vm-lcsampler: LLM Sampler with LangChain chat models.

Overview

This repository provides tools for sampling text and structured data using Large Language Models (LLMs) by prompt engineering on LangChain.

Install

This repository is managed using poetry. You can install with the following command.

pip install -e .

Examples

In the following examples, OpenAI API key must be set in .env file according to ./.env.org.

Raw text sampling

The full code is in examples/02_chatmodel_structure_sampler.py.

from dotenv import dotenv_values
from langchain_openai import ChatOpenAI
from vm_lcsampler.chatmodel_samplers import ChatModelChunkedTextEnumerator

# Set up LLM
llm = ChatOpenAI(
    api_key=str(dotenv_values()["OPENAI_API_KEY"]),
    temperature=0.0,  # default to 0.7
)

# Set up sampler
sampler = ChatModelChunkedTextEnumerator(llm=llm)
enumerator = sampler.enumerate(
    category_name="cat breeds",
    category_description=None,
    chunk_size=5,
    num_chunk=4,
    few_shot_chunked_samples=[
        [
            "Persian",
            "Siamese",
            "Maine Coon",
            "Bengal",
            "Sphynx",
        ],
    ],
)

# Perform sampling
for i, text in enumerator:
    print(text)

Structured object sampling

See examples/02_chatmodel_structure_sampler.py