生效） ======================
|
|||
|
|
api_key: Optional[str] = None # API密钥（在线模型必填）
|
|||
|
|
base_url: Optional[str] = None # API代理地址（如 https://api.openai-proxy.org/v1）
|
|||
|
|
# timeout: int = 30 # 请求超时时间（秒）
|
|||
|
|
max_retries: int = 3 # 最大重试次数
|
|||
|
|
api_version: Optional[str] = None # API版本（如 OpenAI 的 2024-02-15-preview）
|
|||
|
|
|
|||
|
|
# ====================== 本地模型专属参数（非必填，仅本地模型生效） ======================
|
|||
|
|
model_path: Optional[str] = None # 本地模型文件路径（本地模型必填）
|
|||
|
|
device: str = "cpu" # 运行设备（cpu/cuda/mps）
|
|||
|
|
n_ctx: int = 2048 # 上下文窗口大小
|
|||
|
|
n_threads: int = 8 # 推理线程数
|
|||
|
|
quantization: str = "q4_0" # 量化级别（q4_0/q8_0/f16）
|
|||
|
|
load_in_8bit: bool = False # 是否8bit加载
|
|||
|
|
load_in_4bit: bool = False # 是否4bit加载
|
|||
|
|
prompt_template: Optional[str] = None # 自定义Prompt模板
|
|||
|
|
|
|||
|
|
# ====================== 嵌入式模型专属参数（非必填，仅嵌入式模型生效） ======================
|
|||
|
|
normalize_embeddings: bool = True # 是否归一化向量
|
|||
|
|
batch_size: int = 32 # 批量编码大小
|
|||
|
|
encode_kwargs: Dict[str, Any] = field(default_factory=dict) # 编码扩展参数
|
|||
|
|
dimension: Optional[int] = None # 向量维度（如 768）
|
|||
|
|
|
|||
|
|
# ====================== 元数据字段（数据库自动维护） ======================
|
|||
|
|
extra_config: Dict[str, Any] = field(default_factory=dict) # 额外扩展配置
|
|||
|
|
usage_count: int = 0 # 使用次数
|
|||
|
|
last_used_at: Optional[datetime] = None # 最后使用时间
|
|||
|
|
created_at: Optional[datetime] = None # 创建时间
|
|||
|
|
updated_at: Optional[datetime] = None # 更新时间
|
|||
|
|
created_by: Optional[int] = None # 创建人ID
|
|||
|
|
updated_by: Optional[int] = None # 更新人ID
|
|||
|
|
|
|||
|
|
api_key_masked: Optional[str] = "" # 掩码后的API密钥（数据库存储）
|
|||
|
|
|
|||
|
|
# ====================== 核心工具方法 ======================
|
|||
|
|
def __post_init__(self):
|
|||
|
|
"""后置初始化：自动校验和修正配置"""
|
|||
|
|
# 1. 嵌入式模型强制清空推理参数（避免误用）
|
|||
|
|
if self.is_embedding:
|
|||
|
|
self.max_tokens = 0
|
|||
|
|
self.temperature = 0.0
|
|||
|
|
self.top_p = 0.0
|
|||
|
|
|
|||
|
|
# 2. 校验必填参数（按模型类型）
|
|||
|
|
self._validate_required_fields()
|
|||
|
|
|
|||
|
|
def _validate_required_fields(self):
|
|||
|
|
"""按模型类型校验必填参数"""
|
|||
|
|
# 在线模型校验
|
|||
|
|
if not self.is_embedding and self.provider in ['openai', 'zhipu', 'baidu', 'anthropic']:
|
|||
|
|
if not self.api_key:
|
|||
|
|
raise ValueError(f"[{self.name}] 在线模型({self.provider})必须配置 api_key")
|
|||
|
|
|
|||
|
|
# 本地模型校验
|
|||
|
|
if not self.is_embedding and self.provider in ['llama', 'qwen', 'yi', 'glm', 'mistral']:
|
|||
|
|
if not self.model_path:
|
|||
|
|
raise ValueError(f"[{self.name}] 本地模型({self.provider})必须配置 model_path")
|
|||
|
|
|
|||
|
|
def to_dict(self) -> Dict[str, Any]:
|
|||
|
|
"""转换为字典（用于存入/更新数据库）"""
|
|||
|
|
return {
|
|||
|
|
key: value for key, value in self.__dict__.items()
|
|||
|
|
if not key.startswith('_') # 排除私有属性
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
@classmethod
|
|||
|
|
def from_db_dict(cls, db_dict: Dict[str, Any]) -> "LLMConfig_DataClass":
|
|||
|
|
"""从数据库字典初始化配置（核心方法）"""
|
|||
|
|
# 1. 时间字段转换：字符串 → datetime
|
|||
|
|
time_fields = ['last_used_at', 'created_at', 'updated_at']
