langchain 初窥门径

Embedding

扼要言之，embedding 的作用即为使高维疏松的语言转变为低维稠密的向量。原理是矩阵的乘法。

Langchain的小demo

运行环境为 Python 3.11.7 本地conda虚拟环境名字langchain

构建了一个简单的对话程

import os
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.output_parsers import StrOutputParser
from langchain_openai import ChatOpenAI
api_key = "sk-X6k4gks4hxg7IkmGFLYvEdlB1LNx6IIfISlb9KbcpvEKLkiB"
base_url ="https://api.chatanywhere.com.cn"
prompt = ChatPromptTemplate.from_messages([
    ("system", "使用中文回答"),
    ("user", "{input}")
])
llm = ChatOpenAI(api_key=api_key,base_url=base_url)
output_parser = StrOutputParser()
chain = prompt | llm | output_parser
print(chain.invoke({"input": "你好"}))

基于 web 文档内容实施问答的程序。

这里有一个小问题base_url ="https://api.chatanywhere.com.cn/v1/"这段记得加/v1/

from langchain_core.prompts import ChatPromptTemplate
from langchain_openai import ChatOpenAI,OpenAIEmbeddings
from langchain_community.document_loaders import web_base
from langchain_community.vectorstores import faiss
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain.chains.combine_documents import create_stuff_documents_chain
from langchain_core.documents import Document
from langchain.chains import retrieval
api_key = "sk-X6k4gks4hxg7IkmGFLYvEdlB1LNx6IIfISlb9KbcpvEKLkiB"
base_url ="https://api.chatanywhere.com.cn/v1/"
# 创建embeddings接口
embeddings = OpenAIEmbeddings(base_url=base_url,api_key=api_key)
# 导入文档
loader = web_base.WebBaseLoader("https://docs.smith.langchain.com/user_guide")
docs = loader.load()
# 构建向量数据库
text_splitter = RecursiveCharacterTextSplitter()
documents = text_splitter.split_documents(docs)
vector = faiss.FAISS.from_documents(documents, embeddings)
# 定义模型
llm = ChatOpenAI(api_key=api_key,base_url=base_url)
# 模板
prompt = ChatPromptTemplate.from_template("""Answer the following question based only on the provided context:
<context>
{context}
</context>
Question: {input}""")
# 构建链
document_chain = create_stuff_documents_chain(llm, prompt)
retriever = vector.as_retriever()
retrieval_chain = retrieval.create_retrieval_chain(retriever, document_chain)
# 运行
response = retrieval_chain.invoke({"input": "how can langsmith help with testing?"})
print(response["answer"])

添加了上下文功能

from langchain_core.prompts import ChatPromptTemplate
from langchain_openai import ChatOpenAI, OpenAIEmbeddings
from langchain_community.document_loaders import web_base
from langchain_community.vectorstores import faiss
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain.chains.combine_documents import create_stuff_documents_chain
from langchain.chains import retrieval
from langchain.chains import history_aware_retriever
from langchain_core.prompts import MessagesPlaceholder
from langchain_core.messages import HumanMessage, AIMessage
api_key = "sk-X6k4gks4hxg7IkmGFLYvEdlB1LNx6IIfISlb9KbcpvEKLkiB"
base_url = "https://api.chatanywhere.com.cn/v1/"
# 创建embeddings接口
embeddings = OpenAIEmbeddings(base_url=base_url, api_key=api_key)
# 导入文档
loader = web_base.WebBaseLoader("https://docs.smith.langchain.com/user_guide")
docs = loader.load()
# 构建向量数据库
text_splitter = RecursiveCharacterTextSplitter()
documents = text_splitter.split_documents(docs)
vector = faiss.FAISS.from_documents(documents, embeddings)
# 定义模型
llm = ChatOpenAI(api_key=api_key, base_url=base_url)
# 模板
prompt = ChatPromptTemplate.from_messages([
    ("system", "Answer the user's questions based on the below context:\n\n{context}"),
    MessagesPlaceholder(variable_name="chat_history"),
    ("user", "{input}"),
])
# 构建链
retriever = vector.as_retriever()
retriever_chain = history_aware_retriever.create_history_aware_retriever(llm, retriever, prompt)
document_chain = create_stuff_documents_chain(llm, prompt)
retrieval_chain =retrieval.create_retrieval_chain(retriever_chain, document_chain)
# 运行
chat_history = [HumanMessage(content="LangSmith能帮助测试我的LLM应用程序吗?使用中文回答"), AIMessage(content="好的!")]
response =retrieval_chain.invoke({
    "context":retriever,
    "chat_history": chat_history,
    "input": "告诉我怎么做"
})
print(response["answer"])

CNN

官方文档

https://python.langchain.com/docs/get_started/quickstart/#agent

【卷积神经网络】8分钟搞懂CNN，动画讲解喜闻乐见_哔哩哔哩_bilibili

• 卷积层 概括特征
• 池化层 进一步扩大特征
• 全连接层 对前面的特征进行汇总，进行最终的判断