基于Langchain使用开源大模型结合本地知识库实现问答

准备工作

1. 首先在HuggingFace中下载ChatGLM-6B权重，并在GitHub中下载ChatGLM-6B源码。
2. 环境安装，控制台中进入ChatGLM源码并输入：

   pip install -r requirements.txt

3. 修改源码中api.py文件中的模型路径。
   

4. 然后在ChatGLM-6B工程中执行：python api.py 开启ChatGLM api。
   

加载模型

from langchain.llms import ChatGLM

model_path = 'E:\hf_model\chatglm-6b'
api = 'http://127.0.0.1:8000'

llm = ChatGLM(endpoint_url=api,
             max_token=80000,
             top_p=0.7)

读取知识库

from langchain.document_loaders import DirectoryLoader

loader = DirectoryLoader('knowledge_base')
documents = loader.load()

文档切分

from langchain.text_splitter import CharacterTextSplitter

text_splitter = CharacterTextSplitter(chunk_size=256,   # 每个切片的token大小
                                      chunk_overlap=0)  # 相邻切片的重复token大小

split_documents = text_splitter.split_documents(documents)

Created a chunk of size 279, which is longer than the specified 256
Created a chunk of size 426, which is longer than the specified 256
Created a chunk of size 302, which is longer than the specified 256
Created a chunk of size 282, which is longer than the specified 256
Created a chunk of size 273, which is longer than the specified 256
Created a chunk of size 289, which is longer than the specified 256

文本向量化

from langchain.embeddings.huggingface import HuggingFaceEmbeddings

# 加载Embedding模型
encode_kwargs = {'normalize_embeddings': False}
model_kwargs = {'device': 'cuda:0'}
embeddings = HuggingFaceEmbeddings(model_name="shibing624/text2vec-base-chinese",
                     cache_folder="E:\hf_model",
                     model_kwargs=model_kwargs,
                     encode_kwargs=encode_kwargs)

from langchain.vectorstores import Chroma
import os

db = Chroma.from_documents(split_documents, embedding=embeddings, persist_directory='VectorStore')

本地知识库问答

from langchain.chains import RetrievalQA

retriever = db.as_retriever()

qa = RetrievalQA.from_chain_type(llm=llm,
                                chain_type='stuff',
                                retriever=retriever)

response = qa.run('小王子为什么离开玫瑰？')

print(response)

小王子离开了他的玫瑰 because he realized that she was more important to him than any other person or thing in the world.