langchain调用本地ollama语言模型和嵌入模型

参考：ollama兼容OpenAIEmbeddings的解决思路

解决代码：

访问embedding模型代码

# 测试以下两个引用都可以
from langchain_openai import OpenAIEmbeddings
#from langchain_community.embeddings import OpenAIEmbeddings 
from typing import List, Optionalclass OllamaCompatibleEmbeddings(OpenAIEmbeddings):def _tokenize(self, texts: List[str], chunk_size: int) -> tuple:"""禁用 Tokenization，直接返回原始文本和索引"""indices = list(range(len(texts)))return (range(0, len(texts), chunk_size), texts, indices)def _get_len_safe_embeddings(self, texts: List[str], *, engine: str, chunk_size: Optional[int] = None) -> List[List[float]]:"""直接传递原始文本，跳过 Token 化步骤"""_chunk_size = chunk_size or self.chunk_sizebatched_embeddings: List[List[float]] = []# 直接遍历原始文本分块for i in range(0, len(texts), _chunk_size):chunk = texts[i: i + _chunk_size]# 关键修改：input 直接使用文本列表response = self.client.create(input=chunk,  # 直接使用原始文本列表model=self.model,  # 显式传递模型参数**{k: v for k, v in self._invocation_params.items() if k != "model"})if not isinstance(response, dict):response = response.model_dump()batched_embeddings.extend(r["embedding"] for r in response["data"])# 跳过空文本处理（Ollama 不需要）return batched_embeddingsasync def _aget_len_safe_embeddings(self, texts: List[str], *, engine: str, chunk_size: Optional[int] = None) -> List[List[float]]:"""异步版本处理逻辑"""_chunk_size = chunk_size or self.chunk_sizebatched_embeddings: List[List[float]] = []for i in range(0, len(texts), _chunk_size):chunk = texts[i: i + _chunk_size]response = await self.async_client.create(input=chunk,model=self.model,**{k: v for k, v in self._invocation_params.items() if k != "model"})if not isinstance(response, dict):response = response.model_dump()batched_embeddings.extend(r["embedding"] for r in response["data"])  # 注意: 实际应为 "embedding"return batched_embeddingsembeddings = OllamaCompatibleEmbeddings(model="bge-m3:latest",openai_api_base='http://localhost:11434/v1',#说明一下，openai会自己拼接成http://127.0.0.1:11434/v1/embeddingsapi_key="ollama",# 任意字符串即可chunk_size=512)
output = embeddings.embed_query("你好")   
print(output)

访问LLM模型代码

from langchain_openai import ChatOpenAI
from langchain.schema import HumanMessage# 配置ChatOpenAI类指向本地Ollama API
llm = ChatOpenAI(model_name="deepseek-r1:1.5b",  # 指定Ollama部署的模型名称openai_api_base="http://localhost:11434/v1",  # Ollama API地址api_key="ollama",  # 任意字符串，仅用于兼容格式temperature=0.7,  # 控制输出随机性，0为确定性，1为最大随机性max_tokens=2000,  # 最大生成长度
)# 定义提问函数
def ask_llm(question: str) -> str:"""向本地DeepSeek-R1模型提问并获取回答"""# 创建HumanMessage对象messages = [HumanMessage(content=question)]# 调用模型生成回答response = llm.invoke(messages)# 提取回答内容return response.content# 测试示例
if __name__ == "__main__":question = "请简要介绍量子计算的基本原理"answer = ask_llm(question)print(f"问题: {question}")print("\n回答:")print(answer)