llm-functions 0.1.0

Build semantic functions.

LLM Functions

Define custom LLM functions with minimal code.

How to Define a LLM Function

Here is an example of a directory with files that defines a simple sentiment classification function.

└─ sentiment
   ├── args.json
   └── template.txt

The content of template.txt:

Aalyze and determine whether the sentiment of the following sentence is positive, negative, or neutral.
Sentence:
{sentence}

The content of args.json:

{
    "temperature": 0.0,
    "model": "gpt-3.5-turbo",
    "function_name": "sentiment_classifier",
    "description": "Classify the sentiment of a sentence",
    "properties": {
        "rationale": {
            "type": "string",
            "description": "Your rationale when classifying sentiment of the given sentence."
        },
        "sentiment": {
            "type": "string",
            "enum": ["positive", "negative", "neutral"],
            "description": "The sentiment of the given sentence."
        }
    }
}

Load the LLM function:

from llm_functions import LLMFunction

sentiment = LLMFunction.from_dir("sentiment/")
pred = sentiment(sentence="I am super happy!")

# Output:
# {
#   "rationale": "The sentiment in the sentence is positive",
#   "sentiment": "positive"
# }