Installation & Setup

Installation & Setup#

Install UV#

To begin, configure your environment with uv, a fast and versatile Python package manager written in Rust. Please follow the installation instructions on the official uv website.

On Linux or macOS, you can install uv with a single command (requires an internet connection):

curl -LsSf https://astral.sh/uv/install.sh | sh

It is highly recommended that configuring high-performance mirrors based on your IP location. For example, for users in China, you cloud using the mirror provided by TUNA.

# Add the following lines into ~/.config/uv/uv.toml
[[index]]
url = "https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple/"
default = true

Create deeph python virtual environment#

Create python 3.13 environments with uv:

mkdir ~/.uvenv
cd ~/.uvenv
uv venv deeph --python=3.13 # Create `deeph` venv in current dir

Then, the uv virtual environment can be activate with command,

source ~/.uvenv/deeph/bin/activate

Conveniently, all files installed into the deeph venv will be located in the ~/.uvenv/deeph directory.

Install DeepH-pack#

Ensure you’ve activated the uv environment as described in the previous section, and that you’re currently in the deeph environment you created.

uv pip install ./deepx-1.0.6+light-py3-none-any.whl[gpu] --extra-index-url https://download.pytorch.org/whl/cpu

This will install all the dependencies, including GPU version of jax, flax, etc. For CPU-only platforms, replace [gpu] with [cpu].

NOTE:

  • ./deepx-1.0.6+light-py3-none-any.whl is the Python wheel file available for download from the official DeepH-pack website.

  • The [gpu] extra dependency tag indicates the GPU-accelerated version of the package, which is strongly recommended for optimal performance. If your system only supports CPU computation, replace [gpu] with [cpu].

  • The --extra-index-url flag is used to specify an additional package index (in this case, PyTorch’s official repository) for resolving certain dependencies. The url ended with "cpu" because pytorch is ony used on the CPU side.

  • Please note that the DeepH-pack referred to in this documentation is distinct from other open-source projects sharing similar names. To avoid confusion, all software packages in this documentation use the namespace deepx (short for deeph-jax). This naming convention clearly separates this new JAX-based public release from earlier experimental versions (such as DeepH and DeepH-E3), which use package names like deeph or deephe3. This distinction will remain consistent in all future releases.

Contents