使用LangChain+OpenAI生成文本时LLMChain验证错误求助

问题背景 #

使用Python 3.10，依赖openai==1.3.5、langchain==0.0.345，尝试用LangChain结合OpenAI生成文本时，运行代码出现ValidationError。

原代码 #

#openai==1.3.5
#langchain==0.0.345

from transformers import pipeline
from langchain import PromptTemplate, LLMChain, OpenAI

import os
os.environ['OPENAI_API_KEY'] = '*****'

def generate_story (scenario):
  template = ''' 
          You are a story teller. Please make a story based on the context below:
          CONTEXT: {scenario}
          STORY: 
  '''

  prompt = PromptTemplate (template = template, input_variables = ['scenario'])
  story_llm = LLMChain(LLM = OpenAI(
       model_name = 'gpt-3.5-turbo',
       temperature = 1),
        prompt = prompt
  )
  story = story_llm.predict(scenario = scenario)
  print(story)
  return story


scenario = 'A person is skiing down a snowy slope.'
story = generate_story (scenario)
print(story)

错误信息 #

/usr/local/lib/python3.10/dist-packages/pydantic/main.cpython-310-x86_64-linux-gnu.so in pydantic.main.BaseModel.__init__()

ValidationError: 2 validation errors for LLMChain
llm
  field required (type=value_error.missing)
LLM
  extra fields not permitted (type=value_error.extra)

问题原因 #

LangChain的LLMChain类初始化参数要求传入小写的llm，你写成了大写的LLM，导致Pydantic验证时：

• 找不到必填的llm字段，抛出"field required"错误
• 大写的LLM属于未定义的额外字段，抛出"extra fields not permitted"错误

修正后的代码 #

修改LLMChain初始化时的参数名，将LLM改为llm：

#openai==1.3.5
#langchain==0.0.345

from transformers import pipeline
from langchain import PromptTemplate, LLMChain, OpenAI

import os
os.environ['OPENAI_API_KEY'] = '*****'

def generate_story (scenario):
  template = ''' 
          You are a story teller. Please make a story based on the context below:
          CONTEXT: {scenario}
          STORY: 
  '''

  prompt = PromptTemplate (template = template, input_variables = ['scenario'])
  # 这里将LLM改为llm
  story_llm = LLMChain(llm = OpenAI(
       model_name = 'gpt-3.5-turbo',
       temperature = 1),
        prompt = prompt
  )
  story = story_llm.predict(scenario = scenario)
  print(story)
  return story


scenario = 'A person is skiing down a snowy slope.'
story = generate_story (scenario)
print(story)

额外说明 #

gpt-3.5-turbo属于Chat模型，在LangChain中更适合用ChatOpenAI类而非OpenAI类，当前版本OpenAI类虽能兼容，但如果后续出现问题，可以替换为：

from langchain.chat_models import ChatOpenAI

# 替换OpenAI为ChatOpenAI
story_llm = LLMChain(llm = ChatOpenAI(
     model_name = 'gpt-3.5-turbo',
     temperature = 1),
      prompt = prompt
)

内容的提问来源于stack exchange，提问作者zesla