PACEvolve: Enabling Long-Horizon Progress-Aware Consistent Evolution

Table of Contents

1. About the Project
2. Prerequisites
3. Installation & Usage
4. Support & Contribution
5. License

---

About the Project

This repo contains implementation for the PACEvolve paper.

---

Prerequisites

Before installing PACEvolve you need:

• Python 3.9 or later
• We support OpenAI / Anthropic / Google Gemini APIs. You should have at least one API key available, more installation instructions can be found below.
• Git

The project relies on specific directory structures to locate tasks and configurations. Ensure your project root contains:

• workflows/ (where this script resides)
• tasks/ (containing task definitions and their respective environment installation requirements)

---

Installation & Usage

1. Installation

Clone the repository and install the required dependencies.

git clone [https://github.com/MinghaoYan/PACEvolve.git](https://github.com/MinghaoYan/PACEvolve.git)
cd auto-evo
pip install -r requirements.txt

---

2. Setup API Keys & Dependencies

You must configure the API keys and install the necessary packages for the models you intend to use. We currently support Google Gemini, OpenAI, and Anthropic.

1. Set your API Keys Export the environment variables for the providers you plan to use:

# For Google Gemini
export GOOGLE_API_KEY="your_gemini_key"

# For OpenAI (GPT-4)
export OPENAI_API_KEY="your_openai_key"

# For Anthropic (Claude)
export ANTHROPIC_API_KEY="your_anthropic_key"

# Install all supported clients
pip install google-generativeai openai anthropic

Each task contains a config.yaml file in the config subdirectory, you can change the backbone llm by changing the llm section in the config file.

3. Running Your First Experiment

To run the evolutionary process, execute the script with a specific task_id. This assumes you have a task configuration file located at ../tasks/<task_id>/config/.

python run_experiment.py --task_id "my_task"

---

Support & Contribution

Documentation

Transcript Logs: Detailed logs of the LLM's thought process and code generation are saved to the transcripts defined in your YAML config. Controller Logs: Technical execution logs are saved to controller_verbose_*.log.

If you need help, please open an issue in the repository!

---

License

Open-source project

You are free to copy, modify, and distribute PACEvolve with attribution under the terms of the Apache 2.0 license. See the LICENSE file for details.

---

This is not an officially supported Google product. This project is not eligible for the Google Open Source Software Vulnerability Rewards Program.

This project is intended for demonstration purposes only. It is not intended for use in a production environment.