patching LLMs api for structured output with Instructor library

instructor_llm.py

import instructor
from openai import OpenAI
from anthropic import Anthropic
from pydantic import BaseModel

class UserInfo(BaseModel):
    field1: str
    field2: int

# Initialize an AI client with various providers ....

# client = instructor.from_openai(OpenAI())
# model = "gpt-4o-mini"

client = instructor.from_anthropic(Anthropic())
model = "claude-3-5-haiku-20241022"

system_template = "Your task is to extract the information about a user from the prompt text. extract name into field1 and born year into field2"
prompt_template = "The {user} was born in {year} and now is {age} years old"

data  = {"user":"Joe", "year":1995, "age":30}

messages = [
    {"role": "system", "content": system_template},
    {"role": "user", "content": prompt_template.format(**data)}
]

answer = client.chat.completions.create(
    model=model,
    response_model=UserInfo, # obligatory for <create> method
    messages=messages,
    max_tokens = 1024 # obligatory for Anthropic
)

print(answer)
usage = answer._raw_response.usage
print(usage.input_tokens,usage.output_tokens)