FunMinks: Genetic Algorithm Utilities for LLMs

FunMinks is a Python package for symbolic regression and genetic algorithm experimentation using Large Language Models (LLMs). It provides tools to define, mutate, and evolve code-based individuals ("minks") and manage populations, leveraging LLMs for code generation and mutation.

What is FunMinks?

FunMinks enables evolutionary computation on code snippets, using LLMs to mutate and optimize Python functions to fit target data. It is designed for symbolic regression, automated code improvement, and research in genetic programming with LLMs.

• Mink: An individual represented by a genome (Python code), evaluated for fitness against datapoints.
• Population: A collection of minks, managed for selection, mutation, and evolution.
• LLM Endpoint: Utilities to load and interact with local LLM models for code generation.

Features

• Symbolic regression using genetic algorithms and LLMs
• Population management and selection
• Code mutation via LLM prompts
• Fitness evaluation against custom datapoints
• Extensible for custom genetic operations

Installation

It is recommended to use a virtual environment for isolation:

# Create and activate a virtual environment
python -m venv .venv
.venv\Scripts\Activate.ps1

# Install FunMinks and dependencies
pip install -e .

Or install from source:

pip install git+https://github.com/UGarCil/GeneticAlgoLLM.git

Requirements

• Python >= 3.10
• transformers
• pandas
• huggingface_hub
• torch (optional, see below)

If you need GPU support, install the appropriate version of torch:

pip install torch --extra-index-url https://download.pytorch.org/whl/cu118

Usage

1. Prepare Data and Templates

• Place your datapoints in templates/datapoints.csv (two columns: x, y)
• Use or modify templates/prompt_template.txt for LLM mutation prompts
• Use templates/seed_code_template.txt as the initial genome

2. Load LLM Endpoint

from FunMinks.llm_endpoint import load_llm
llm_params = load_llm('path/to/your/model')

3. Create a Population

from FunMinks.population import Population
from FunMinks.mink_llm_based import Mink

# Load templates and datapoints
with open('templates/prompt_template.txt') as f:
    prompt = f.read()
with open('templates/seed_code_template.txt') as f:
    seed_code = f.read()
with open('templates/datapoints.csv') as f:
    datapoints = f.read()

# Initialize population
pop = Population(
    pop_size=20,
    seed_code=seed_code,
    datapoints=datapoints,
    prompt=prompt,
    llm_params=llm_params
)

4. Run Evolution

for generation in range(10):
    pop.run_episode()
    print(f"Generation {generation}: Avg fitness = {pop.avg_fitness}")

Project Structure

FunMinks/
    mink_llm_based.py   # Mink class (individual)
    population.py       # Population management
    llm_endpoint.py     # LLM loading utilities
icons/
    main_logo.png       # Project logo
templates/
    datapoints.csv      # Example data
    prompt_template.txt # LLM prompt template
    seed_code_template.txt # Initial code genome

License

This project is licensed under the MIT License. See LICENSE for details.

Author

Uriel Garcilazo Cruz

Links

• Homepage
• Repository
• Bug Tracker