Research CommonsResearch Commons
gpu-train/Installation

Installation

Install gpu-train from the GitHub Releases wheel, with optional per-provider extras.

gpu-train is not yet on PyPI. Install it from the prebuilt wheel attached to the GitHub Releases of the repository. The core install gives you the Python API and the zero-cost local provider; cloud providers and the dashboard are opt-in extras.

Requirements

  • Python 3.10+
  • ssh and rsync on your PATH (used to drive remote boxes; preinstalled on macOS and most Linux)
  • A credential for at least one cloud provider (or just use local)

Install

Install the latest release wheel directly from GitHub:

pip install "gpu-train[runpod,wandb] @ https://github.com/Research-Commons/gpu-train/releases/download/v0.0.3/gpu_train-0.0.3-py3-none-any.whl"

Or, if you have the repository checked out, install from the vendored wheel:

pip install "gpu-train[server,runpod] @ file://$(pwd)/vendor/wheels/gpu_train-0.0.3-py3-none-any.whl"
Pin it in a project

In a consuming project's requirements.txt, pin the release URL so installs are reproducible:

gpu-train[server,runpod] @ https://github.com/Research-Commons/gpu-train/releases/download/v0.0.3/gpu_train-0.0.3-py3-none-any.whl

Optional extras

The base install includes the core control plane + the local provider. Everything below is opt-in — add an extra only when you need that capability.

# Providers
pip install "gpu-train[runpod]"   # RunPod (GraphQL + SSH)
pip install "gpu-train[vastai]"   # Vast.ai marketplace
pip install "gpu-train[gcp]"      # Google Compute Engine
pip install "gpu-train[colab]"    # Colab tunnel connector
 
# Execution + tracking
pip install "gpu-train[ray]"      # Ray runtime backend
pip install "gpu-train[wandb]"    # Weights & Biases tracking
 
# Dashboard server (bundles the prebuilt UI — no Node.js needed)
pip install "gpu-train[server]"
 
# Everything
pip install "gpu-train[all]"

Combine them freely, e.g. gpu-train[server,runpod,wandb].

Development install

To work on the package itself, clone the repo and install in editable mode with the dev extras:

git clone https://github.com/Research-Commons/gpu-train
cd gpu-train
python -m pip install -e ".[dev,server,runpod]"

CI runs the same gate locally:

ruff check src tests
mypy src
pytest --cov --cov-fail-under=90

The dashboard is a Next.js app under ui/; rebuild and re-bundle the static export with:

python scripts/build_ui.py

Verify

import gpu_train
print(gpu_train.__version__)

Once installed, head to the Quickstart.