Cupybara

Installation

• Clone and cd into the repo
• chmod +x ./build.sh
• ./build.sh

Possible Errors:

• Make sure to this script is run in an environment managed by pip

Usage:

For usages look in the testing folder for how to interface with the cupybara library

Performance

Performance Measured On 1000x1000 Matrices Hardware: Ryzen 5 3600 & NVIDIA RTX 3070

Cupybara Python Front End: 7.742296567166108 GFLOPS

Pytorch Python Front End: 0.5086737813983089 GFLOPS

Cupybara CUDA Back End: 421354 GFLOPS

Performance Measured On 200x1 Vectors Cupybara Python Front End AVX: 2.755241411022795e-06 GFLOPS

Pytorch Python Front End: 1.6040844970771753e-06 GFLOPS

Uninstall

• Run pip uninstall cupybara
• Run sudo apt remove cupybara