Installation

Laptop, no GPU #

Enough to serve, record, report, and evolve a harness against a hosted model.

For a released version:

pip install reef-infra        # the import package is `reef`

Or, to work on Reef itself:

git clone https://github.com/Human-Agent-Society/reef.git && cd reef
pip install -e .
git lfs install

git lfs is required: Reef keeps release history in Git, with weight files under LFS. The checkout includes the core recipe example, the paper-backed methods under recipes/, and the harness evolution demo under tutorials/harness_evolve/.

Client only #

A client that only talks to a Reef deployment does not need the repository or the full install. Install the stdlib-only wire client instead:

pip install reef-client

GPU image, for weight training #

Weight training runs Ray, a Slime driver, and Reef together, against CUDA-specific builds of torch, SGLang, Megatron-Core, and FlashAttention. The supported way to get them is the image:

docker build -f docker/Dockerfile.reef -t reef .

docker/README.md covers the GPU prerequisites and how to start the container; Evolve your model picks up from there.

Inside the container, to install from source instead of using what the image already carries:

git submodule update --init --recursive
pip install -e ".[slime]"
pip install --no-deps --group runtime

Keep --no-deps: the CUDA-specific packages are already in the image, and resolving them again replaces working builds. The slime extra installs only the Python-side dependencies; the runtime group pins the training runtime itself.