alpha-evolve

Automate population-based evolution of ML models within a fixed compute budget.

150|19|Updated Jun 15, 2026

Search Tags:#parallel processing #autoresearch #ML evolution #population-based search #model evolution

Ready to run

npx skills add https://github.com/gaasher/Agent-Loop-Skills --skill alpha-evolve

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: alpha-evolve
Source: https://github.com/gaasher/Agent-Loop-Skills/tree/main/loops/alpha-evolve
Command: npx skills add https://github.com/gaasher/Agent-Loop-Skills --skill alpha-evolve

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) and assets (resource) components.

What problem does it solve?

This Skill automates the evolution of ML models/programs through a population-based search, allowing for parallel exploration and the preservation of diversity.

Core Features & Use Cases

Population-Based Evolution: Uses a generational evolution approach with parallel proposers and a cascade-evaluated training run.
Diversity-Preserving Search: Ensures diverse high performers co-exist instead of one local optimum winning.
Use Case: Ideal for parallel, diversity-preserving search over a model/program where many variants explore at once, and the archive keeps the illuminated frontier.

Quick Start

Run the alpha-evolve skill with the following command: alpha-evolve.run --total_budget <total_budget> --concurrency <concurrency>.

Frequently Asked Questions about alpha-evolve

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I automate ML model evolution using population-based search?▼

Population-based search automates ML model evolution by using generational evolution with parallel proposers and cascade-evaluated training to explore and preserve diverse high-performing variants.

What is the best way to run parallel evolution without converging on a single local optimum?▼

Diversity-preserving search ensures diverse high performers co-exist instead of one local optimum winning, maintaining an illuminated frontier in the archive during parallel exploration.

Do I need Python 3.9 to run population-based model evolution?▼

Yes, population-based model evolution requires Python 3.9 or higher to execute the bounded-parallelism re-creation of AlphaEvolve/OpenEvolve for ML autoresearch.

How do I start a parallel ML autoresearch task with a fixed compute budget?▼

Start parallel ML autoresearch by running the command with specified total budget and concurrency flags, allowing the process to operate within fixed compute limits or until interrupted.

What are the limitations of bounded-parallelism search for ML programs?▼

Bounded-parallelism search operates within a fixed compute budget or until interrupted, limiting infinite exploration while tailoring the AlphaEvolve/OpenEvolve recreation for ML autoresearch.