Skip to content

Latest commit

 

History

History
55 lines (40 loc) · 1.5 KB

DETAILED_INSTALL.md

File metadata and controls

55 lines (40 loc) · 1.5 KB
Raw

Detailed installation guide

Conda virtual environment

We recommend using uv for package management. These instructions should allow you to install and run mava.

  1. Install uv
curl -LsSf https://astral.sh/uv/install.sh | sh
  1. Clone mava
git clone https://github.com/instadeepai/Mava.git
cd mava
  1. Create and activate a virtual environment and install requirements
uv venv -p=3.12
source .venv/bin/activate
uv pip install -e .
  1. Install jax on your accelerator. The example below is for an NVIDIA GPU, please the official install guide for other accelerators. Note that the Jax version we use will change over time, please check the requirements.txt for our latest tested Jax verion.
uv pip install "jax[cuda12]==0.4.30"
  1. Run a system!
python mava/systems/ppo/anakin/ff_ippo.py env=rware

Docker

If you are having trouble with dependencies we recommend using our docker image and provided Makefile.

  1. Build the docker image using the make command:

    To build the docker image run

    make build

  2. Run a system:

    make run example=dir/to/system.py

    For example, make run example=mava/systems/ppo/ff_ippo.py.

    Alternatively, run bash inside a docker container with Mava installed by running make bash, and from there systems can be run as follows: python dir/to/system.py.