Skip to main content

Installation

This page covers environment setup, dependency installation, and common installation/runtime issues.

Requirements​

  • Linux with NVIDIA GPU(s) (recommended for training/evolution workloads)
  • CUDA-compatible PyTorch environment
  • Python 3.12 (project default)

Tip
If you only want to get started quickly, complete this page and then jump to quick-start.

Install AlphaApollo​

1) Clone repository​

git clone https://github.com/tmlr-group/AlphaApollo.git
cd AlphaApollo

2) Create and activate environment​

conda create -n alphaapollo python==3.12 -y
conda activate alphaapollo

3) Run project installer​

bash installation.sh

This script is expected to install dependencies used by AlphaApollo workflows (including Verl-related runtime requirements).

If no matching wheel exists for your environment, disable flash-attn path and continue with a standard attention backend.