/pyskip

Preview Release: numpy-like tensor library accelerated by run-length compressible math

Primary LanguageJupyter NotebookMIT LicenseMIT

PySkip

Summary

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.

Run Illustration
Highlighting compressible runs within a minecraft voxel grid

To install as a Python extension

> pip install .

To build and run tests

> python -m venv .venv/
> source .venv/Scripts/activate
> pip install -r requirements.txt
> python tools/build.py

System requirements

  1. Python >=3.6
  2. C++ compiler toolchain
  3. CMake
  4. CUDA toolkit (optional)