Installation
============

Requirements
------------

- **Python**: Version >= 3.9
- **CUDA**: Version >= 12.2
- **PyTorch**: Version >= 2.6.0
- **vLLM**: Version >= 0.8.5

Setup
------------

You can either build docker image from source or install dependencies inside your python env:

🐳 Option 1: Build docker image from source
::::::::::::::::::::::::::::::::::::::::::::

.. code-block:: bash

    docker build -t cosmos-rl-dev:dev .
    docker run -it --gpus all --shm-size=24G -w /workspace/cosmos-rl cosmos-rl-dev:dev

.. note::
    EFA driver is included in the docker image specifically for aws instances with EFA net interface (Sagemaker AI Pod).

🔨 Option 2: Run in your own environment
:::::::::::::::::::::::::::::::::::::::::

.. code-block:: bash

    apt-get update && apt-get install redis-server
    pip install -r requirements.txt
    pip install -e .