LangChain

langchain.py

llm = OpenAI(temperature =0)
class Action(BaseModel):     
   action: str = Field(description="action to take")     
   action_input: str = Field(description="input to the action")   
parser = PydanticOutputParser(pydantic_object=Action)

# the parser provides natural language formatting instruction for use in LLM. 
# The process is centered around prompts.

prompt = PromptTemplate(template="Answer the user query.      
     \n{format_instructions}\n{query}\n", 
     input_variables=["query"],
     partial_variables={"format_instructions": parser.get_format_instructions()}
)
prompt_value = prompt.format_prompt(query="who is leo di caprios gf?")
model_response = model(prompt_value.to_string())

# the parsing aspect and the querying aspect is detached in LangChain
parsed_model = parser.parse(model_response)

# if the parsing process results in an exception, users could use the retry parser that internally appends the parsing error and prompts the LM to retry it again
retry_parser = RetryWithErrorOutputParser.from_llm(parser=parser, llm=llm)
retry_parser.parse_with_prompt(model_response, prompt_value)