|
|||
|
|
for field_name in time_fields:
|
|||
|
|
val = db_dict.get(field_name)
|
|||
|
|
if val and isinstance(val, str):
|
|||
|
|
try:
|
|||
|
|
db_dict[field_name] = datetime.fromisoformat(val.replace('Z', '+00:00'))
|
|||
|
|
except (ValueError, TypeError):
|
|||
|
|
db_dict[field_name] = None
|
|||
|
|
|
|||
|
|
# 2. 过滤数据库中无关字段（如 api_key_masked）
|
|||
|
|
valid_fields = cls.__dataclass_fields__.keys()
|
|||
|
|
filtered_dict = {k: v for k, v in db_dict.items() if k in valid_fields}
|
|||
|
|
|
|||
|
|
# 3. 初始化并返回配置实例
|
|||
|
|
return cls(**filtered_dict)
|
|||
|
|
|
|||
|
|
def get_model_type(self) -> str:
|
|||
|
|
"""快速判断模型类型（返回：online/local/embedding）"""
|
|||
|
|
if self.is_embedding:
|
|||
|
|
return "embedding"
|
|||
|
|
if self.provider in ['openai', 'zhipu', 'baidu', 'anthropic']:
|
|||
|
|
return "online"
|
|||
|
|
if self.provider in ['llama', 'qwen', 'yi', 'glm', 'mistral']:
|
|||
|
|
return "local"
|
|||
|
|
return "unknown"
|
|||
|
|
|
|||
|
|
|
|||
|
|
class BaseLLM(BaseChatModel):
|
|||
|
|
"""
|
|||
|
|
继承 LangChain 的 BaseChatModel（BaseLanguageModel 的子类）
|
|||
|
|
使其能直接用于 create_agent
|
|||
|
|
"""
|
|||
|
|
# 配置参数（通过 __init__ 初始化）
|
|||
|
|
config: Any = None
|
|||
|
|
model: Any = None
|
|||
|
|
|
|||
|
|
def __init__(self, config):
|
|||
|
|
super().__init__() # 必须调用父类构造函数
|
|||
|
|
self.config = config
|
|||
|
|
self.model = None
|
|||
|
|
self._validate_config()
|
|||
|
|
logger.info(f"初始化 {self.__class__.__name__}，模型: {config.model_name}")
|
|||
|
|
|
|||
|
|
# ---------------------- 必须实现的核心抽象方法（LangChain 协议） ----------------------
|
|||
|
|
def _generate(
|
|||
|
|
self,
|
|||
|
|
messages: List[BaseMessage],
|
|||
|
|
stop: Optional[List[str]] = None,
|
|||
|
|
run_manager: Optional[CallbackManagerForLLMRun] = None,
|
|||
|
|
**kwargs: Any,
|
|||
|
|
) -> ChatResult:
|
|||
|
|
"""
|
|||
|
|
核心同步生成方法（LangChain 要求必须实现）
|
|||
|
|
messages: 消息列表（如 [HumanMessage(content="你好")]）
|
|||
|
|
返回 ChatResult 类型（LangChain 标准输出）
|
|||
|
|
"""
|
|||
|
|
logger.error(f"{self.__class__.__name__} 未实现 同步 _generate 方法")
|
|||
|
|
|
|||
|
|
async def _agenerate(
|
|||
|
|
self,
|
|||
|
|
messages: List[BaseMessage],
|
|||
|
|
stop: Optional[List[str]] = None,
|
|||
|
|
run_manager: Optional[CallbackManagerForLLMRun] = None,
|
|||
|
|
** kwargs: Any,
|
|||
|
|
) -> ChatResult:
|
|||
|
|
"""异步生成方法（LangChain 异步协议）"""
|
|||
|
|
logger.error(f"{self.__class__.__name__} 未实现 异步 _agenerate 方法")
|
|||
|
|
|
|||
|
|
@property
|
|||
|
|
def _llm_type(self) -> str:
|
|||
|
|
"""返回模型类型标识（如 "openai"、"llama"、"bge"）"""
|
|||
|
|
return self.__class__.__name__
|
|||
|
|
|
|||
|
|
def load_model(self) -> None:
|
|||
|
|
"""加载模型（自定义逻辑）"""
|
|||
|
|
logger.error(f"{self.__class__.__name__} 未实现 load_model 方法")
|
|||
|
|
|
|||
|
|
def close(self) -> None:
|
|||
|
|
"""释放资源（自定义逻辑）"""
|
|||
|
|
if self.model:
|
|||
|
|
logger.info(f"释放 {self.__class__.__name__} 模型资源")
|
|||
|
|
self.model = None
|
|||
|
|
|
|||
|
|
def __enter__(self):
|
|||
|
|
self.load_model()
|
|||
|
|
return self
|
|||
|
|
|
|||
|
|
def __exit__(self, exc_type, exc_val, exc_tb):
|
|||
|
|
self.close()
|