Skip to content

Installation

DEQN-JAX requires Python 3.10+.

From source (alpha)

git clone https://github.com/deqn-jax/deqn-jax.git
cd deqn-jax
uv sync

For editable installs (hacking on the framework):

uv pip install -e .

CUDA

For GPU acceleration on Linux (x86_64 or aarch64):

uv pip install -U "jax[cuda13]"  # CUDA 13
# or
uv pip install -U "jax[cuda12]"  # CUDA 12

Optional dependencies

Group Purpose Install
logging TensorBoard / Orbax checkpoints uv pip install -e ".[logging]"
wandb Weights & Biases experiment tracking uv pip install -e ".[wandb]"
docs mkdocs-material site (this site) uv pip install -e ".[docs]"
all Everything uv pip install -e ".[all]"

Verify

uv run deqn-jax check
uv run deqn-jax list

check confirms JAX detects your devices; list shows available models.