alphaevolve

Replicating alphaevolve from AlphaEvolve: A coding agent for scientific and algorithmic discovery.

WIP.

local dev

Get a token from huggingface hub before you run, especially if you plan to use gated models. Then put the token in a .env file, following the format in .env.example.

uv sync
source .venv/bin/activate

Run main.py

# Run with default settings
uv run main.py

# Run with custom settings
uv run main.py --model-id "google/gemma-2b-it" --population-size 10 --num-generations 50 --num-parent-context 3

Command-line arguments

• --model-id: HuggingFace model ID to use (default: "google/gemma-2b-it")
• --population-size: Number of candidate programs in the population (default: 5)
• --num-generations: Number of generations to run the evolutionary search (default: 50)
• --num-parent-context: Number of best programs to include in LLM context for generation (default: 2)