📦Installation

System Dependencies

# uv - fast Python package manager (handles Python 3.12+ automatically)
curl -LsSf https://astral.sh/uv/install.sh | sh

# swig - required for Box2D environments (LunarLander, BipedalWalker)
brew install swig        # macOS
# apt-get install -y swig  # Linux/WSL

Windows: Use WSL2 and follow Linux instructions.

Install SLM Lab

# Clone
git clone https://github.com/kengz/SLM-Lab.git
cd SLM-Lab

# Install
uv sync
uv tool install --editable .

# Verify
slm-lab --help                                 # list all commands
slm-lab run --help                             # options for run command

# Troubleshoot: if slm-lab not found, use uv run
uv run slm-lab --help

MuJoCo Playground

To use MuJoCo Playground environments (JAX/MJWarp GPU-accelerated simulation with 54 environments):

uv sync --group playground

This adds JAX, MuJoCo Playground, and MJWarp dependencies. Requires a CUDA GPU for GPU-accelerated simulation.

Minimal Install

For remote training in the cloud with dstack, a minimal installation without ML dependencies is available:

git clone https://github.com/kengz/SLM-Lab.git
cd SLM-Lab
uv sync --no-default-groups
uv tool install dstack

Docker

For containerized runs without local setup, use the prebuilt image:

docker pull ghcr.io/kengz/slm-lab:latest
docker run -it ghcr.io/kengz/slm-lab:latest uv run slm-lab --help
docker run -v $(pwd)/data:/app/data ghcr.io/kengz/slm-lab:latest \
    uv run slm-lab run slm_lab/spec/benchmark/ppo/ppo_cartpole.json ppo_cartpole train