A PyTorch-like tensor library which maintains end-to-end run-length compression to reduce operator costs and memory pressure. With highly compressible tensors like Minecraft voxel grids we achieve near-GPU like performance for convolutions.
Highlighting compressible runs within a minecraft voxel grid
> pip install .
> python -m venv .venv/
> source .venv/Scripts/activate
> pip install -r requirements.txt
> python tools/build.py
- Python >=3.6
- C++ compiler toolchain
- CMake
- CUDA toolkit (optional)