使用GPT3.5模型构建油管频道问答机器人

在 chatgpt api（也就是 GPT-3.5-Turbo）模型出来后，因钱少活好深受大家喜爱，所以 LangChain 也加入了专属的链和模型，我们来跟着这个例子看下如何使用他。

  1. import os
  3. from langchain.document_loaders import YoutubeLoader
  4. from langchain.embeddings.openai import OpenAIEmbeddings
  5. from langchain.vectorstores import Chroma
  6. from langchain.text_splitter import RecursiveCharacterTextSplitter
  7. from langchain.chains import ChatVectorDBChain, ConversationalRetrievalChain
  9. from langchain.chat_models import ChatOpenAI
  10. from langchain.prompts.chat import (
  11.   ChatPromptTemplate,
  12.   SystemMessagePromptTemplate,
  13.   HumanMessagePromptTemplate
  14. )
  16. # 加载 youtube 频道
  17. loader = YoutubeLoader.from_youtube_url('https://www.youtube.com/watch?v=Dj60HHy-Kqk')
  18. # 将数据转成 document
  19. documents = loader.load()
  21. # 初始化文本分割器
  22. text_splitter = RecursiveCharacterTextSplitter(
  23.   chunk_size=1000,
  24.   chunk_overlap=20
  25. )
  27. # 分割 youtube documents
  28. documents = text_splitter.split_documents(documents)
  30. # 初始化 openai embeddings
  31. embeddings = OpenAIEmbeddings()
  33. # 将数据存入向量存储
  34. vector_store = Chroma.from_documents(documents, embeddings)
  35. # 通过向量存储初始化检索器
  36. retriever = vector_store.as_retriever()
  38. system_template = """
  39. Use the following context to answer the user's question.
  40. If you don't know the answer, say you don't, don't try to make it up. And answer in Chinese.
  41. -----------
  42. {question}
  43. -----------
  44. {chat_history}
  45. """
  47. # 构建初始 messages 列表，这里可以理解为是 openai 传入的 messages 参数
  48. messages = [
  49.   SystemMessagePromptTemplate.from_template(system_template),
  50.   HumanMessagePromptTemplate.from_template('{question}')
  51. ]
  53. # 初始化 prompt 对象
  54. prompt = ChatPromptTemplate.from_messages(messages)
  57. # 初始化问答链
  58. qa = ConversationalRetrievalChain.from_llm(ChatOpenAI(temperature=0.1,max_tokens=2048),retriever,condense_question_prompt=prompt)
  61. chat_history = []
  62. while True:
  63.   question = input('问题：')
  64.   # 开始发送问题 chat_history 为必须参数,用于存储对话历史
  65.   result = qa({'question': question, 'chat_history': chat_history})
  66.   chat_history.append((question, result['answer']))
  67.   print(result['answer'])

我们可以看到他能很准确的围绕这个油管视频进行问答

image-20230406211923672

使用流式回答也很方便

  1. from langchain.callbacks.base import CallbackManager
  2. from langchain.callbacks.streaming_stdout import StreamingStdOutCallbackHandler
  4. chat = ChatOpenAI(streaming=True, callback_manager=CallbackManager([StreamingStdOutCallbackHandler()]), verbose=True, temperature=0)
  5. resp = chat(chat_prompt_with_values.to_messages())