struct-gpt 2.2.7

structured llm outputs

struct-gpt

Get structured output from LLM's.

from struct_gpt import OpenAiBase
from pydantic import Field


class SentimentSchema(OpenAiBase):
    """
    Determine the sentiment of the given text:

    {content}
    """
    # 👆this becomes the prompt

    sentiment: str = Field(description="Either -1, 0, or 1.")


result = SentimentSchema.from_openai(content="I love pizza!").json()
                                     # 👆this ends up in the prompt
print(result)

outputs:

{
  "sentiment": "1"
}

Features

• Easy creation of custom models using the OpenAI API
• Integration with Pydantic for model validation and serialization
• Flexible configuration with retries and temperature settings

Installation

pip install struct-gpt

Usage

• Template variables in the class' docstring are replaced with the keyword arguments passed to from_openai
• Classes can reference one another
• You can use the OpenAiMixin to add functionality to existing Pydantic classes

from struct_gpt import OpenAiBase, OpenAiMixin
from pydantic import Field, BaseModel
from typing import Mapping


class SentimentSchema(OpenAiBase):
    sentiment: str = Field(description="Either -1, 0, or 1.")


class SentimentAnalysis(BaseModel, OpenAiMixin):
    """
    Determine the sentiment of each word in the following: {text}

    Also determine the overall sentiment of the text and who the narrator is.
    """

    words_to_sentiment: Mapping[str, SentimentSchema]
    overall_sentiment: SentimentSchema
    narrator: str


print(
    SentimentAnalysis.from_openai(
        text="As president, I loved the beautiful scenery, but the long hike was exhausting."
    ).json(indent=2)
)

See Output

{
  "words_to_sentiment": {
    "As": {
      "sentiment": "0"
    },
    "president,": {
      "sentiment": "1"
    },
    "I": {
      "sentiment": "0"
    },
    "loved": {
      "sentiment": "1"
    },
    "the": {
      "sentiment": "0"
    },
    "beautiful": {
      "sentiment": "1"
    },
    "scenery,": {
      "sentiment": "1"
    },
    "but": {
      "sentiment": "-1"
    },
    "long": {
      "sentiment": "-1"
    },
    "hike": {
      "sentiment": "-1"
    },
    "was": {
      "sentiment": "0"
    },
    "exhausting.": {
      "sentiment": "-1"
    }
  },
  "overall_sentiment": {
    "sentiment": "0"
  },
  "narrator": "president"
}

Improving reliability with examples

from_openai can accept a list of examples to guide the model and improve its accuracy.

Each example is a dictionary containing an input and output key.

output should be an instance of the model, a dictionary, or its json string representation,.

from struct_gpt import OpenAiBase
from pydantic import Field


class SentimentSchema(OpenAiBase):
    """
    Determine the sentiment of the given text:

    {content}
    """

    sentiment: str = Field(description="Either -1, 0, or 1.")


examples = [
    {
        "input": "I love the beach!",
        "output": {"sentiment": "1"},
    },
    {
        "input": "don't touch that",
        "output": '{"sentiment": "-1"}',
    },
    {
        "input": "this library is neat!",
        "output": SentimentSchema(sentiment="1"),
    },
]

print(SentimentSchema.from_openai(content="I love pizza!", examples=examples).json())

outputs:

{
  "sentiment": "1"
}