Switch from Instructor to OpenAI Structured Outputs

OpenAI recently released structured outputs, which removes some of the complexity of using Instructor or other structured output libraries.

This pattern will transform your existing code to use the new structured outputs API.

Apply with the Grit CLI

grit apply instructor_to_openai

Sample Usage

Previously you would use Instructor to patch create and add a response_format arg.

BEFORE

import instructor
from pydantic import BaseModel
from openai import OpenAI


# Define your desired output structure
class UserInfo(BaseModel):
    name: str
    age: int


client = instructor.from_openai(OpenAI())

# Extract structured data from natural language
user_info = client.chat.completions.create(
    model="gpt-3.5-turbo",
    response_model=UserInfo,
    messages=[{"role": "user", "content": "John Doe is 30 years old."}],
)

print(user_info.name)
#> John Doe
print(user_info.age)
#> 30

Now you just need to use the parse method directly on the OpenAI client.

AFTER

from pydantic import BaseModel
from openai import OpenAI


# Define your desired output structure
class UserInfo(BaseModel):
    name: str
    age: int

client = OpenAI()

# Extract structured data from natural language
user_info = client.beta.chat.completions.parse(
    model="gpt-3.5-turbo",
    response_format=UserInfo,
    messages=[{"role": "user", "content": "John Doe is 30 years old."}],
)

if user_info.choices[0].message.refusal:
    raise Exception(f"GPT refused to comply! {user_info.choices[0].message.refusal}")
user_info = user_info.choices[0].message.parsed

print(user_info.name)
#> John Doe
print(user_info.age)
#> 30