Installation¶
Basic install (CPU)¶
pip install vibespatial
Or with uv:
uv add vibespatial
This gives you a fully functional GeoPandas drop-in. All operations work via CPU fallback.
GPU support¶
GPU acceleration requires an NVIDIA GPU with compatible drivers. Install the extra matching your CUDA version:
pip install vibespatial[cu12] # CUDA 12
pip install vibespatial[cu13] # CUDA 13
This pulls in CuPy, cuda-python, CUDA CCCL, pylibcudf, and nvidia-ml-py.
Note: The CUDA toolkit version installed by the extras must not exceed your NVIDIA driver version. Check with
nvidia-smi— the “CUDA Version” shown is the maximum supported.
Development install¶
git clone https://github.com/jarmak-personal/vibeSpatial.git
cd vibeSpatial
uv sync # CPU only
uv sync --extra cu12 # with CUDA 12 GPU deps
Verify GPU availability¶
import vibespatial
sel = vibespatial.select_runtime()
print(sel)
# RuntimeSelection(requested=auto, selected=gpu, reason=...)
If no GPU is available, selected will be cpu and all operations
will use the CPU fallback path transparently.